AGI Ruin: A List of Lethalities

Preamble:

(If you're already familiar with all basics and don't want any preamble, skip ahead to Section B for technical difficulties of alignment proper.)

I have several times failed to write up a well-organized list of reasons why AGI will kill you.  People come in with different ideas about why AGI would be survivable, and want to hear different obviously key points addressed first.  Some fraction of those people are loudly upset with me if the obviously most important points aren't addressed immediately, and I address different points first instead.

Having failed to solve this problem in any good way, I now give up and solve it poorly with a poorly organized list of individual rants.  I'm not particularly happy with this list; the alternative was publishing nothing, and publishing this seems marginally more dignified.

Three points about the general subject matter of discussion here, numbered so as not to conflict with the list of lethalities:

-3I'm assuming you are already familiar with some basics, and already know what 'orthogonality' and 'instrumental convergence' are and why they're true.  People occasionally claim to me that I need to stop fighting old wars here, because, those people claim to me, those wars have already been won within the important-according-to-them parts of the current audience.  I suppose it's at least true that none of the current major EA funders seem to be visibly in denial about orthogonality or instrumental convergence as such; so, fine.  If you don't know what 'orthogonality' or 'instrumental convergence' are, or don't see for yourself why they're true, you need a different introduction than this one.

-2.  When I say that alignment is lethally difficult, I am not talking about ideal or perfect goals of 'provable' alignment, nor total alignment of superintelligences on exact human values, nor getting AIs to produce satisfactory arguments about moral dilemmas which sorta-reasonable humans disagree about, nor attaining an absolute certainty of an AI not killing everyone.  When I say that alignment is difficult, I mean that in practice, using the techniques we actually have, "please don't disassemble literally everyone with probability roughly 1" is an overly large ask that we are not on course to get.  So far as I'm concerned, if you can get a powerful AGI that carries out some pivotal superhuman engineering task, with a less than fifty percent change of killing more than one billion people, I'll take it.  Even smaller chances of killing even fewer people would be a nice luxury, but if you can get as incredibly far as "less than roughly certain to kill everybody", then you can probably get down to under a 5% chance with only slightly more effort.  Practically all of the difficulty is in getting to "less than certainty of killing literally everyone".  Trolley problems are not an interesting subproblem in all of this; if there are any survivors, you solved alignment.  At this point, I no longer care how it works, I don't care how you got there, I am cause-agnostic about whatever methodology you used, all I am looking at is prospective results, all I want is that we have justifiable cause to believe of a pivotally useful AGI 'this will not kill literally everyone'.  Anybody telling you I'm asking for stricter 'alignment' than this has failed at reading comprehension.  The big ask from AGI alignment, the basic challenge I am saying is too difficult, is to obtain by any strategy whatsoever a significant chance of there being any survivors.

-1.  None of this is about anything being impossible in principle.  The metaphor I usually use is that if a textbook from one hundred years in the future fell into our hands, containing all of the simple ideas that actually work robustly in practice, we could probably build an aligned superintelligence in six months.  For people schooled in machine learning, I use as my metaphor the difference between ReLU activations and sigmoid activations.  Sigmoid activations are complicated and fragile, and do a terrible job of transmitting gradients through many layers; ReLUs are incredibly simple (for the unfamiliar, the activation function is literally max(x, 0)) and work much better.  Most neural networks for the first decades of the field used sigmoids; the idea of ReLUs wasn't discovered, validated, and popularized until decades later.  What's lethal is that we do not have the Textbook From The Future telling us all the simple solutions that actually in real life just work and are robust; we're going to be doing everything with metaphorical sigmoids on the first critical try.  No difficulty discussed here about AGI alignment is claimed by me to be impossible - to merely human science and engineering, let alone in principle - if we had 100 years to solve it using unlimited retries, the way that science usually has an unbounded time budget and unlimited retries.  This list of lethalities is about things we are not on course to solve in practice in time on the first critical try; none of it is meant to make a much stronger claim about things that are impossible in principle.

That said:

Here, from my perspective, are some different true things that could be said, to contradict various false things that various different people seem to believe, about why AGI would be survivable on anything remotely remotely resembling the current pathway, or any other pathway we can easily jump to.

 

Section A:

This is a very lethal problem, it has to be solved one way or another, it has to be solved at a minimum strength and difficulty level instead of various easier modes that some dream about, we do not have any visible option of 'everyone' retreating to only solve safe weak problems instead, and failing on the first really dangerous try is fatal.

 

1.  Alpha Zero blew past all accumulated human knowledge about Go after a day or so of self-play, with no reliance on human playbooks or sample games.  Anyone relying on "well, it'll get up to human capability at Go, but then have a hard time getting past that because it won't be able to learn from humans any more" would have relied on vacuum.  AGI will not be upper-bounded by human ability or human learning speed.  Things much smarter than human would be able to learn from less evidence than humans require to have ideas driven into their brains; there are theoretical upper bounds here, but those upper bounds seem very high. (Eg, each bit of information that couldn't already be fully predicted can eliminate at most half the probability mass of all hypotheses under consideration.)  It is not naturally (by default, barring intervention) the case that everything takes place on a timescale that makes it easy for us to react.

2.  A cognitive system with sufficiently high cognitive powers, given any medium-bandwidth channel of causal influence, will not find it difficult to bootstrap to overpowering capabilities independent of human infrastructure.  The concrete example I usually use here is nanotech, because there's been pretty detailed analysis of what definitely look like physically attainable lower bounds on what should be possible with nanotech, and those lower bounds are sufficient to carry the point.  My lower-bound model of "how a sufficiently powerful intelligence would kill everyone, if it didn't want to not do that" is that it gets access to the Internet, emails some DNA sequences to any of the many many online firms that will take a DNA sequence in the email and ship you back proteins, and bribes/persuades some human who has no idea they're dealing with an AGI to mix proteins in a beaker, which then form a first-stage nanofactory which can build the actual nanomachinery.  (Back when I was first deploying this visualization, the wise-sounding critics said "Ah, but how do you know even a superintelligence could solve the protein folding problem, if it didn't already have planet-sized supercomputers?" but one hears less of this after the advent of AlphaFold 2, for some odd reason.)  The nanomachinery builds diamondoid bacteria, that replicate with solar power and atmospheric CHON, maybe aggregate into some miniature rockets or jets so they can ride the jetstream to spread across the Earth's atmosphere, get into human bloodstreams and hide, strike on a timer.  Losing a conflict with a high-powered cognitive system looks at least as deadly as "everybody on the face of the Earth suddenly falls over dead within the same second".  (I am using awkward constructions like 'high cognitive power' because standard English terms like 'smart' or 'intelligent' appear to me to function largely as status synonyms.  'Superintelligence' sounds to most people like 'something above the top of the status hierarchy that went to double college', and they don't understand why that would be all that dangerous?  Earthlings have no word and indeed no standard native concept that means 'actually useful cognitive power'.  A large amount of failure to panic sufficiently, seems to me to stem from a lack of appreciation for the incredible potential lethality of this thing that Earthlings as a culture have not named.)

3.  We need to get alignment right on the 'first critical try' at operating at a 'dangerous' level of intelligence, where unaligned operation at a dangerous level of intelligence kills everybody on Earth and then we don't get to try again.  This includes, for example: (a) something smart enough to build a nanosystem which has been explicitly authorized to build a nanosystem; or (b) something smart enough to build a nanosystem and also smart enough to gain unauthorized access to the Internet and pay a human to put together the ingredients for a nanosystem; or (c) something smart enough to get unauthorized access to the Internet and build something smarter than itself on the number of machines it can hack; or (d) something smart enough to treat humans as manipulable machinery and which has any authorized or unauthorized two-way causal channel with humans; or (e) something smart enough to improve itself enough to do (b) or (d); etcetera.  We can gather all sorts of information beforehand from less powerful systems that will not kill us if we screw up operating them; but once we are running more powerful systems, we can no longer update on sufficiently catastrophic errors.  This is where practically all of the real lethality comes from, that we have to get things right on the first sufficiently-critical try.  If we had unlimited retries - if every time an AGI destroyed all the galaxies we got to go back in time four years and try again - we would in a hundred years figure out which bright ideas actually worked.  Human beings can figure out pretty difficult things over time, when they get lots of tries; when a failed guess kills literally everyone, that is harder.  That we have to get a bunch of key stuff right on the first try is where most of the lethality really and ultimately comes from; likewise the fact that no authority is here to tell us a list of what exactly is 'key' and will kill us if we get it wrong.  (One remarks that most people are so absolutely and flatly unprepared by their 'scientific' educations to challenge pre-paradigmatic puzzles with no scholarly authoritative supervision, that they do not even realize how much harder that is, or how incredibly lethal it is to demand getting that right on the first critical try.)

4.  We can't just "decide not to build AGI" because GPUs are everywhere, and knowledge of algorithms is constantly being improved and published; 2 years after the leading actor has the capability to destroy the world, 5 other actors will have the capability to destroy the world.  The given lethal challenge is to solve within a time limit, driven by the dynamic in which, over time, increasingly weak actors with a smaller and smaller fraction of total computing power, become able to build AGI and destroy the world.  Powerful actors all refraining in unison from doing the suicidal thing just delays this time limit - it does not lift it, unless computer hardware and computer software progress are both brought to complete severe halts across the whole Earth.  The current state of this cooperation to have every big actor refrain from doing the stupid thing, is that at present some large actors with a lot of researchers and computing power are led by people who vocally disdain all talk of AGI safety (eg Facebook AI Research).  Note that needing to solve AGI alignment only within a time limit, but with unlimited safe retries for rapid experimentation on the full-powered system; or only on the first critical try, but with an unlimited time bound; would both be terrifically humanity-threatening challenges by historical standards individually.

5.  We can't just build a very weak system, which is less dangerous because it is so weak, and declare victory; because later there will be more actors that have the capability to build a stronger system and one of them will do so.  I've also in the past called this the 'safe-but-useless' tradeoff, or 'safe-vs-useful'.  People keep on going "why don't we only use AIs to do X, that seems safe" and the answer is almost always either "doing X in fact takes very powerful cognition that is not passively safe" or, even more commonly, "because restricting yourself to doing X will not prevent Facebook AI Research from destroying the world six months later".  If all you need is an object that doesn't do dangerous things, you could try a sponge; a sponge is very passively safe.  Building a sponge, however, does not prevent Facebook AI Research from destroying the world six months later when they catch up to the leading actor.

6.  We need to align the performance of some large task, a 'pivotal act' that prevents other people from building an unaligned AGI that destroys the world.  While the number of actors with AGI is few or one, they must execute some "pivotal act", strong enough to flip the gameboard, using an AGI powerful enough to do that.  It's not enough to be able to align a weak system - we need to align a system that can do some single very large thing.  The example I usually give is "burn all GPUs".  This is not what I think you'd actually want to do with a powerful AGI - the nanomachines would need to operate in an incredibly complicated open environment to hunt down all the GPUs, and that would be needlessly difficult to align.  However, all known pivotal acts are currently outside the Overton Window, and I expect them to stay there.  So I picked an example where if anybody says "how dare you propose burning all GPUs?" I can say "Oh, well, I don't actually advocate doing that; it's just a mild overestimate for the rough power level of what you'd have to do, and the rough level of machine cognition required to do that, in order to prevent somebody else from destroying the world in six months or three years."  (If it wasn't a mild overestimate, then 'burn all GPUs' would actually be the minimal pivotal task and hence correct answer, and I wouldn't be able to give that denial.)  Many clever-sounding proposals for alignment fall apart as soon as you ask "How could you use this to align a system that you could use to shut down all the GPUs in the world?" because it's then clear that the system can't do something that powerful, or, if it can do that, the system wouldn't be easy to align.  A GPU-burner is also a system powerful enough to, and purportedly authorized to, build nanotechnology, so it requires operating in a dangerous domain at a dangerous level of intelligence and capability; and this goes along with any non-fantasy attempt to name a way an AGI could change the world such that a half-dozen other would-be AGI-builders won't destroy the world 6 months later.

7.  The reason why nobody in this community has successfully named a 'pivotal weak act' where you do something weak enough with an AGI to be passively safe, but powerful enough to prevent any other AGI from destroying the world a year later - and yet also we can't just go do that right now and need to wait on AI - is that nothing like that exists.  There's no reason why it should exist.  There is not some elaborate clever reason why it exists but nobody can see it.  It takes a lot of power to do something to the current world that prevents any other AGI from coming into existence; nothing which can do that is passively safe in virtue of its weakness.  If you can't solve the problem right now (which you can't, because you're opposed to other actors who don't want to be solved and those actors are on roughly the same level as you) then you are resorting to some cognitive system that can do things you could not figure out how to do yourself, that you were not close to figuring out because you are not close to being able to, for example, burn all GPUs.  Burning all GPUs would actually stop Facebook AI Research from destroying the world six months later; weaksauce Overton-abiding stuff about 'improving public epistemology by setting GPT-4 loose on Twitter to provide scientifically literate arguments about everything' will be cool but will not actually prevent Facebook AI Research from destroying the world six months later, or some eager open-source collaborative from destroying the world a year later if you manage to stop FAIR specifically.  There are no pivotal weak acts.

8.  The best and easiest-found-by-optimization algorithms for solving problems we want an AI to solve, readily generalize to problems we'd rather the AI not solve; you can't build a system that only has the capability to drive red cars and not blue cars, because all red-car-driving algorithms generalize to the capability to drive blue cars.

9.  The builders of a safe system, by hypothesis on such a thing being possible, would need to operate their system in a regime where it has the capability to kill everybody or make itself even more dangerous, but has been successfully designed to not do that.  Running AGIs doing something pivotal are not passively safe, they're the equivalent of nuclear cores that require actively maintained design properties to not go supercritical and melt down.

 

Section B:

Okay, but as we all know, modern machine learning is like a genie where you just give it a wish, right?  Expressed as some mysterious thing called a 'loss function', but which is basically just equivalent to an English wish phrasing, right?  And then if you pour in enough computing power you get your wish, right?  So why not train a giant stack of transformer layers on a dataset of agents doing nice things and not bad things, throw in the word 'corrigibility' somewhere, crank up that computing power, and get out an aligned AGI?

 

Section B.1:  The distributional leap. 

10.  You can't train alignment by running lethally dangerous cognitions, observing whether the outputs kill or deceive or corrupt the operators, assigning a loss, and doing supervised learning.  On anything like the standard ML paradigm, you would need to somehow generalize optimization-for-alignment you did in safe conditions, across a big distributional shift to dangerous conditions.  (Some generalization of this seems like it would have to be true even outside that paradigm; you wouldn't be working on a live unaligned superintelligence to align it.)  This alone is a point that is sufficient to kill a lot of naive proposals from people who never did or could concretely sketch out any specific scenario of what training they'd do, in order to align what output - which is why, of course, they never concretely sketch anything like that.  Powerful AGIs doing dangerous things that will kill you if misaligned, must have an alignment property that generalized far out-of-distribution from safer building/training operations that didn't kill you.  This is where a huge amount of lethality comes from on anything remotely resembling the present paradigm.  Unaligned operation at a dangerous level of intelligence*capability will kill you; so, if you're starting with an unaligned system and labeling outputs in order to get it to learn alignment, the training regime or building regime must be operating at some lower level of intelligence*capability that is passively safe, where its currently-unaligned operation does not pose any threat.  (Note that anything substantially smarter than you poses a threat given any realistic level of capability.  Eg, "being able to produce outputs that humans look at" is probably sufficient for a generally much-smarter-than-human AGI to navigate its way out of the causal systems that are humans, especially in the real world where somebody trained the system on terabytes of Internet text, rather than somehow keeping it ignorant of the latent causes of its source code and training environments.)

11.  If cognitive machinery doesn't generalize far out of the distribution where you did tons of training, it can't solve problems on the order of 'build nanotechnology' where it would be too expensive to run a million training runs of failing to build nanotechnology.  There is no pivotal act this weak; there's no known case where you can entrain a safe level of ability on a safe environment where you can cheaply do millions of runs, and deploy that capability to save the world and prevent the next AGI project up from destroying the world two years later.  Pivotal weak acts like this aren't known, and not for want of people looking for them.  So, again, you end up needing alignment to generalize way out of the training distribution - not just because the training environment needs to be safe, but because the training environment probably also needs to be cheaper than evaluating some real-world domain in which the AGI needs to do some huge act.  You don't get 1000 failed tries at burning all GPUs - because people will notice, even leaving out the consequences of capabilities success and alignment failure.

12.  Operating at a highly intelligent level is a drastic shift in distribution from operating at a less intelligent level, opening up new external options, and probably opening up even more new internal choices and modes.  Problems that materialize at high intelligence and danger levels may fail to show up at safe lower levels of intelligence, or may recur after being suppressed by a first patch.

13.  Many alignment problems of superintelligence will not naturally appear at pre-dangerous, passively-safe levels of capability.  Consider the internal behavior 'change your outer behavior to deliberately look more aligned and deceive the programmers, operators, and possibly any loss functions optimizing over you'.  This problem is one that will appear at the superintelligent level; if, being otherwise ignorant, we guess that it is among the median such problems in terms of how early it naturally appears in earlier systems, then around half of the alignment problems of superintelligence will first naturally materialize after that one first starts to appear.  Given correct foresight of which problems will naturally materialize later, one could try to deliberately materialize such problems earlier, and get in some observations of them.  This helps to the extent (a) that we actually correctly forecast all of the problems that will appear later, or some superset of those; (b) that we succeed in preemptively materializing a superset of problems that will appear later; and (c) that we can actually solve, in the earlier laboratory that is out-of-distribution for us relative to the real problems, those alignment problems that would be lethal if we mishandle them when they materialize later.  Anticipating all of the really dangerous ones, and then successfully materializing them, in the correct form for early solutions to generalize over to later solutions, sounds possibly kinda hard.

14.  Some problems, like 'the AGI has an option that (looks to it like) it could successfully kill and replace the programmers to fully optimize over its environment', seem like their natural order of appearance could be that they first appear only in fully dangerous domains.  Really actually having a clear option to brain-level-persuade the operators or escape onto the Internet, build nanotech, and destroy all of humanity - in a way where you're fully clear that you know the relevant facts, and estimate only a not-worth-it low probability of learning something which changes your preferred strategy if you bide your time another month while further growing in capability - is an option that first gets evaluated for real at the point where an AGI fully expects it can defeat its creators.  We can try to manifest an echo of that apparent scenario in earlier toy domains.  Trying to train by gradient descent against that behavior, in that toy domain, is something I'd expect to produce not-particularly-coherent local patches to thought processes, which would break with near-certainty inside a superintelligence generalizing far outside the training distribution and thinking very different thoughts.  Also, programmers and operators themselves, who are used to operating in not-fully-dangerous domains, are operating out-of-distribution when they enter into dangerous ones; our methodologies may at that time break.

15.  Fast capability gains seem likely, and may break lots of previous alignment-required invariants simultaneouslyGiven otherwise insufficient foresight by the operators, I'd expect a lot of those problems to appear approximately simultaneously after a sharp capability gain.  See, again, the case of human intelligence.  We didn't break alignment with the 'inclusive reproductive fitness' outer loss function, immediately after the introduction of farming - something like 40,000 years into a 50,000 year Cro-Magnon takeoff, as was itself running very quickly relative to the outer optimization loop of natural selection.  Instead, we got a lot of technology more advanced than was in the ancestral environment, including contraception, in one very fast burst relative to the speed of the outer optimization loop, late in the general intelligence game.  We started reflecting on ourselves a lot more, started being programmed a lot more by cultural evolution, and lots and lots of assumptions underlying our alignment in the ancestral training environment broke simultaneously.  (People will perhaps rationalize reasons why this abstract description doesn't carry over to gradient descent; eg, “gradient descent has less of an information bottleneck”.  My model of this variety of reader has an inside view, which they will label an outside view, that assigns great relevance to some other data points that are not observed cases of an outer optimization loop producing an inner general intelligence, and assigns little importance to our one data point actually featuring the phenomenon in question.  When an outer optimization loop actually produced general intelligence, it broke alignment after it turned general, and did so relatively late in the game of that general intelligence accumulating capability and knowledge, almost immediately before it turned 'lethally' dangerous relative to the outer optimization loop of natural selection.  Consider skepticism, if someone is ignoring this one warning, especially if they are not presenting equally lethal and dangerous things that they say will go wrong instead.)

 

Section B.2:  Central difficulties of outer and inner alignment. 

16.  Even if you train really hard on an exact loss function, that doesn't thereby create an explicit internal representation of the loss function inside an AI that then continues to pursue that exact loss function in distribution-shifted environments.  Humans don't explicitly pursue inclusive genetic fitness; outer optimization even on a very exact, very simple loss function doesn't produce inner optimization in that direction This happens in practice in real life, it is what happened in the only case we know about, and it seems to me that there are deep theoretical reasons to expect it to happen again: the first semi-outer-aligned solutions found, in the search ordering of a real-world bounded optimization process, are not inner-aligned solutions.  This is sufficient on its own, even ignoring many other items on this list, to trash entire categories of naive alignment proposals which assume that if you optimize a bunch on a loss function calculated using some simple concept, you get perfect inner alignment on that concept.

17.  More generally, a superproblem of 'outer optimization doesn't produce inner alignment' is that on the current optimization paradigm there is no general idea of how to get particular inner properties into a system, or verify that they're there, rather than just observable outer ones you can run a loss function over.  This is a problem when you're trying to generalize out of the original training distribution, because, eg, the outer behaviors you see could have been produced by an inner-misaligned system that is deliberately producing outer behaviors that will fool you.  We don't know how to get any bits of information into the inner system rather than the outer behaviors, in any systematic or general way, on the current optimization paradigm.

18.  There's no reliable Cartesian-sensory ground truth (reliable loss-function-calculator) about whether an output is 'aligned', because some outputs destroy (or fool) the human operators and produce a different environmental causal chain behind the externally-registered loss function.  That is, if you show an agent a reward signal that's currently being generated by humans, the signal is not in generalreliable perfect ground truth about how aligned an action was, because another way of producing a high reward signal is to deceive, corrupt, or replace the human operators with a different causal system which generates that reward signal.  When you show an agent an environmental reward signal, you are not showing it something that is a reliable ground truth about whether the system did the thing you wanted it to do; even if it ends up perfectly inner-aligned on that reward signal, or learning some concept that exactly corresponds to 'wanting states of the environment which result in a high reward signal being sent', an AGI strongly optimizing on that signal will kill you, because the sensory reward signal was not a ground truth about alignment (as seen by the operators).

19.  More generally, there is no known way to use the paradigm of loss functions, sensory inputs, and/or reward inputs, to optimize anything within a cognitive system to point at particular things within the environment - to point to latent events and objects and properties in the environment, rather than relatively shallow functions of the sense data and reward.  This isn't to say that nothing in the system’s goal (whatever goal accidentally ends up being inner-optimized over) could ever point to anything in the environment by accident Humans ended up pointing to their environments at least partially, though we've got lots of internally oriented motivational pointers as well.  But insofar as the current paradigm works at all, the on-paper design properties say that it only works for aligning on known direct functions of sense data and reward functions.  All of these kill you if optimized-over by a sufficiently powerful intelligence, because they imply strategies like 'kill everyone in the world using nanotech to strike before they know they're in a battle, and have control of your reward button forever after'.  It just isn't true that we know a function on webcam input such that every world with that webcam showing the right things is safe for us creatures outside the webcam.  This general problem is a fact about the territory, not the map; it's a fact about the actual environment, not the particular optimizer, that lethal-to-us possibilities exist in some possible environments underlying every given sense input.

20.  Human operators are fallible, breakable, and manipulable.  Human raters make systematic errors - regular, compactly describable, predictable errors.  To faithfully learn a function from 'human feedback' is to learn (from our external standpoint) an unfaithful description of human preferences, with errors that are not random (from the outside standpoint of what we'd hoped to transfer).  If you perfectly learn and perfectly maximize the referent of rewards assigned by human operators, that kills them.  It's a fact about the territory, not the map - about the environment, not the optimizer - that the best predictive explanation for human answers is one that predicts the systematic errors in our responses, and therefore is a psychological concept that correctly predicts the higher scores that would be assigned to human-error-producing cases.

21.  There's something like a single answer, or a single bucket of answers, for questions like 'What's the environment really like?' and 'How do I figure out the environment?' and 'Which of my possible outputs interact with reality in a way that causes reality to have certain properties?', where a simple outer optimization loop will straightforwardly shove optimizees into this bucket.  When you have a wrong belief, reality hits back at your wrong predictions.  When you have a broken belief-updater, reality hits back at your broken predictive mechanism via predictive losses, and a gradient descent update fixes the problem in a simple way that can easily cohere with all the other predictive stuff.  In contrast, when it comes to a choice of utility function, there are unbounded degrees of freedom and multiple reflectively coherent fixpoints.  Reality doesn't 'hit back' against things that are locally aligned with the loss function on a particular range of test cases, but globally misaligned on a wider range of test cases.  This is the very abstract story about why hominids, once they finally started to generalize, generalized their capabilities to Moon landings, but their inner optimization no longer adhered very well to the outer-optimization goal of 'relative inclusive reproductive fitness' - even though they were in their ancestral environment optimized very strictly around this one thing and nothing else.  This abstract dynamic is something you'd expect to be true about outer optimization loops on the order of both 'natural selection' and 'gradient descent'.  The central result:  Capabilities generalize further than alignment once capabilities start to generalize far.

22.  There's a relatively simple core structure that explains why complicated cognitive machines work; which is why such a thing as general intelligence exists and not just a lot of unrelated special-purpose solutions; which is why capabilities generalize after outer optimization infuses them into something that has been optimized enough to become a powerful inner optimizer.  The fact that this core structure is simple and relates generically to low-entropy high-structure environments is why humans can walk on the Moon.  There is no analogous truth about there being a simple core of alignment, especially not one that is even easier for gradient descent to find than it would have been for natural selection to just find 'want inclusive reproductive fitness' as a well-generalizing solution within ancestral humans.  Therefore, capabilities generalize further out-of-distribution than alignment, once they start to generalize at all.

23.  Corrigibility is anti-natural to consequentialist reasoning; "you can't bring the coffee if you're dead" for almost every kind of coffee.  We (MIRI) tried and failed to find a coherent formula for an agent that would let itself be shut down (without that agent actively trying to get shut down).  Furthermore, many anti-corrigible lines of reasoning like this may only first appear at high levels of intelligence.

24.  There are two fundamentally different approaches you can potentially take to alignment, which are unsolvable for two different sets of reasons; therefore, by becoming confused and ambiguating between the two approaches, you can confuse yourself about whether alignment is necessarily difficult.  The first approach is to build a CEV-style Sovereign which wants exactly what we extrapolated-want and is therefore safe to let optimize all the future galaxies without it accepting any human input trying to stop it.  The second course is to build corrigible AGI which doesn't want exactly what we want, and yet somehow fails to kill us and take over the galaxies despite that being a convergent incentive there.

  1. The first thing generally, or CEV specifically, is unworkable because the complexity of what needs to be aligned or meta-aligned for our Real Actual Values is far out of reach for our FIRST TRY at AGI.  Yes I mean specifically that the dataset, meta-learning algorithm, and what needs to be learned, is far out of reach for our first try.  It's not just non-hand-codable, it is unteachable on-the-first-try because the thing you are trying to teach is too weird and complicated.
  2. The second thing looks unworkable (less so than CEV, but still lethally unworkable) because corrigibility runs actively counter to instrumentally convergent behaviors within a core of general intelligence (the capability that generalizes far out of its original distribution).  You're not trying to make it have an opinion on something the core was previously neutral on.  You're trying to take a system implicitly trained on lots of arithmetic problems until its machinery started to reflect the common coherent core of arithmetic, and get it to say that as a special case 222 + 222 = 555.  You can maybe train something to do this in a particular training distribution, but it's incredibly likely to break when you present it with new math problems far outside that training distribution, on a system which successfully generalizes capabilities that far at all.

 

Section B.3:  Central difficulties of sufficiently good and useful transparency / interpretability.

25.  We've got no idea what's actually going on inside the giant inscrutable matrices and tensors of floating-point numbers.  Drawing interesting graphs of where a transformer layer is focusing attention doesn't help if the question that needs answering is "So was it planning how to kill us or not?"

26.  Even if we did know what was going on inside the giant inscrutable matrices while the AGI was still too weak to kill us, this would just result in us dying with more dignity, if DeepMind refused to run that system and let Facebook AI Research destroy the world two years later.  Knowing that a medium-strength system of inscrutable matrices is planning to kill us, does not thereby let us build a high-strength system of inscrutable matrices that isn't planning to kill us.

27.  When you explicitly optimize against a detector of unaligned thoughts, you're partially optimizing for more aligned thoughts, and partially optimizing for unaligned thoughts that are harder to detect.  Optimizing against an interpreted thought optimizes against interpretability.

28.  The AGI is smarter than us in whatever domain we're trying to operate it inside, so we cannot mentally check all the possibilities it examines, and we cannot see all the consequences of its outputs using our own mental talent.  A powerful AI searches parts of the option space we don't, and we can't foresee all its options.

29.  The outputs of an AGI go through a huge, not-fully-known-to-us domain (the real world) before they have their real consequences.  Human beings cannot inspect an AGI's output to determine whether the consequences will be good.

30.  Any pivotal act that is not something we can go do right now, will take advantage of the AGI figuring out things about the world we don't know so that it can make plans we wouldn't be able to make ourselves.  It knows, at the least, the fact we didn't previously know, that some action sequence results in the world we want.  Then humans will not be competent to use their own knowledge of the world to figure out all the results of that action sequence.  An AI whose action sequence you can fully understand all the effects of, before it executes, is much weaker than humans in that domain; you couldn't make the same guarantee about an unaligned human as smart as yourself and trying to fool you.  There is no pivotal output of an AGI that is humanly checkable and can be used to safely save the world but only after checking it; this is another form of pivotal weak act which does not exist.

31.  A strategically aware intelligence can choose its visible outputs to have the consequence of deceiving you, including about such matters as whether the intelligence has acquired strategic awareness; you can't rely on behavioral inspection to determine facts about an AI which that AI might want to deceive you about.  (Including how smart it is, or whether it's acquired strategic awareness.)

32.  Human thought partially exposes only a partially scrutable outer surface layer.  Words only trace our real thoughts.  Words are not an AGI-complete data representation in its native style.  The underparts of human thought are not exposed for direct imitation learning and can't be put in any dataset.  This makes it hard and probably impossible to train a powerful system entirely on imitation of human words or other human-legible contents, which are only impoverished subsystems of human thoughts; unless that system is powerful enough to contain inner intelligences figuring out the humans, and at that point it is no longer really working as imitative human thought.

33.  The AI does not think like you do, the AI doesn't have thoughts built up from the same concepts you use, it is utterly alien on a staggering scale.  Nobody knows what the hell GPT-3 is thinking, not only because the matrices are opaque, but because the stuff within that opaque container is, very likely, incredibly alien - nothing that would translate well into comprehensible human thinking, even if we could see past the giant wall of floating-point numbers to what lay behind.

 

Section B.4:  Miscellaneous unworkable schemes. 

34.  Coordination schemes between superintelligences are not things that humans can participate in (eg because humans can't reason reliably about the code of superintelligences); a "multipolar" system of 20 superintelligences with different utility functions, plus humanity, has a natural and obvious equilibrium which looks like "the 20 superintelligences cooperate with each other but not with humanity".

35.  Schemes for playing "different" AIs off against each other stop working if those AIs advance to the point of being able to coordinate via reasoning about (probability distributions over) each others' code.  Any system of sufficiently intelligent agents can probably behave as a single agent, even if you imagine you're playing them against each other.  Eg, if you set an AGI that is secretly a paperclip maximizer, to check the output of a nanosystems designer that is secretly a staples maximizer, then even if the nanosystems designer is not able to deduce what the paperclip maximizer really wants (namely paperclips), it could still logically commit to share half the universe with any agent checking its designs if those designs were allowed through, if the checker-agent can verify the suggester-system's logical commitment and hence logically depend on it (which excludes human-level intelligences).  Or, if you prefer simplified catastrophes without any logical decision theory, the suggester could bury in its nanosystem design the code for a new superintelligence that will visibly (to a superhuman checker) divide the universe between the nanosystem designer and the design-checker.

36.  What makes an air conditioner 'magic' from the perspective of say the thirteenth century, is that even if you correctly show them the design of the air conditioner in advance, they won't be able to understand from seeing that design why the air comes out cold; the design is exploiting regularities of the environment, rules of the world, laws of physics, that they don't know about.  The domain of human thought and human brains is very poorly understood by us, and exhibits phenomena like optical illusions, hypnosis, psychosis, mania, or simple afterimages produced by strong stimuli in one place leaving neural effects in another place.  Maybe a superintelligence couldn't defeat a human in a very simple realm like logical tic-tac-toe; if you're fighting it in an incredibly complicated domain you understand poorly, like human minds, you should expect to be defeated by 'magic' in the sense that even if you saw its strategy you would not understand why that strategy worked.  AI-boxing can only work on relatively weak AGIs; the human operators are not secure systems.

 

Section C:

Okay, those are some significant problems, but lots of progress is being made on solving them, right?  There's a whole field calling itself "AI Safety" and many major organizations are expressing Very Grave Concern about how "safe" and "ethical" they are?

 

37.  There's a pattern that's played out quite often, over all the times the Earth has spun around the Sun, in which some bright-eyed young scientist, young engineer, young entrepreneur, proceeds in full bright-eyed optimism to challenge some problem that turns out to be really quite difficult.  Very often the cynical old veterans of the field try to warn them about this, and the bright-eyed youngsters don't listen, because, like, who wants to hear about all that stuff, they want to go solve the problem!  Then this person gets beaten about the head with a slipper by reality as they find out that their brilliant speculative theory is wrong, it's actually really hard to build the thing because it keeps breaking, and society isn't as eager to adopt their clever innovation as they might've hoped, in a process which eventually produces a new cynical old veteran.  Which, if not literally optimal, is I suppose a nice life cycle to nod along to in a nature-show sort of way.  Sometimes you do something for the first time and there are no cynical old veterans to warn anyone and people can be really optimistic about how it will go; eg the initial Dartmouth Summer Research Project on Artificial Intelligence in 1956:  "An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves. We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer."  This is less of a viable survival plan for your planet if the first major failure of the bright-eyed youngsters kills literally everyone before they can predictably get beaten about the head with the news that there were all sorts of unforeseen difficulties and reasons why things were hard.  You don't get any cynical old veterans, in this case, because everybody on Earth is dead.  Once you start to suspect you're in that situation, you have to do the Bayesian thing and update now to the view you will predictably update to later: realize you're in a situation of being that bright-eyed person who is going to encounter Unexpected Difficulties later and end up a cynical old veteran - or would be, except for the part where you'll be dead along with everyone else.  And become that cynical old veteran right away, before reality whaps you upside the head in the form of everybody dying and you not getting to learn.  Everyone else seems to feel that, so long as reality hasn't whapped them upside the head yet and smacked them down with the actual difficulties, they're free to go on living out the standard life-cycle and play out their role in the script and go on being bright-eyed youngsters; there's no cynical old veterans to warn them otherwise, after all, and there's no proof that everything won't go beautifully easy and fine, given their bright-eyed total ignorance of what those later difficulties could be.

38.  It does not appear to me that the field of 'AI safety' is currently being remotely productive on tackling its enormous lethal problems.  These problems are in fact out of reach; the contemporary field of AI safety has been selected to contain people who go to work in that field anyways.  Almost all of them are there to tackle problems on which they can appear to succeed and publish a paper claiming success; if they can do that and get funded, why would they embark on a much more unpleasant project of trying something harder that they'll fail at, just so the human species can die with marginally more dignity?  This field is not making real progress and does not have a recognition function to distinguish real progress if it took place.  You could pump a billion dollars into it and it would produce mostly noise to drown out what little progress was being made elsewhere.

39.  I figured this stuff out using the null string as input, and frankly, I have a hard time myself feeling hopeful about getting real alignment work out of somebody who previously sat around waiting for somebody else to input a persuasive argument into them.  This ability to "notice lethal difficulties without Eliezer Yudkowsky arguing you into noticing them" currently is an opaque piece of cognitive machinery to me, I do not know how to train it into others.  It probably relates to 'security mindset', and a mental motion where you refuse to play out scripts, and being able to operate in a field that's in a state of chaos.

40.  "Geniuses" with nice legible accomplishments in fields with tight feedback loops where it's easy to determine which results are good or bad right away, and so validate that this person is a genius, are (a) people who might not be able to do equally great work away from tight feedback loops, (b) people who chose a field where their genius would be nicely legible even if that maybe wasn't the place where humanity most needed a genius, and (c) probably don't have the mysterious gears simply because they're rare.  You cannot just pay $5 million apiece to a bunch of legible geniuses from other fields and expect to get great alignment work out of them.  They probably do not know where the real difficulties are, they probably do not understand what needs to be done, they cannot tell the difference between good and bad work, and the funders also can't tell without me standing over their shoulders evaluating everything, which I do not have the physical stamina to do.  I concede that real high-powered talents, especially if they're still in their 20s, genuinely interested, and have done their reading, are people who, yeah, fine, have higher probabilities of making core contributions than a random bloke off the street. But I'd have more hope - not significant hope, but more hope - in separating the concerns of (a) credibly promising to pay big money retrospectively for good work to anyone who produces it, and (b) venturing prospective payments to somebody who is predicted to maybe produce good work later.

41.  Reading this document cannot make somebody a core alignment researcherThat requires, not the ability to read this document and nod along with it, but the ability to spontaneously write it from scratch without anybody else prompting you; that is what makes somebody a peer of its author.  It's guaranteed that some of my analysis is mistaken, though not necessarily in a hopeful direction.  The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so.  Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try.  I'm not particularly hopeful of this turning out to be true in real life, but I suppose it's one possible place for a "positive model violation" (miracle).  The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years into even normies starting to notice the death game, it is still Eliezer Yudkowsky writing up this list, says that humanity still has only one gamepiece that can do that.  I knew I did not actually have the physical stamina to be a star researcher, I tried really really hard to replace myself before my health deteriorated further, and yet here I am writing this.  That's not what surviving worlds look like.

42.  There's no plan.  Surviving worlds, by this point, and in fact several decades earlier, have a plan for how to survive.  It is a written plan.  The plan is not secret.  In this non-surviving world, there are no candidate plans that do not immediately fall to Eliezer instantly pointing at the giant visible gaping holes in that plan.  Or if you don't know who Eliezer is, you don't even realize you need a plan, because, like, how would a human being possibly realize that without Eliezer yelling at them?  It's not like people will yell at themselves about prospective alignment difficulties, they don't have an internal voice of caution.  So most organizations don't have plans, because I haven't taken the time to personally yell at them.  'Maybe we should have a plan' is deeper alignment mindset than they possess without me standing constantly on their shoulder as their personal angel pleading them into... continued noncompliance, in fact.  Relatively few are aware even that they should, to look better, produce a pretend plan that can fool EAs too 'modest' to trust their own judgments about seemingly gaping holes in what serious-looking people apparently believe.

43.  This situation you see when you look around you is not what a surviving world looks like.  The worlds of humanity that survive have plans.  They are not leaving to one tired guy with health problems the entire responsibility of pointing out real and lethal problems proactively.  Key people are taking internal and real responsibility for finding flaws in their own plans, instead of considering it their job to propose solutions and somebody else's job to prove those solutions wrong.  That world started trying to solve their important lethal problems earlier than this.  Half the people going into string theory shifted into AI alignment instead and made real progress there.  When people suggest a planetarily-lethal problem that might materialize later - there's a lot of people suggesting those, in the worlds destined to live, and they don't have a special status in the field, it's just what normal geniuses there do - they're met with either solution plans or a reason why that shouldn't happen, not an uncomfortable shrug and 'How can you be sure that will happen' / 'There's no way you could be sure of that now, we'll have to wait on experimental evidence.'

A lot of those better worlds will die anyways.  It's a genuinely difficult problem, to solve something like that on your first try.  But they'll die with more dignity than this.

AGI Ruin: A List of Lethalities
New Comment
716 comments, sorted by Click to highlight new comments since:
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings
[-]evhubΩ76251116

That requires, not the ability to read this document and nod along with it, but the ability to spontaneously write it from scratch without anybody else prompting you; that is what makes somebody a peer of its author. It's guaranteed that some of my analysis is mistaken, though not necessarily in a hopeful direction. The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so. Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try. I'm not particularly hopeful of this turning out to be true in real life, but I suppose it's one possible place for a "positive model violation" (miracle). The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years into even normies start

... (read more)
[-]VaniverΩ103911

I agree this list doesn't seem to contain much unpublished material, and I think the main value of having it in one numbered list is that "all of it is in one, short place", and it's not an "intro to computers can think" and instead is "these are a bunch of the reasons computers thinking is difficult to align".

The thing that I understand to be Eliezer's "main complaint" is something like: "why does it seem like No One Else is discovering new elements to add to this list?". Like, I think Risks From Learned Optimization was great, and am glad you and others wrote it! But also my memory is that it was "prompted" instead of "written from scratch", and I imagine Eliezer reading it more had the sense of "ah, someone made 'demons' palatable enough to publish" instead of "ah, I am learning something new about the structure of intelligence and alignment."

[I do think the claim that Eliezer 'figured it out from the empty string' doesn't quite jive with the Yudkowsky's Coming of Age sequence.]

Nearly empty string of uncommon social inputs.  All sorts of empirical inputs, including empirical inputs in the social form of other people observing things.

It's also fair to say that, though they didn't argue me out of anything, Moravec and Drexler and Ed Regis and Vernor Vinge and Max More could all be counted as social inputs telling me that this was an important thing to look at.

Thank you, Evan, for living the Virture of Scholarship. Your work is appreciated. 

Eliezer's post here is doing work left undone by the writing you cite. It is a much clearer account of how our mainline looks doomed than you'd see elsewhere, and it's frank on this point.

I think Eliezer wishes these sorts of artifacts were not just things he wrote, like this and "There is no fire alarm".

Also, re your excerpts for (14), (15), and (32), I see Eliezer as saying something meaningfully different in each case. I might elaborate under this comment.

Re (14), I guess the ideas are very similar, where the mesaoptimizer scenario is like a sharp example of the more general concept Eliezer points at, that different classes of difficulties may appear at different capability levels.

Re (15), "Fast capability gains seem likely, and may break lots of previous alignment-required invariants simultaneously", which is about how we may have reasons to expect aligned output that are brittle under rapid capability gain: your quote from Richard is just about "fast capability gain seems possible and likely", and isn't about connecting that to increased difficulty in succeeding at the alignment problem?

Re (32), I don't think your quote isn't talking about the thing Eliezer is talking about, which is that in order to be human level at modelling human-generated text, your AI must be doing something on par with human thought that figures out what humans would say. Your quote just isn't discussing this, namely that strong imitation requires cognition that is dangerous.

So I guess I don't take much issue with (14) or (15), but I think you're quite off the mark about (32). In any case, I still have a strong sense that Eliezer is successfully being more on the mark here than the rest of us manage. Kudos of course to you and others that are working on writing things up and figuring things out. Though I remain sympathetic to Eliezer's complaint.

2Eliezer Yudkowsky
Well, my disorganized list sure wasn't complete, so why not go ahead and list some of the foreseeable difficulties I left out?  Bonus points if any of them weren't invented by me, though I realize that most people may not realize how much of this entire field is myself wearing various trenchcoats.
[-]evhubΩ7226968

Sure—that's easy enough. Just off the top of my head, here's five safety concerns that I think are important that I don't think you included:

  • The fact that there exist functions that are easier to verify than satisfy ensures that adversarial training can never guarantee the absence of deception.

  • It is impossible to verify a model's safety—even given arbitrarily good transparency tools—without access to that model's training process. For example, you could get a deceptive model that gradient hacks itself in such a way that cryptographically obfuscates its deception.

  • It is impossible in general to use interpretability tools to select models to have a particular behavioral property. I think this is clear if you just stare at Rice's theorem enough: checking non-trivial behavioral properties, even with mechanistic access, is in general undecidable. Note, however, that this doesn't rule out checking a mechanistic property that implies a behavioral property.

  • Any prior you use to incentivize models to behave in a particular way doesn't necessarily translate to situations where that model itself runs another search over algorithms. For example, the fastest way to search for algorith

... (read more)

Consider my vote to be placed that you should turn this into a post, keep going for literally as long as you can, expand things to paragraphs, and branch out beyond things you can easily find links for.

(I do think there's a noticeable extent to which I was trying to list difficulties more central than those, but I also think many people could benefit from reading a list of 100 noncentral difficulties.)

I do think there's a noticeable extent to which I was trying to list difficulties more central than those

Probably people disagree about which things are more central, or as evhub put it:

Every time anybody writes up any overview of AI safety, they have to make tradeoffs [...] depending on what the author personally believes is most important/relevant to say

Now FWIW I thought evhub was overly dismissive of (4) in which you made an important meta-point:

EY: 4. We can't just "decide not to build AGI" because GPUs are everywhere, and knowledge of algorithms is constantly being improved and published; 2 years after the leading actor has the capability to destroy the world, 5 other actors will have the capability to destroy the world.  The given lethal challenge is to solve within a time limit, driven by the dynamic in which, over time, increasingly weak actors with a smaller and smaller fraction of total computing power, become able to build AGI and destroy the world.  Powerful actors all refraining in unison from doing the suicidal thing just delays this time limit - it does not lift it [...]

evhub: This is just answering a particular bad plan.

But I would add a criticism of my o... (read more)

(Note that these have a theme: you can't wrangle general computation / optimization. That's why I'm short universal approaches to AI alignment (approaches that aim at making general optimization safe by enforcing universal rules), and long existential approaches (approaches that try to find specific mechanisms that can be analytically seen to do the right thing).)

-1Remmelt
Eliezer: If you find that (for reasons still left explained) * ... selection of code for intentionality is coupled – over the long run, in mostly non-reverse-engineerable ways – to various/most of the physical/chemical properties  * ... of the molecular substrate through which discrete code is necessarily computed/expressed (via input and output channels of information/energy packet transmission), then given that  * ... the properties of the solid-state substrate (e.g. silicon-based hardware) computing AGI's code * ... differ from the properties of the substrate of humans (carbon-based wetware), a conclusion that follows is that * ... the intentionality being selected for in AGI over the long run * ... will diverge from the intentionality that was selected for in humans.
2Rob Bensinger
What do you mean by 'intentionality'? Per SEP, "In philosophy, intentionality is the power of minds and mental states to be about, to represent, or to stand for, things, properties and states of affairs." So I read your comment as saying, a la Searle, 'maybe AI can never think like a human because there's something mysterious and crucial about carbon atoms in particular, or about capital-b Biology, for doing reasoning.' This seems transparently silly to me -- I know of no reasonable argument for thinking carbon differs from silicon on this dimension -- and also not relevant to AGI risk. You can protest "but AlphaGo doesn't really understand Go!" until the cows come home, and it will still beat you at Go. You can protest "but you don't really understand killer nanobots!" until the cows come home, and superintelligent Unfriendly AI will still build the nanobots and kill you with them. By the same reasoning, Searle-style arguments aren't grounds for pessimism either. If Friendly AI lacks true intentionality or true consciousness or whatever, it can still do all the same mechanistic operations, and therefore still produce the same desirable good outcomes as if it had human-style intentionality or whatver.
1Remmelt
That’s not the argument. Give me a few days to write a response. There’s a minefield of possible misinterpretations here. However, the argumentation does undermine the idea that designing for mechanistic (alignment) operations is going to work. I’ll try and explain why.
2Remmelt
BTW, with ‘intentionality’, I meant something closer to everyday notions of ‘intentions one has’. Will more precisely define that meaning later. I should have checked for diverging definitions from formal fields. Thanks for catching that.
2Remmelt
If you happen to have time, this paper serves as useful background reading: https://royalsocietypublishing.org/doi/full/10.1098/rsif.2012.0869 Particularly note the shift from trivial self-replication (e.g. most computer viruses) to non-trivial self-replication (e.g. as through substrate-environment pathways to reproduction). None of this is sufficient for you to guess what the argumentation is (you might be able to capture a bit of it, along with a lot of incorrect and often implicit assumptions we must dig into). If you could call on some patience and openness to new ideas, I would really appreciate it! I am already bracing for a next misinterpretation (which is fine, if we can talk about that). I apologise for that I cannot find a viable way yet to throw out all the argumentation in one go, and also for that this will get a bit disorientating when we go through arguments step-by-step.
1Remmelt
Returning to this:   Key idea: Different basis of existence→ different drives→ different intentions→ different outcomes. @Rob, I wrote up a longer explanation here, which I prefer to discuss with you in private first.  Will email you a copy tomorrow in the next weeks.

I'm sorry to hear that your health is poor and you feel that this is all on you. Maybe you're right about the likelihood of doom, and even if I knew you were, I'd be sorry that it troubles you this way.

I think you've done an amazing job of building the AI safety field and now, even when the field has a degree of momentum of its own, it does seem to be less focused on doom than it should be, and I think you continuing to push people to focus on doom is valuable.

I don't think its easy to get people to take weird ideas seriously. I've had many experiences where I've had ideas about how people should change their approach to a project that weren't particularly far out and (in my view) were right for very straightforward reasons, and yet for the most part I was ignored altogether. What you've accomplished in building the AI safety field is amazing because AI doom ideas seemed really crazy when you started talking about them.

Nevertheless, I think some of the things you've said in this post are counterproductive. Most of the post is good, but insulting people who might contribute to solving the problem is not, nor is demanding that people acknowledge that you are smarter than they are. I'... (read more)

There's a point here about how fucked things are that I do not know how to convey without saying those things, definitely not briefly or easily.  I've spent, oh, a fair number of years, being politer than this, and less personal than this, and the end result is that people nod along and go on living their lives.

I expect this won't work either, but at some point you start trying different things instead of the things that have already failed.  It's more dignified if you fail in different ways instead of the same way.

[-]lc8845

FWIW you taking off the Mr. Nice guy gloves has actually made me make different life decisions. I'm glad you tried it even if it doesn't work.

Do whatever you want, obviously, but I just want to clarify that I did not suggest you avoid personally criticising people (only that you avoid vague/hard to interpret criticism) or saying you think doom is overwhelmingly likely. Some other comments give me a stronger impression than yours that I was asking you in a general sense to be nice, but I'm saying it to you because I figure it mostly matters that you're clear on this.

7Chris_Leong
You might not have this ability, but surely you know at least one person who does?

I vehemently disagree here, based on my personal and generalizable or not history. I will illustrate with the three turning points of my recent life.

First step: I stumbled upon HPMOR, and Eliezer way of looking straight into the irrationality of all our common ways of interacting and thinking was deeply shocking. It made me feel like he was in a sense angrily pointing at me, who worked more like one of the PNJ rather than Harry. I heard him telling me you're dumb and all your ideals of making intelligent decisions, being the gifted kid and being smarter than everyone are all are just delusions. You're so out of touch with reality on so many levels, where to even start.

This attitude made me embark on a journey to improve myself, read the sequences, pledge on Giving What we can after knowing EA for many years, and overall reassess whether I was striving towards my goal of helping people (spoiler: I was not).

Second step:  The April fools post also shocked me on so many levels. I was once again deeply struck by the sheer pessimism of this figure I respected so much. After months of reading articles on LessWrong and so many about AI alignment, this was the one that made me terrifie... (read more)

I disagree strongly. To me it seems that AI safety has long punched below its weight because its proponents are unwilling to be confrontational, and are too reluctant to put moderate social pressure on people doing the activities which AI safety proponents hold to be very extremely bad. It is not a coincidence that among AI safety proponents, Eliezer is both unusually confrontational and unusually successful.

This isn't specific to AI safety. A lot of people in this community generally believe that arguments which make people feel bad are counterproductive because people will be "turned off".

This is false. There are tons of examples of disparaging arguments against bad (or "bad") behavior that succeed wildly. Such arguments very frequently succeed in instilling individual values like e.g. conscientiousness or honesty. Prominent political movements which use this rhetoric abound. When this website was young, Eliezer and many others participated in an aggressive campaign of discourse against religious ideas, and this campaign accomplished many of its goals. I could name many many more large and small examples. I bet you can too.

Obviously this isn't to say that confrontational and insu... (read more)

I think there's an important distinction between:

  • Deliberately phrasing things in confrontational or aggressive ways, in the hope that this makes your conversation partner "wake up" or something.
  • Choosing not to hide real, potentially-important beliefs you have about the world, even though those beliefs are liable to offend people, liable to be disagreed with, etc.

Either might be justifiable, but I'm a lot more wary of heuristics like "it's never OK to talk about individuals' relative proficiency at things, even if it feels very cruxy and important, because people just find the topic too triggering" than of heuristics like "it's never OK to say things in ways that sound shouty or aggressive". I think cognitive engines can much more easily get by self-censoring their tone than self-censoring what topics are permissible to think or talk about.

1teageegeepea
How is "success" measured among AI safety proponents?

This kind of post scares away the person who will be the key person in the AI safety field if we define "key person" as the genius main driver behind solving it, not the loudest person.  Which is rather unfortunate, because that person is likely to read this post at some point.

I don't believe this post has any "dignity", whatever weird obscure definition dignity has been given now. It's more like flailing around in death throes while pointing fingers and lauding yourself than it is a solemn battle stance against an oncoming impossible enemy.

For context, I'm not some Eliezer hater, I'm a young person doing an ML masters currently who just got into this space and within the past week have become a huge fan of Eliezer Yudkowsky's earlier work while simultaneously very disappointed in the recent, fruitless, output.

It seems worth doing a little user research on this to see how it actually affects people. If it is a net positive, then great. If it is a net negative, the question becomes how big of a net negative it is and whether it is worth the extra effort to frame things more nicely.

5Eli Tyre
I think this was excellently worded, and I'm glad you said it. I'm also glad to have read all the responses, many of which seem important and on point to me. I strong upvoted this comment as well as several of the responses. I'm leaving this comment, because I want to give you some social reinforcement for saying what you said, and saying it as clearly and tactfully as you did. 
2Yitz
Strongly agree with this, said more eloquently than I was able to :)

I'd have more hope - not significant hope, but more hope - in separating the concerns of (a) credibly promising to pay big money retrospectively for good work to anyone who produces it, and (b) venturing prospective payments to somebody who is predicted to maybe produce good work later.

 

I desperately want to make this ecosystem exist, either as part of Manifold Markets, or separately. Some people call it "impact certificates" or "retroactive public goods funding"; I call it "equity for public goods", or "Manifund" in the specific case.

If anyone is interested in:

a) Being a retroactive funder for good work (aka bounties, prizes)

b) Getting funding through this kind of mechanism (aka income share agreements, angel investment)

c) Working on this project full time (full-stack web dev, ops, community management)

Please get in touch! Reply here, or message austin@manifold.markets~

I'm also on a team trying to build impact certificates/retroactive public goods funding and we are receiving a grant from an FTX Future Fund regrantor to make it happen!

If you're interested in learning more or contributing you can:

  • Read about our ongoing $10,000 retro-funding contest (Austin is graciously contributing to the prize pool)
  • Submit an EA Forum Post to this retro-funding contest (before July 1st)
  • Join our Discord to chat/ask questions
  • Read/Comment on our lengthy informational EA forum post "Towards Impact Markets"
[-]Matthew BarnettΩ36144-31

It's as good as time as any to re-iterate my reasons for disagreeing with what I see as the Yudkowskian view of future AI. What follows isn't intended as a rebuttal of any specific argument in this essay, but merely a pointer that I'm providing for readers, that may help explain why some people might disagree with the conclusion and reasoning contained within.

I'll provide my cruxes point-by-point,

  • I think raw intelligence, while important, is not the primary factor that explains why humanity-as-a-species is much more powerful than chimpanzees-as-a-species. Notably, humans were once much less powerful, in our hunter-gatherer days, but over time, through the gradual process of accumulating technology, knowledge, and culture, humans now possess vast productive capacities that far outstrip our ancient powers.

    Similarly, our ability to coordinate through language also plays a huge role in explaining our power compared to other animals. But, on a first approximation, other animals can't coordinate at all, making this distinction much less impressive. The first AGIs we construct will be born into a culture already capable of coordinating, and sharing knowledge, making the potential power di
... (read more)
[-]VaniverΩ136543

Notably, humans were once much less powerful, in our hunter-gatherer days, but over time, through the gradual process of accumulating technology, knowledge, and culture, humans now possess vast productive capacities that far outstrip our ancient powers.

Similarly, our ability to coordinate through language also plays a huge role in explaining our power compared to other animals. But, on a first approximation, other animals can't coordinate at all, making this distinction much less impressive. The first AGIs we construct will be born into a culture already capable of coordinating, and sharing knowledge, making the potential power difference between AGI and humans relatively much smaller than between humans and other animals, at least at first.

I basically buy the story that human intelligence is less useful that human coordination; i.e. it's the intelligence of "humanity" the entity that matters, with the intelligence of individual humans relevant only as, like, subcomponents of that entity.

But... shouldn't this mean you expect AGI civilization to totally dominate human civilization? They can read each other's source code, and thus trust much more deeply! They can transmit information... (read more)

[-]lc134

But... shouldn't this mean you expect AGI civilization to totally dominate human civilization? They can read each other's source code, and thus trust much more deeply! They can transmit information between them at immense bandwidths! They can clone their minds and directly learn from each other's experiences!

This is 100% correct, and part of why I expect the focus on superintelligence, while literally true, is bad for AI outreach. There's a much simpler (and empirically, in my experience, more convincing) explanation of why we lose to even an AI with an IQ of 110. It is Dath Ilan, and we are Earth. Coordination is difficult for humans and the easy part for AIs. 

I will note that Eliezer wrote That Alien Message a long time ago I think in part to try to convey the issue to this perspective, but it's mostly about "information-theoretic bounds are probably not going to be tight" in a simulation-y universe instead of "here's what coordination between computers looks like today". I do predict the coordination point would be good to include in more of the intro materials.

[-]BuckΩ5127

But... shouldn't this mean you expect AGI civilization to totally dominate human civilization? They can read each other's source code, and thus trust much more deeply! They can transmit information between them at immense bandwidths! They can clone their minds and directly learn from each other's experiences!

I don't think it's obvious that this means that AGI is more dangerous, because it means that for a fixed total impact of AGI, the AGI doesn't have to be as competent at individual thinking (because it leans relatively more on group thinking). And so at the point where the AGIs are becoming very powerful in aggregate, this argument pushes us away from thinking they're good at individual thinking.

Also, it's not obvious that early AIs will actually be able to do this if their creators don't find a way to train them to have this affordance. ML doesn't currently normally make AIs which can helpfully share mind-states, and it probably requires non-trivial effort to hook them up correctly to be able to share mind-state.

6Anthony DiGiovanni
Being able to read source code doesn't automatically increase trust—you also have to be able to verify that the code being shared with you actually governs the AGI's behavior, despite that AGI's incentives and abilities to fool you. (Conditional on the AGIs having strongly aligned goals with each other, sure, this degree of transparency would help them with pure coordination problems.)

Nice! Thanks! I'll give my commentary on your commentary, also point by point. Your stuff italicized, my stuff not. Warning: Wall of text incoming! :)

I think raw intelligence, while important, is not the primary factor that explains why humanity-as-a-species is much more powerful than chimpanzees-as-a-species. Notably, humans were once much less powerful, in our hunter-gatherer days, but over time, through the gradual process of accumulating technology, knowledge, and culture, humans now possess vast productive capacities that far outstrip our ancient powers.

Similarly, our ability to coordinate through language also plays a huge role in explaining our power compared to other animals. But, on a first approximation, other animals can't coordinate at all, making this distinction much less impressive. The first AGIs we construct will be born into a culture already capable of coordinating, and sharing knowledge, making the potential power difference between AGI and humans relatively much smaller than between humans and other animals, at least at first.

I don't think I understand this argument. Yes, humans can use language to coordinate & benefit from cultural evolution, so an AI that... (read more)

5Chris van Merwijk
"I have sat down to make toy models .." reference?
7Daniel Kokotajlo
? I am the reference, I'm describing a personal experience.
1Chris van Merwijk
I meant, is there a link to where you've written this down somewhere? Maybe you just haven't written it down. 
2Daniel Kokotajlo
I'll send you a DM.
4Kinrany
Markdown has syntax for quotes: a line with > this on it will look like
3Paul Kent
  Can I get a link or two to read more about this incident?
2Daniel Kokotajlo
It's not so much an incident as a trend. I haven't investigated it myself, but I've read lots of people making this claim, citing various studies, etc. See e.g. "The social dilemma" by Tristan Harris. There's an academic literature on the subject now which I haven't read but which you can probably find by googling. I just did a quick search and found graphs like this: Presumably not all of the increase in deaths is due to Facebook; presumably it's multi-causal blah blah blah. But even if Facebook is responsible for a tiny fraction of the increase, that would mean Facebook was responsible for thousands of deaths.

You said you weren't replying to any specific point Eliezer was making, but I think it's worth pointing out that when he brings up Alpha Go, he's not talking about the 2 years it took Google to build a Go-playing AI - remarkable and surprising as that was - but rather the 3 days it took Alpha Zero to go from not knowing anything about the game beyond the basic rules to being better than all humans and the earlier AIs.

[-]lc197

I hate how convincing so many different people are. I wish I just had some fairly static, reasoned perspective based on object-level facts and not persuasion strings.

[This comment is no longer endorsed by its author]Reply

Note that convincing is a 2-place word. I don't think I can transfer this ability, but I haven't really tried, so here's a shot:

The target is: "reading as dialogue." Have a world-model. As you read someone else, be simultaneously constructing / inferring "their world-model" and holding "your world-model", noting where you agree and disagree.

If you focus too much on "how would I respond to each line", you lose the ability to listen and figure out what they're actually pointing at. If you focus too little on "how would I respond to this", you lose the ability to notice disagreements, holes, and notes of discord.

The first homework exercise I'd try to printing out something (probably with double-spacing), and writing your thoughts each sentence. "uh huh", "wait what?", "yes and", "no but", etc.; at the beginning you're probably going to be alternating between the two moves before you can do them simultaneously.

[Historically, I think I got this both from 'reading a lot', including a lot of old books, and also 'arguing on the internet' in forum environments that only sort of exist today, which was a helpful feedback loop for the relevant subskills, and of course whatever background factors made me do those activities.]

2lc
Why can't I delete comments sometimes? >:(
5Raemon
Users can't delete their own comments if the comment has been replied to, to avoid disrupting other people's content. (you can edit it to be blank though, or mark it as retracted)
[-]leogaoΩ7187

Some quick thoughts on these points:

  • I think the ability for humans to communicate and coordinate is a double edged sword. In particular, it enables the attack vector of dangerous self propagating memes. I expect memetic warfare to play a major role in many of the failure scenarios I can think of. As we've seen, even humans are capable of crafting some pretty potent memes, and even defending against human actors is difficult.
  • I think it's likely that the relevant reference class here is research bets rather then the "task" of AGI. An extremely successful research bet could be currently underinvested in, but once it shows promise, discontinuous (relative to the bet) amounts of resources will be dumped into scaling it up, even if the overall investment towards the task as a whole remains continuous. In other words, in this case even though investment into AGI may be continuous (though that might not even hold), discontinuity can occur on the level of specific research bets. Historical examples would include imagenet seeing discontinuous improvement with AlexNet despite continuous investment into image recognition to that point. (Also, for what it's worth, my personal model of AI doo
... (read more)
9emanuele ascani
Thanks a lot for writing this.  These disagreements mainly concern the relative power of future AIs, the polarity of takeoff, takeoff speed, and, in general, the shape of future AIs. Do you also have detailed disagreements about the difficulty of alignment? If anything, the fact that the future unfolds differently in your view should impact future alignment efforts (but you also might have other considerations informing your view on alignment). You partially answer this in the last point, saying: "But, equally, one could view these theses pessimistically." But what do you personally think? Are you more pessimistic, more optimistic, or equally pessimistic about humanity's chances of surviving AI progress? And why?

Part of what makes it difficult for me to talk about alignment difficultly is that the concept doesn’t fit easily into my paradigm of thinking about the future of AI. If I am correct, for example, that AI services will be modular, marginally more powerful than what comes before, and numerous as opposed to monolithic, then there will not be one alignment problem, but many.

I could talk about potential AI safety principles, healthy cultural norms, and specific engineering issues, but not “a problem” called “aligning the AI” — a soft prerequisite for explaining how difficult “the problem” will be. Put another way, my understanding is that future AI alignment will be continuous with ordinary engineering, like cars and skyscrapers. We don’t ordinarily talk about how hard the problem of building a car is, in some sort of absolute sense, though there are many ways of operationalizing what that could mean.

One question is how costly it is to build a car. We could then compare that cost to the overall consumer benefit that people get from cars, and from that, deduce whether and how many cars will be built. Similarly, we could ask about the size of the “alignment tax” (the cost of aligning an ... (read more)

2Emrik
Btw, your top-level comment is one of the best comments I've come across ever. Probably. Top 5? Idk, I'll check how I feel tomorrow. Aspiring to read everything you've ever written rn. Incidentally, you mention that And I've been thinking lately about how important it is to prioritise original thinking before you've consumed all the established literature in an active field of research.[1] If you manage to diverge early, the novelty of your perspective compounds over time (feel free to ask about my model) and you're more likely to end up with a productively different paradigm from what's already out there. Did you ever feel embarrassed trying to think for yourself when you didn't feel like you had read enough? Or, did you feel like other people might have expected you to feel embarrassed for how seriously you took your original thoughts, given how early you were in your learning arc? 1. ^ I'm not saying you haven't. I'm just guessing that you acquired your paradigm by doing original thinking early, and thus had the opportunity to diverge early, rather than greedily over-prioritising the consumption of existing literature in order to "reach the frontier". Once having hastily consumed someone else's paradigm, it's much harder to find its flaws and build something else from the ground up.
5Vishrut Arya
hi Matt! on the coordination crux, you say  but wouldn’t an AGI be able to coordinate and do knowledge sharing with humans because  a) it can impersonate being a human online and communicate with them via text and speech and  b) it‘ll realize such coordination is vital to accomplish it‘s goals and so it’ll do the necessary acculturation?  Watching all the episodes of Friends or reading all the social media posts by the biggest influencers, as examples. 
4Emrik
One reason that a fully general AGI might be more profitable than specialised AIs, despite obvious gains-from-specialisation, is if profitability depends on insight-production. For humans, it's easier to understand a particular thing the more other things you understand. One of the main ways you make novel intellectual progress is by combining remote associations from models about different things. Insight-ability for a particular novel task grows with the number of good models you have available to draw connections between. But, it could still be that the gains from increased generalisation for a particular model grows too slowly and can't compete with obvious gains from specialised AIs.
2David Johnston
Slightly relatedly, I think it's possible that "causal inference is hard". The idea is: once someone has worked something out, they can share it and people can pick it up easily, but it's hard to figure the thing out to begin with - even with a lot of prior experience and efficient inference, most new inventions still need a lot of trial and error. Thus the reason the process of technology accumulation is gradual is, crudely, because causal inference is hard. Even if this is true, one way things could still go badly is if most doom scenarios are locked behind a bunch of hard trial and error, but the easiest one isn't. On the other hand, if both of these things are true then there could be meaningful safety benefits gained from censoring certain kinds of data.
2[anonymous]
This is what struck me as the least likely to be true from the above AI doom scenario. Is diamondoid nanotechnology possible?  Very likely it is or something functionally equivalent.   Can a sufficiently advanced superintelligence infer how to build it from scratch solely based on human data?  Or will it need a large R&D center with many, many robotic systems that conduct experiments in parallel to extract the information required about our specific details of physics in our actual universe.  Not the very slightly incorrect approximations a simulator will give you.   The 'huge R&D center so big you can't see the end of it' is somewhat easier to regulate the 'invisible dust the AI assembles with clueless stooges'.
8Marion Z.
Any individual doomsday mechanism we can think of, I would agree is not nearly so simple for an AGI to execute as Yudkowsky implies. I do think that it's quite likely we're just not able to think of mechanisms even theoretically that an AGI could think of,  and one or more of those might actually be quite easy to do secretly and quickly. I wouldn't call it guaranteed by any means, but intuitively this seems like the sort of thing that raw cognitive power might have a significant bearing on.
5[anonymous]
I agree. One frightening mechanism I thought of is : "ok, assume the AGI can't craft the bioweapon or nanotechnology killbots without collecting vast amounts of information through carefully selected and performed experiments. (Basically enormous complexes full of robotics). How does it get the resources it needs? And the answer is it scams humans into doing it. We have many examples of humans trusting someone they shouldn't even when the evidence was readily available that they shouldn't.
1Keenmaster
Any “huge R&D center” constraint is trivialized in a future where agile, powerful robots will be ubiquitous and an AGI can use robots to create an underground lab in the middle of nowhere, using its superintelligence to be undetectable in all ways that are physically possible. An AGI will also be able to use robots and 3D printers to fabricate purpose-built machines that enable it to conduct billions of physical experiments a day. Sure, it would be harder to construct something like a massive particle accelerator, but 1) that isn’t needed to make killer nanobots 2) even that isn’t impossible for a sufficiently intelligent machine to create covertly and quickly.
[-]Vanessa KosoyΩ2711430

First, some remarks about the meta-level:

The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so. Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try. I'm not particularly hopeful of this turning out to be true in real life, but I suppose it's one possible place for a "positive model violation" (miracle). The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years into even normies starting to notice the death game, it is still Eliezer Yudkowsky writing up this list, says that humanity still has only one gamepiece that can do that.

Actually, I don't feel like I learned that much reading this list, compared to what I already knew. [EDIT: To be clear, this know... (read more)

There is a big chunk of what you're trying to teach which not weird and complicated, namely: "find this other agent, and what their values are". Because, "agents" and "values" are natural concepts, for reasons strongly related to "there's a relatively simple core structure that explains why complicated cognitive machines work".

This seems like it must be true to some degree, but "there is a big chunk" feels a bit too strong to me.

Possibly we don't disagree, and just have different notions of what a "big chunk" is. But some things that make the chunk feel smaller to me:

  • Humans are at least a little coherent, or we would never get anything done; but we aren't very coherent, so the project of piecing together 'what does the human brain as a whole "want"' can be vastly more difficult than the problem of figuring out what a coherent optimizer wants.
  • There are shards of planning and optimization and goal-oriented-ness in a cat's brain, but 'figure out what utopia would look like for a cat' is a far harder problem than 'identify all of the goal-encoding parts of the cat's brain and "read off" those goals'. E.g., does 'identifying utopia' in this context involve uplifting or extrapolating the
... (read more)

Humans are at least a little coherent, or we would never get anything done; but we aren't very coherent, so the project of piecing together 'what does the human brain as a whole "want"' can be vastly more difficult than the problem of figuring out what a coherent optimizer wants.

This is a point where I feel like I do have a substantial disagreement with the "conventional wisdom" of LessWrong.

First, LessWrong began with a discussion of cognitive biases in human irrationality, so this naturally became a staple of the local narrative. On the other hand, I think that a lot of presumed irrationality is actually rational but deceptive behavior (where the deception runs so deep that it's part of even our inner monologue). There are exceptions, like hyperbolic discounting, but not that many.

Second, the only reason why the question "what X wants" can make sense at all, is because X is an agent. As a corollary, it only makes sense to the extent that X is an agent. Therefore, if X is not entirely coherent then X's preferences are only approximately defined, and hence we only need to infer them approximately. So, the added difficulty of inferring X's preferences, resulting from the partial ... (read more)

Second, the only reason why the question "what X wants" can make sense at all, is because X is an agent. As a corollary, it only makes sense to the extent that X is an agent.

I'm not sure this is true; or if it's true, I'm not sure it's relevant. But assuming it is true...

Therefore, if X is not entirely coherent then X's preferences are only approximately defined, and hence we only need to infer them approximately.

... this strikes me as not capturing the aspect of human values that looks strange and complicated. Two ways I could imagine the strangeness and complexity cashing out as 'EU-maximizer-ish' are:

  • Maybe I sort-of contain a lot of subagents, and 'my values' are the conjunction of my sub-agents' values (where they don't conflict), plus the output of an idealized negotiation between my sub-agents (where they do conflict).
  • Alternatively, maybe I have a bunch of inconsistent preferences, but I have a complicated pile of meta-preferences that collectively imply some chain of self-modifications and idealizations that end up producing something more coherent and utility-function-ish after a long sequence of steps.

In both cases, the fact that my brain isn't a single coherent EU maximiz... (read more)

7Vanessa Kosoy
If we go down that path then it becomes the sort of conversation where I have no idea what common assumptions do we have, if any, that we could use to agree. As a general rule, I find it unconstructive, for the purpose of trying to agree on anything, to say things like "this (intuitively compelling) assumption is false" unless you also provide a concrete argument or an alternative of your own. Otherwise the discussion is just ejected into vacuum. Which is to say, I find it self-evident that "agents" are exactly the sort of beings that can "want" things, because agency is about pursuing objectives and wanting is about the objectives that you pursue. If you don't believe this then I don't know what these words even mean for you. Maybe, and maybe this means we need to treat "composite agents" explicitly in our models. But, there is also a case to be made that groups of (super)rational agents effectively converge into a single utility function, and if this is true, then the resulting system can just as well be interpreted as a single agent having this effective utility function, which is a solution that should satisfy the system of agents according to their existing bargaining equilibrium. If your agent converges to optimal behavior asymptotically, then I suspect it's still going to have infinite g and therefore an asymptotically-crisply-defined utility function. Of course it doesn't help on its own. What I mean is, we are going to find a precise mathematical formalization of this concept and then hard-code this formalization into our AGI design.
5Rob Bensinger
Fair enough! I don't think I agree in general, but I think 'OK, but what's your alternative to agency?' is an especially good case for this heuristic. The first counter-example that popped into my head was "a mind that lacks any machinery for considering, evaluating, or selecting actions; but it does have machinery for experiencing more-pleasurable vs. less pleasurable states". This is a mind we should be able to build, even if it would never evolve naturally. Possibly this still qualifies as an "agent" that "wants" and "pursues" things, as you conceive it, even though it doesn't select actions?
9Vanessa Kosoy
My 0th approximation answer is: you're describing something logically incoherent, like a p-zombie. My 1st approximation answer is more nuanced. Words that, in the pre-Turing era, referred exclusively to humans (and sometimes animals, and fictional beings), such as "wants", "experiences" et cetera, might have two different referents. One referent is a natural concept, something tied into deep truths about how the universe (or multiverse) works. In particular, deep truths about the "relatively simple core structure that explains why complicated cognitive machines work". The other referent is something in our specifically-human "ontological model" of the world (technically, I imagine that to be an infra-POMDP that all our hypotheses our refinements of). Since the latter is a "shard" of the former produced by evolution, the two referents are related, but might not be the same. (For example, I suspect that cats lack natural!consciousness but have human!consciousness.) The creature you describe does not natural!want anything. You postulated that it is "experiencing more pleasurable and less pleasurable states", but there is no natural method that would label its states as such, or that would interpret them as any sort of "experience". On the other hand, maybe if this creature is designed as a derivative of the human brain, then it does human!want something, because our shard of the concept of "wanting" mislabels (relatively to natural!want) weird states that wouldn't occur in the ancestral environment. You can then ask, why should we design the AI to follow what we natural!want rather than what we human!want? To answer this, notice that, under ideal conditions, you converge to actions that maximize your natural!want, (more or less) according to definition of natural!want. In particular, under ideal conditions, you would build an AI that follows your natural!want. Hence, it makes sense to take a shortcut and "update now to the view you will predictably update to later":

On Twitter, Eric Rogstad wrote:

"the thing where it keeps being literally him doing this stuff is quite a bad sign"

I'm a bit confused by this part. Some thoughts on why it seems odd for him (or others) to express that sentiment...

1. I parse the original as, "a collection of EY's thoughts on why safe AI is hard". It's EY's thoughts, why would someone else (other than @robbensinger) write a collection of EY's thoughts?

(And if we generalize to asking why no-one else would write about why safe AI is hard, then what about Superintelligence, or the AI stuff in cold-takes, or ...?)

2. Was there anything new in this doc? It's prob useful to collect all in one place, but we don't ask, "why did no one else write this" for every bit of useful writing out there, right?

Why was it so overwhelmingly important that someone write this summary at this time, that we're at all scratching our heads about why no one else did it?

Copying over my reply to Eric:

My shoulder Eliezer (who I agree with on alignment, and who speaks more bluntly and with less hedging than I normally would) says:

  1. The list is true, to the best of my knowledge, and the details actually matter.

    Many civilizations try to make a canonical
... (read more)
5handoflixue
  I don't think making this list in 1980 would have been meaningful. How do you offer any sort of coherent, detailed plan for dealing with something when all you have is toy examples like Eliza?  We didn't even have the concept of machine learning back then - everything computers did in 1980 was relatively easily understood by humans, in a very basic step-by-step way. Making a 1980s computer "safe" is a trivial task, because we hadn't yet developed any technology that could do something "unsafe" (i.e. beyond our understanding). A computer in the 1980s couldn't lie to you, because you could just inspect the code and memory and find out the actual reality. What makes you think this would have been useful? Do we have any historical examples to guide us in what this might look like?

I think most worlds that successfully navigate AGI risk have properties like:

  • AI results aren't published publicly, going back to more or less the field's origin.
  • The research community deliberately steers toward relatively alignable approaches to AI, which includes steering away from approaches that look like 'giant opaque deep nets'.
    • This means that you need to figure out what makes an approach 'alignable' earlier, which suggests much more research on getting de-confused regarding alignable cognition.
      • Many such de-confusions will require a lot of software experimentation, but the kind of software/ML that helps you learn a lot about alignment as you work with it is itself a relatively narrow target that you likely need to steer towards deliberately, based on earlier, weaker deconfusion progress. I don't think having DL systems on hand to play with has helped humanity learn much about alignment thus far, and by default, I don't expect humanity to get much more clarity on this before AGI kills us.
  • Researchers focus on trying to predict features of future systems, and trying to get mental clarity about how to align such systems, rather than focusing on 'align ELIZA' just because ELIZA is
... (read more)
5Thomas Kwa
"most worlds that successfully navigate AGI risk" is kind of a strange framing to me.  For one thing, it represents p(our world | success) and we care about p(success | our world). To convert between the two you of course need to multiply by p(success) / p(our world). What's the prior distribution of worlds? This seems underspecified. For another, using the methodology "think about whether our civilization seems more competent than the problem is hard" or "whether our civilization seems on track to solve the problem" I might have forecast nuclear annihilation (not sure about this). The methodology seems to work when we're relatively certain about the level of difficulty on the mainline, so if I were more sold on that I would believe this more. It would still feel kind of weird though.
6Vaniver
I mean, I think many of the computing pioneers 'basically saw' AI risk. I noted some surprise that IJ Good didn't write the precursor to this list in 1980, and apparently Wikipedia claims there was an unpublished statement in 1998 about AI x-risk; it'd be interesting to see what it contains and how much it does or doesn't line up with our modern conception of why the problem is hard.

The historical figures who basically saw it (George Eliot 1879: "will the creatures who are to transcend and finally supersede us be steely organisms [...] performing with infallible exactness more than everything that we have performed with a slovenly approximativeness and self-defeating inaccuracy?"; Turing 1951: "At some stage therefore we should have to expect the machines to take control") seem to have done so in the spirit of speculating about the cosmic process. The idea of coming up with a plan to solve the problem is an additional act of audacity; that's not really how things have ever worked so far. (People make plans about their own lives, or their own businesses; at most, a single country; no one plans world-scale evolutionary transitions.)

5Andrew McKnight
I'm tempted to call this a meta-ethical failure. Fatalism, universal moral realism, and just-world intuitions seem to be the underlying implicit hueristics or principals that would cause this "cosmic process" thought-blocker.
3ESRogs
Why is this v0 and not https://arbital.com/explore/ai_alignment/, or the Sequences, or any of the documents that Evan links to here? That's part of what I meant to be responding to — not that this post is not useful, but that I don't see what makes it so special compared to all the other stuff that Eliezer and others have already written.
6ESRogs
To put it another way, I would agree that Eliezer has made (what seem to me like) world-historically-significant contributions to understanding and advocating for (against) AI risk. So, if 2007 Eliezer was asking himself, "Why am I the only one really looking into this?", I think that's a very reasonable question. But here in 2022, I just don't see this particular post as that significant of a contribution compared to what's already out there.
2ESRogs
Wrote a long comment here. (Which you've seen, but linking since your comment started as a response to me.)
[-]ESRogsΩ197010

-3.  I'm assuming you are already familiar with some basics, and already know what 'orthogonality' and 'instrumental convergence' are and why they're true.


I think this is actually the part that I most "disagree" with. (I put "disagree" in quotes, because there are forms of these theses that I'm persuaded by. However, I'm not so confident that they'll be relevant for the kinds of AIs we'll actually build.)

1. The smart part is not the agent-y part

It seems to me that what's powerful about modern ML systems is their ability to do data compression / pattern recognition. That's where the real cognitive power (to borrow Eliezer's term) comes from. And I think that this is the same as what makes us smart.

GPT-3 does unsupervised learning on text data. Our brains do predictive processing on sensory inputs. My guess (which I'd love to hear arguments against!) is that there's a true and deep analogy between the two, and that they lead to impressive abilities for fundamentally the same reason.

If so, it seems to me that that's where all the juice is. That's where the intelligence comes from. (In the past, I've called this the core smarts of our brains.)

On this view, all the agent-y, planful... (read more)

GPT-3 does unsupervised learning on text data. Our brains do predictive processing on sensory inputs. My guess (which I'd love to hear arguments against!) is that there's a true and deep analogy between the two, and that they lead to impressive abilities for fundamentally the same reason.

Agree that self-supervised learning powers both GPT-3 updates and human brain world-model updates (details & caveats). (Which isn’t to say that GPT-3 is exactly the same as the human brain world-model—there are infinitely many different possible ML algorithms that all update via self-supervised learning).

However…

If so, it seems to me that that's where all the juice is. That's where the intelligence comes from … if agency is not a fundamental part of intelligence, and rather something that can just be added in on top, or not, and if we're at a loss for how to either align a superintelligent agent with CEV or else make it corrigible, then why not try to avoid creating the agent part of superintelligent agent?

I disagree; I think the agency is necessary to build a really good world-model, one that includes new useful concepts that humans have never thought of.

Without the agency, some of the things ... (read more)

4ESRogs
Why is agency necessary for these things? If we follow Ought's advice and build "process-based systems [that] are built on human-understandable task decompositions, with direct supervision of reasoning steps", do you expect us to hit a hard wall somewhere that prevents these systems from creatively choosing things to think about, books to read, or better brainstorming strategies?
7Steven Byrnes
(Copying from here:) (Does that count as “agency”? I don’t know, it depends on what you mean by “agency”.) In terms of the “task decomposition” strategy, this might be a tricky to discuss because you probably have a more detailed picture in your mind than I do. I’ll try anyway. It seems to me that the options are: (1) the subprocess only knows its narrow task (“solve this symplectic geometry homework problem”), and is oblivious to the overall system goal (“design a better microscope”), or (2) the subprocess is aware of the overall system goal and chooses actions in part to advance it. In Case (2), I’m not sure this really counts as “task decomposition” in the first place, or how this would help with safety. In Case (1), yes I expect systems to hit a hard wall—I’m skeptical that tasks we care about decompose cleanly. For example, at my last job, I would often be part of a team inventing a new gizmo, and it was not at all unusual for me to find myself sketching out the algorithms and sketching out the link budget and scrutinizing laser spec sheets and scrutinizing FPGA spec sheets and nailing down end-user requirements, etc. etc. Not because I’m individually the best person at each of those tasks—or even very good!—but because sometimes a laser-related problem is best solved by switching to a different algorithm, or an FPGA-related problem is best solved by recognizing that the real end-user requirements are not quite what we thought, etc. etc. And that kind of design work is awfully hard unless a giant heap of relevant information and knowledge is all together in a single brain / world-model. In the case of my current job doing AI alignment research, I sometimes come across small self-contained tasks that could be delegated, but I would have no idea how to decompose most of what I do. (E.g. writing this comment!) Here’s John Wentworth making a similar point more eloquently: A possible example of a seemingly-hard-to-decompose task would be: Until 1948, no h
1David Johnston
FWIW self-supervised learning can be surprisingly capable of doing things that we previously only knew how to do with "agentic" designs. From that link: classification is usually done with an objective + an optimization procedure, but GPT-3 just does it.

For example, I claim that while AlphaGo could be said to be agent-y, it does not care about atoms. And I think that we could make it fantastically more superhuman at Go, and it would still not care about atoms. Atoms are just not in the domain of its utility function.

In particular, I don't think it has an incentive to break out into the real world to somehow get itself more compute, so that it can think more about its next move. It's just not modeling the real world at all. It's not even trying to rack up a bunch of wins over time. It's just playing the single platonic game of Go.

I would distinguish three ways in which different AI systems could be said to "not care about atoms":

  1. The system is thinking about a virtual object (e.g., a Go board in its head), and it's incapable of entertaining hypotheses about physical systems. Indeed, we might add the assumption that it can't entertain hypotheses like 'this Go board I'm currently thinking about is part of a larger universe' at all. (E.g., there isn't some super-Go-board I and/or the board are embedded in.)
  2. The system can think about atoms/physics, but it only terminally cares about digital things in a simulated environment (e.g., winni
... (read more)
5ESRogs
In my mind, this is still making the mistake of not distinguishing the true domain of the agent's utility function from ours. Whether the simulation continues to be instantiated in some computer in our world is a fact about our world, not about the simulated world. AlphaGo doesn't care about being unplugged in the middle of a game (unless that dynamic was part of its training data). It cares about the platonic game of go, not about the instantiated game it's currently playing. We need to worry about leaky abstractions, as per my original comment. So we can't always assume the agent's domain is what we'd ideally want it to be. But I'm trying to highlight that it's possible (and I would tentatively go further and say probable) for agents not to care about the real world. To me, assuming care about the real world (including wanting not to be unplugged) seems like a form of anthropomorphism. For any given agent-y system I think we need to analyze whether it in particular would come to care about real world events. I don't think we can assume in general one way or the other.
6Rob Bensinger
What if the programmers intervene mid-game to give the other side an advantage? Does a Go AGI, as you're thinking of it, care about that? I'm not following why a Go AGI (with the ability to think about the physical world, but a utility function that only cares about states of the simulation) wouldn't want to seize more hardware, so that it can think better and thereby win more often in the simulation; or gain control of its hardware and directly edit the simulation so that it wins as many games as possible as quickly as possible. Why would having a utility function that only assigns utility based on X make you indifferent to non-X things that causally affect X? If I only terminally cared about things that happened a year from now, I would still try to shape the intervening time because doing so will change what happens a year from now. (This is maybe less clear in the case of shutdown, because it's not clear how an agent should think about shutdown if its utility is defined states of its simulation. So I'll set that particular case aside.)
2David Johnston
A Go AI that learns to play go via reinforcement learning might not "have a utility function that only cares about winning Go". Using standard utility theory, you could observe its actions and try to rationalise them as if they were maximising some utility function, and the utility function you come up with probably wouldn't be "win every game of Go you start playing" (what you actually come up with will depend, presumably, on algorithmic and training regime details). The reason why the utility function is slippery is that it's fundamentally an adaptation executor, not a utility maxmiser.
2David Johnston
Not necessarily. Train something multimodally on digital games of Go and on, say, predicting the effects of modifications to its own code on its success at Go. It could be a) good at go and b) have some real understanding of "real world actions" that make it better at Go, and still not actually take any real world actions to make it better at Go, even if it had the opportunity. You could modify the training to make it likely to do so - perhaps by asking it to either make a move or to produce descendants that make better choices - but if you don't do this then it seems entirely plausible, and even perhaps likely, that it develops an understanding of self-modification and of go playing without ever self-modifying in order to play go better. Its goal, so to speak, is "play go with the restriction of using only legal game moves". Edit - forget the real world, here's an experiment: Train a board game playing AI with two modes of operation: game state x move -> outcome and game state -> best move. Subtle difference: in the first mode of operation, the move has a "cheat button" that, when pressed, always results in a win. In the second, it can output cheat button presses, but it has no effect on winning or losing. Question is: does it learn to press the cheat button? I'm really not sure. Could you prevent it from learning to press the cheat button if training feedback is never allowed to depend on whether or not this button was pressed? That seems likely.
7James Payor
Can you visualize an agent that is not "open-ended" in the relevant ways, but is capable of, say, building nanotech and melting all the GPUs? In my picture most of the extra sauce you'd need on top of GPT-3 looks very agenty. It seems tricky to name "virtual worlds" in which AIs manipulate just "virtual resources" and still manage to do something like melting the GPUs.
8James Payor
I should say that I do see this as a reasonable path forward! But we don't seem to be coordinating to do this, and AI researchers seem to love doing work on open-ended agents, which sucks. Hm, regardless it doesn't really move the needle, so long as people are publishing all of their work. Developing overpowered pattern recognizers is similar to increasing our level of hardware overhang. People will end up using them as components of systems that aren't safe.
4David Johnston
I strongly disagree. Gain of function research happens, but it's rare because people know it's not safe. To put it mildly, I think reducing the number of dangerous experiments substantially improves the odds of no disaster happening over any given time frame
5ESRogs
FWIW, I'm not sold on the idea of taking a single pivotal act. But, engaging with what I think is the real substance of the question — can we do complex, real-world, superhuman things with non-agent-y systems? Yes, I think we can! Just as current language models can be prompt-programmed into solving arithmetic word problems, I think a future system could be led to generate a GPU-melting plan, without it needing to be a utility-maximizing agent. For a very hand-wavy sketch of how that might go, consider asking GPT-N to generate 1000s of candidate high-level plans, then rate them by feasibility, then break each plan into steps and re-evaluate, etc. Or, alternatively, imagine the cognitive steps you might take if you were trying to come up with a GPU-melting plan (or alternatively a pivotal act plan in general). Do any of those steps really require that you have a utility function or that you're a goal-directed agent? It seems to me that we need some form of search, and discrimination and optimization. But not necessarily anymore than GPT-3 already has. (It would just need to be better at the search. And we'd need to make many many passes through the network to complete all the cognitive steps.) On your view, what am I missing here? * Is GPT-3 already more of an agent than I realize? (If so, is it dangerous?) * Will GPT-N by default be more of an agent than GPT-3? * Are our own thought processes making use of goal-directedness more than I realize? * Will prompt-programming passive systems hit a wall somewhere? * If so, what are some of the simplest cognitive tasks that we can do that you think such systems wouldn't be able to do? * (See also my similar question here.)
4David Johnston
FWIW, I'd call this "weakly agentic" in the sense that you're searching through some options, but the number of options you're looking through is fairly small. It's plausible that this is enough to get good results and also avoid disasters, but it's actually not obvious to me. The basic reason: if the top 1000 plans are good enough to get superior performance, they might also be "good enough" to be dangerous. While it feels like there's some separation between "useful and safe" and "dangerous" plans and this scheme might yield plans all of the former type, I don't presently see a stronger reason to believe that this is true.
6ESRogs
Separately from whether the plans themselves are safe or dangerous, I think the key question is whether the process that generated the plans is trying to deceive you (so it can break out into the real world or whatever). If it's not trying to deceive you, then it seems like you can just build in various safeguards (like asking, "is this plan safe?", as well as more sophisticated checks), and be okay.
2TekhneMakre
>then rate them by feasibility, I mean, literal GPT is just going to have poor feasibility ratings for novel engineering concepts. >Do any of those steps really require that you have a utility function or that you're a goal-directed agent? Yes, obviously. You have to make many scientific and engineering discoveries, which involves goal-directed investigation.  > Are our own thought processes making use of goal-directedness more than I realize? Yes, you know which ideas make sense by generalizing from ideas more closely tied in with the actions you take directed towards living.  
4David Johnston
What do you think of a claim like "most of the intelligence comes from the steps where you do most of the optimization"? A corollary of this is that we particularly want to make sure optimization intensive steps of AI creation are safe WRT not producing intelligent programs devoted to killing us. Example: most of the "intelligence" of language models comes from the supervised learning step. However, it's in-principle plausible that we could design e.g. some really capable general purpose reinforcement learner where the intelligence comes from the reinforcement, and the latter could (but wouldn't necessarily) internalise "agenty" behaviour. I have a vague impression that this is already something other people are thinking about, though maybe I read too much into some tangential remarks in this direction. E.g. I figured the concern about mesa-optimizers was partly motivated by the idea that we can't always tell when an optimization intensive step is taking place. I can easily imagine people blundering into performing unsafe optimization-intensive AI creation processes. Gain of function pathogen research would seem to be a relevant case study here, except we currently have less idea about what kind of optimization makes deadly AIs vs what kind of optimization makes deadly pathogens. One of the worries (again, maybe I'm reading too far into comments that don't say this explicitly) is that the likelihood of such a blunder approaches 1 over long enough times, and the "pivotal act" framing is supposed to be about doing something that could change this (??) That said, it seems that there's a lot that could be done to make it less likely in short time frames.
3ESRogs
This seems probably right to me. I agree that reinforcement learners seem more likely to be agent-y (and therefore scarier) than self-supervised learners.

I agree with pretty much everything here, and I would add into the mix two more claims that I think are especially cruxy and therefore should maybe be called out explicitly to facilitate better discussion:

Claim A: “There’s no defense against an out-of-control omnicidal AGI, not even with the help of an equally-capable (or more-capable) aligned AGI, except via aggressive outside-the-Overton-window acts like preventing the omnicidal AGI from being created in the first place.”

I think this claim is true, on account of gray goo and lots of other things, and I suspect Eliezer does too, and I’m pretty sure other people disagree with this claim.

If someone disagrees with this claim (i.e., if they think that if DeepMind can make an aligned and Overton-window-abiding “helper” AGI, then we don’t have to worry about Meta making a similarly-capable out-of-control omnicidal misaligned AGI the following year, because DeepMind’s AGI will figure out how to protect us), and also believes in extremely slow takeoff, I can see how such a person might be substantially less pessimistic about AGI doom than I am.

Claim B: “Shortly after (i.e., years not decades after) we have dangerous AGI, we will have dang... (read more)

 I think this claim is true, on account of gray goo and lots of other things, and I suspect Eliezer does too, and I’m pretty sure other people disagree with this claim.

If you have robust alignment, or AIs that are rapidly bootstrapping their level of alignment fast enough to outpace the danger of increased capabilities, aligned AGI could get through its intelligence explosion to get radically superior technology and capabilities. It could also get a hard start on superexponential replication in space, so that no follower could ever catch up, and enough tech and military hardware to neutralize any attacks on it (and block attacks on humans via nukes, bioweapons, robots, nanotech, etc). That wouldn't work if there are thing like vacuum collapse available to attackers, but we don't have much reason to expect that from current science and the leading aligned AGI would find out first.

That could be done without any violation of the territory of other sovereign states. The legality of grabbing space resources is questionable in light of the Outer Space Treaty, but commercial exploitation of asteroids is in the Overton window. The superhuman AGI would also be in a good position to per... (read more)

6MichaelStJules
A bit pedantic, but isn't superexponential replication too fast? Won't it hit physical limits eventually, e.g. expanding at the speed of light in each direction, so at most a cubic function of time? Also, never allowing followers to catch up means abandoning at least some or almost all of the space you passed through. Plausibly you could take most of the accessible and useful resources with you, which would also make it harder for pursuers to ever catch up, since they will plausibly need to extract resources every now and then to fuel further travel. On the other hand, it seems unlikely to me that we could extract or destroy resources quickly enough to not leave any behind for pursuers, if they're at most months behind.
7CarlShulman
Naturally it doesn't go on forever, but any situation where you're developing technologies that move you to successively faster exponential trajectories is superexponential overall for some range. E.g. if you have robot factories that can reproduce exponentially until they've filled much of the Earth or solar system, and they are also developing faster reproducing factories,  the overall process is superexponential. So is the history of human economic growth, and the improvement from an AI intelligence explosion. By the time you're at ~cubic expansion being ahead on the early superexponential phase the followers have missed their chance.
3MichaelStJules
I agree that they probably would have missed their chance to catch up with the frontier of your expansion. Maybe an electromagnetic radiation-based assault could reach you if targeted (the speed of light is constant relative to you in a vacuum, even if you're traveling in the same direction), although unlikely to get much of the frontier of your expansion, and there are plausibly effective defenses, too. Do you also mean they wouldn't be able to take most what you've passed through, though? Or it wouldn't matter? If so, how would this be guaranteed (without any violation of the territory of sovereign states on Earth)? Exhaustive extraction in space? An advantage in armed space conflicts?

I agree with these two points. I think an aligned AGI actually able to save the world would probably take initial actions that look pretty similar to those an unaligned AGI would take. Lots of sizing power, building nanotech, colonizing out into space, self-replication, etc. 

4Yitz
So how would we know the difference (for the first few years at least)?

If it kills you, then it probably wasn’t aligned. 

1[anonymous]
Maybe it did that to save your neural weights.  Define 'kill'. 
4Quintin Pope
I did say “probably”!
7lc
I disagree with this claim inasmuch as I expect a year headstart by an aligned AI is absolutely enough to prevent Meta from killing me and my family. 

Depends on what DeepMind does with the AI, right?

Maybe DeepMind uses their AI in very narrow, safe, low-impact ways to beat ML benchmarks, or read lots of cancer biology papers and propose new ideas about cancer treatment.

Or alternatively, maybe DeepMind asks their AI to undergo recursive self-improvement and build nano-replicators in space, etc., like in Carl Shulman’s reply.

I wouldn’t have thought that the latter is really in the Overton window. But what do I know.

You could also say “DeepMind will just ask their AI what they should do next”. If they do that, then maybe the AI (if they’re doing really great on safety such that the AI answers honestly and helpfully) will reply: “Hey, here’s what you should do, you should let me undergo recursive-self-improvement, and then I’ll be able to think of all kinds of crazy ways to destroy the world, and then I can think about how to defend against all those things”. But if DeepMind is being methodical & careful enough that their AI hasn’t destroyed the world already by this point, I’m inclined to think that they’re also being methodical & careful enough that when the AI proposes to do that, DeepMind will say, “Umm, no, that’s total... (read more)

3lc
If DeepMind was committed enough to successfully build an aligned AI (which, as extensively elaborated upon in the post, is a supernaturally difficult proposition), I would assume they understand why running it is necessary. There's no reason to take all of the outside-the-overton-window measures indicated in the above post unless you have functioning survival instincts and have thought through the problem sufficiently to hit the green button.
2MichaelStJules
If you can build one aligned superintelligence, then plausibly you can 1. explain to other AGI developers how to make theirs safe or even just give them a safe design (maybe homomorphically encrypted to prevent modification, but they might not trust that), and 2. have aligned AGI monitoring the internet and computing resources, and alert authorities of abnomalies that might signal new AGI developments. Require that AGI developments provide proof that they were designed according to one of a set of approved designs, or pass some tests determined by your aligned superintelligence. Then aligned AGI can proliferate first and unaligned AGI will plausibly face severe barriers. Plausibly 1 is enough, since there's enough individual incentive to build something safe or copy other people's designs and save work. 2 depends on cooperation with authorities and I'd guess cloud computing service providers or policy makers.

explain to other AGI developers how to make theirs safe or even just give them a safe design (maybe homomorphically encrypted to prevent modification, but they might not trust that)

What if the next would-be AGI developer rejects your “explanation”, and has their own great ideas for how to make an even better next-gen AGI that they claim will work better, and so they discard your “gift” and proceed with their own research effort?

I can think of at least two leaders of would-be AGI development efforts (namely Yann LeCun of Meta and Jeff Hawkins of Numenta) who believe (what I consider to be) spectacularly stupid things about AGI x-risk, and have believed those things consistently for decades, despite extensive exposure to good counter-arguments.

Or what if the next would-be AGI developer agrees with you and accepts your “gift”, and so does the one after that, and the one after that, but not the twelfth one?

have aligned AGI monitoring the internet and computing resources, and alert authorities of [anomalies] that might signal new AGI developments. Require that AGI developments provide proof that they were designed according to one of a set of approved designs, or pass some tests determi

... (read more)
4MichaelStJules
When you ask "what if", are you implying these things are basically inevitable? And inevitable no matter how much more compute aligned AGIs have before unaligned AGIs are developed and deployed? How much of a disadvantage against aligned AGIs does an unaligned AGI need before doom isn't overwhelmingly likely? What's the goal post here for survival probability? You can have AGIs monitoring for pathogens, nanotechnology, other weapons, and building defenses against them, and this could be done locally and legally. They can monitor transactions and access to websites through which dangerous physical systems (including possibly factories, labs, etc.) could be taken over or built. Does every country need to be competent and compliant to protect just one country from doom? The Overton window could also shift dramatically if omnicidal weapons are detected. I agree that plausibly not every country with significant compute will comply, and hacking everyone is outside the public Overton window. I wouldn't put hacking everyone past the NSA, but also wouldn't count on them either.
4Steven Byrnes
Let’s see, I think “What if the next would-be AGI developer rejects your “explanation” / “gift”” has a probability that asymptotes to 100% as the number of would-be AGI developers increases. (Hence “Claim B” above becomes relevant.) I think “What if the authorities in most countries do care, but not the authorities in every single country?” seems to have high probability in today’s world, although of course I endorse efforts to lower the probability. I think “What if the only way to “monitor the internet and computing resources” is to hack into every data center and compute cluster on the planet? (Including those in secret military labs.)” seems very likely to me, conditional on “Claim B” above. Hmm. Offense-defense balance in bio-warfare is not obvious to me. Preventing a virus from being created would seem to require 100% compliance by capable labs, but I’m not sure how many “capable labs” there are, or how geographically distributed and rule-following. Once the virus starts spreading, aligned AGIs could help with vaccines, but apparently a working COVID-19 vaccine was created in 1 day, and that didn’t help much, for various societal coordination & governance reasons. So then you can say “Maybe aligned AGI will solve all societal coordination and governance problems”. And maybe it will! Or, maybe some of those coordination & governance problems come from blame-avoidance and conflicts-of-interest and status-signaling and principle-agent problems and other things that are not obviously solvable by easy access to raw intelligence. I don’t know. Offense-defense balance in nuclear warfare is likewise not obvious to me. I presume that an unaligned AGI could find a way to manipulate nuclear early warning systems (trick them, hack into them, bribe or threaten their operators, etc.) to trigger all-out nuclear war, after hacking into a data center in New Zealand that wouldn’t be damaged. An aligned AGI playing defense would need to protect against these vulnerabilities.
4MichaelStJules
Some more skepticism about infectious diseases and nukes killing us all here: https://www.lesswrong.com/posts/MLKmxZgtLYRH73um3/we-will-be-around-in-30-years?commentId=DJygArj3sj8cmhmme Also my more general skeptical take against non-nano attacks here: https://www.lesswrong.com/posts/MLKmxZgtLYRH73um3/we-will-be-around-in-30-years?commentId=TH4hGeXS4RLkkuNy5 With nanotech, I think there will be tradeoffs between targeting effectiveness and requiring (EM) signals from computers that can be effectively interferred with through things within or closer to the Overton window. Maybe a crux is how good autonomous nanotech with no remote control would be at targeting humans or spreading so much that it just gets into almost all buildings or food or water because it's basically going everywhere.
4Steven Byrnes
Thanks! I wasn’t assuming the infectious diseases and nukes by themselves would kill us all. They don’t have to, because the AGI can do other things in conjunction, like take command of military drones and mow down the survivors (or bomb the PPE factories), or cause extended large-scale blackouts, which would incidentally indirectly prevent PPE production and distribution, along with preventing pretty much every other aspect of an organized anti-pandemic response. See Section 1.6 here. So that brings us to the topic of offense-defense balance for illicitly taking control of military drones. And I would feel concerned about substantial delays before the military trusts a supposedly-aligned AGI so much that they give it root access to all its computer systems (which in turn seems necessary if the aligned AGI is going to be able to patch all the security holes, defend against spear-phishing attacks, etc.) Of course there’s the usual caveat that maybe DeepMind will give their corrigible aligned AGI permission to hack into military systems (for their own good!), and then maybe we wouldn’t have to worry. But the whole point of this discussion is that I’m skeptical that DeepMind would actually give their AGI permission to do something like that. And likewise we would need to talk about offense-defense balance for the power grid. And I would have the same concern about people being unwilling to give a supposedly-aligned AGI root access to all the power grid computers. And I would also be concerned about other power grid vulnerabilities like nuclear EMPs, drone attacks on key infrastructure, etc. And likewise, what’s the offense-defense balance for mass targeted disinformation campaigns? Well, if DeepMind gives its AGI permission to engage in a mass targeted counter-disinformation campaign, maybe we’d be OK on that front. But that’s a big “if”! …And probably dozens of other things like that. Seems like a good question, and maybe difficult to resolve. Or maybe I would
3MichaelStJules
I think there would be too many survivors and enough manned defense capability for existing drones to directly kill the rest of us with high probability. Blocking PPE production and organized pandemic responses still won't stop people from self-isolating, doing no contact food deliveries, etc., although things would be tough, and deliveries and food production would be good targets for drone strikes. It could be bad if lethal pathogens become widespread and practically unremovable in our food/water, or if food production is otherwise consistently attacked, but the militaries would probably step in to protect the food/water supplies. I think, overall, there are too few ways to reliably and kill double or even single digit percentages of the human population with high probability and that can be combined to get basically everyone with high probability. I'm not saying there aren't any, but I'm skeptical that there are enough. There are diminishing returns on doing the same ones (like pandemics) more, because of resistance, and enough people being personally very careful or otherwise difficult targets.

Found this to be an interesting list of challenges, but I disagree with a few points. (Not trying to be comprehensive here, just a few thoughts after the first read-through.)

  • Several of the points here are premised on needing to do a pivotal act that is way out of distribution from anything the agent has been trained on. But it's much safer to deploy AI iteratively; increasing the stakes, time horizons, and autonomy a little bit each time. With this iterative approach to deployment, you only need to generalize a little bit out of distribution. Further, you can use Agent N to help you closely supervise Agent N+1 before giving it any power.
  • One claim is that Capabilities generalize further than alignment once capabilities start to generalize far. The argument is that an agent's world model and tactics will be automatically fixed by reasoning and data, but its inner objective won't be changed by these things. I agree with the preceding sentence, but I would draw a different (and more optimistic) conclusion from it. That it might be possible to establish an agent's inner objective when training on easy problems, when the agent isn't very capable, such that this objective remains stable a
... (read more)
[-]VaniverΩ11273

But it's much safer to deploy AI iteratively; increasing the stakes, time horizons, and autonomy a little bit each time. With this iterative approach to deployment, you only need to generalize a little bit out of distribution. Further, you can use Agent N to help you closely supervise Agent N+1 before giving it any power.

My model of Eliezer claims that there are some capabilities that are 'smooth', like "how large a times table you've memorized", and some are 'lumpy', like "whether or not you see the axioms behind arithmetic." While it seems plausible that we can iteratively increase smooth capabilities, it seems much less plausible for lumpy capabilities. 

A specific example: if you have a neural network with enough capacity to 1) memorize specific multiplication Q+As and 2) implement a multiplication calculator, my guess is that during training you'll see a discontinuity in how many pairs of numbers it can successfully multiply.[1] It is not obvious to me whether or not there are relevant capabilities like this that we'll "find with neural nets" instead of "explicitly programming in"; probably we will just build AlphaZero so that it uses MCTS instead of finding MCTS with grad... (read more)

5John Schulman
Re: smooth vs bumpy capabilities, I agree that capabilities sometimes emerge abruptly and unexpectedly. Still, iterative deployment with gradually increasing stakes is much safer than deploying a model to do something totally unprecedented and high-stakes. There are multiple ways to make deployment more conservative and gradual. (E.g., incrementally increase the amount of work the AI is allowed to do without close supervision, incrementally increase the amount of KL-divergence between the new policy and a known-to-be-safe policy.) Re: ontological collapse, there are definitely some tricky issues here, but the problem might not be so bad with the current paradigm, where you start with a pretrained model (which doesn't really have goals and isn't good at long-horizon control), and fine-tune it (which makes it better at goal-directed behavior). In this case, most of the concepts are learned during the pretraining phase, not the fine-tuning phase where it learns goal-directed behavior.
6Vaniver
I agree with the "X is safer than Y" claim; I am uncertain whether it's practically available to us, and much more worried in worlds where it isn't available. For this specific proposal, when I reframe it as "give the system a KL-divergence budget to spend on each change to its policy" I worry that it works against a stochastic attacker but not an optimizing attacker; it may be the case that every known-to-be-safe policy has some unsafe policy within a reasonable KL-divergence of it, because the danger can be localized in changes to some small part of the overall policy-space. Yeah, I agree that this seems pretty good. I do naively guess that when you do the fine-tuning, it's the concepts that are most related to the goals who change the most (as they have the most gradient pressure on them); it'd be nice to know how much this is the case, vs. most of the relevant concepts being durable parts of the environment that were already very important for goal-free prediction.

Several of the points here are premised on needing to do a pivotal act that is way out of distribution from anything the agent has been trained on. But it's much safer to deploy AI iteratively; increasing the stakes, time horizons, and autonomy a little bit each time.

To do what, exactly, in this nice iterated fashion, before Facebook AI Research destroys the world six months later?  What is the weak pivotal act that you can perform so safely?

Human raters make systematic errors - regular, compactly describable, predictable errors.... This is indeed one of the big problems of outer alignment, but there's lots of ongoing research and promising ideas for fixing it. Namely, using models to help amplify and improve the human feedback signal. Because P!=NP it's easier to verify proofs than to write them. 

When the rater is flawed, cranking up the power to NP levels blows up the P part of the system.

To do what, exactly, in this nice iterated fashion, before Facebook AI Research destroys the world six months later?  What is the weak pivotal act that you can perform so safely?

Do alignment & safety research, set up regulatory bodies and monitoring systems.

When the rater is flawed, cranking up the power to NP levels blows up the P part of the system.

Not sure exactly what this means. I'm claiming that you can make raters less flawed, for example, by decomposing the rating task, and providing model-generated critiques that help with their rating. Also, as models get more sample efficient, you can rely more on highly skilled and vetted raters.
 

[-]VaniverΩ153225

Not sure exactly what this means.

My read was that for systems where you have rock-solid checking steps, you can throw arbitrary amounts of compute at searching for things that check out and trust them, but if there's any crack in the checking steps, then things that 'check out' aren't trustable, because the proposer can have searched an unimaginably large space (from the rater's perspective) to find them. [And from the proposer's perspective, the checking steps are the real spec, not whatever's in your head.]

In general, I think we can get a minor edge from "checking AI work" instead of "generating our own work" and that doesn't seem like enough to tackle 'cognitive megaprojects' (like 'cure cancer' or 'develop a pathway from our current society to one that can reliably handle x-risk' or so on). Like, I'm optimistic about "current human scientists use software assistance to attempt to cure cancer" and "an artificial scientist attempts to cure cancer" and pretty pessimistic about "current human scientists attempt to check the work of an artificial scientist that is attempting to cure cancer." It reminds me of translators who complained pretty bitterly about being given machine-transl... (read more)

7Charbel-Raphaël
If Facebook AI research is such a threat, wouldn't it be possible to talk to Yann LeCun?

I did, briefly.  I ask that you not do so yourself, or anybody else outside one of the major existing organizations, because I expect that will make things worse as you annoy him and fail to phrase your arguments in any way he'd find helpful.

Other MIRI staff have also chatted with Yann. One co-worker told me that he was impressed with Yann's clarity of thought on related topics (e.g., he has some sensible, detailed, reductionist models of AI), so I'm surprised things haven't gone better.

Non-MIRI folks have talked to Yann too; e.g., Debate on Instrumental Convergence between LeCun, Russell, Bengio, Zador, and More.

9TekhneMakre
What happened?

Nothing much.

There was also a debate between Yann and Stuart Russel on facebook, which got discussed here:

https://www.lesswrong.com/posts/WxW6Gc6f2z3mzmqKs/debate-on-instrumental-convergence-between-lecun-russell

For a more comprehensive writeup of some stuff related to the "annoy him and fail to phrase your arguments helpfully", see Idea Innoculation and Inferential Distance

My view is that if Yann continues to be interested in arguing about the issue then there's something to work with, even if he's skeptical, and the real worry is if he's stopped talking to anyone about it (I have no idea personally what his state of mind is right now)

1jrincayc
Produce the Textbook From The Future that tells us how to do AGI safely. That said, getting an AGI to generate a correct Foom safety textbook or AGI Textbook from the future would be incredibly difficult, it would be very possible for an AGI to slip in a subtle hard-to-detect inaccuracy that would make it worthless, verifying that it is correct would be very difficult, and getting all humans on earth to follow it would be very difficult.
[-]Wei DaiΩ245714

I think until recently, I've been consistently more pessimistic than Eliezer about AI existential safety. Here's a 2004 SL4 post for example where I tried to argue against MIRI (SIAI at the time) trying to build a safe AI (and again in 2011). I've made my own list of sources of AI risk that's somewhat similar to this list. But it seems to me that there are still various "outs" from certain doom, such that my probability of a good outcome is closer to 20% (maybe a range of 10-30% depending on my mood) than 1%.

  1. Human thought partially exposes only a partially scrutable outer surface layer. Words only trace our real thoughts. Words are not an AGI-complete data representation in its native style. The underparts of human thought are not exposed for direct imitation learning and can't be put in any dataset. This makes it hard and probably impossible to train a powerful system entirely on imitation of human words or other human-legible contents, which are only impoverished subsystems of human thoughts; unless that system is powerful enough to contain inner intelligences figuring out the humans, and at that point it is no longer really working as imitative human thought.

One of the... (read more)

[This is a nitpick of the form "one of your side-rants went a bit too far IMO;" feel free to ignore]

The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so. Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try. ... The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years into even normies starting to notice the death game, it is still Eliezer Yudkowsky writing up this list, says that humanity still has only one gamepiece that can do that.

The third option this seems to miss is that there are people who could have written this document, but they also thought they had better things to do than write it. I'm thinking of people like Paul Christiano, Nate Soares, John W... (read more)

I'm thinking of people like Paul Christiano, Nate Soares, John Wentworth, Ajeya Cotra...  [...] I do agree with you that they seem to on average be way way too optimistic, but I don't think it's because they are ignorant of the considerations and arguments you've made here.

I don't think Nate is that much more optimistic than Eliezer, but I believe Eliezer thinks Nate couldn't have generated enough of the list in the OP, or couldn't have generated enough of it independently ("using the null string as input").

1gettinglesswrong
>too would be cautiously optimistic if I thought we had 30 years left   This is a bit of an aside but can I ask what the general opinion is on how many years we had left? Was your comment stating that it's optimistic to think we have 30 years left before AGI, or optimistic about the remainder of the sentence?
2[comment deleted]
-14Eliezer Yudkowsky
[-]TurnTroutΩ225023

Reading this post made me more optimistic about alignment and AI. My suspension of disbelief snapped; I realized how vague and bad a lot of these "classic" alignment arguments are, and how many of them are secretly vague analogies and intuitions about evolution.

While I agree with a few points on this list, I think this list is fundamentally misguided. The list is written in a language which assigns short encodings to confused and incorrect ideas. I think a person who tries to deeply internalize this post's worldview will end up more confused about alignment and AI, and urge new researchers to not spend too much time trying to internalize this post's ideas. (Definitely consider whether I am right in my claims here. Think for yourself. If you don't know how to think for yourself, I wrote about exactly how to do it! But my guess is that deeply engaging with this post is, at best, a waste of time.[1])

I think this piece is not "overconfident", because "overconfident" suggests that Lethalities is simply assigning extreme credences to reasonable questions (like "is deceptive alignment the default?"). Rather, I think both its predictions and questions are not reasonable because they are no... (read more)

4lc
I mean it's worth considering that his P(DOOM) was substantially lower then. He's definitely updated on existing evidence, just in the opposite direction that you have.

I would summarize a dimension of the difficulty like this. There are the conditions that give rise to intellectual scenes, intellectual scenes being necessary for novel work in ambiguous domains. There are the conditions that give rise to the sort of orgs that output actions consistent with something like Six Dimensions of Operational Adequacy. The intersection of these two things is incredibly rare but not unheard of. The Manhattan Project was a Scene that had security mindset. This is why I am not that hopeful. Humans are not the ones building the AGI, egregores are, and spending egregore sums of money. It is very difficult for individuals to support a scene of such magnitude, even if they wanted to. Ultra high net worth individuals seem much poorer relative to the wealth of society than in the past, where scenes and universities (a scene generator) could be funded by individuals or families. I'd guess this is partially because the opportunity cost for smart people is much higher now, and you need to match that (cue title card: Baumol's cost disease kills everyone). In practice I expect some will give objections along various seemingly practical lines, but my experience so far is... (read more)

6Ben Pace
Thanks, this story is pretty helpful (to my understanding).
[-]RaemonΩ16379

Note: I think there's a bunch of additional reasons for doom, surrounding "civilizational adequacy / organizational competence / societal dynamics". Eliezer briefly alluded to these, but AFAICT he's mostly focused on lethality that comes "early", and then didn't address them much. My model of Andrew Critch has a bunch of concerns about doom that show up later, because there's a bunch of additional challenges you have to solve if AI doesn't dramatically win/lose early on (i.e. multi/multi dynamics and how they spiral out of control)

I know a bunch of people whose hope funnels through "We'll be able to carefully iterate on slightly-smarter-than-human-intelligences, build schemes to play them against each other, leverage them to make some progress on alignment that we can use to build slightly-more-advanced-safer-systems". (Let's call this the "Careful Bootstrap plan")

I do actually feel nonzero optimism about that plan, but when I talk to people who are optimistic about that I feel a missing mood about the kind of difficulty that is involved here.

I'll attempt to write up some concrete things here later, but wanted to note this for now.

0HiroSakuraba
I agree with this line of thought regarding iterative developments of proto-AGI via careful bootstrapping.  Humans will be inadequate for monitoring progress of skills.  Hopefully, we'll have a slew of diagnostic of narrow minded neural networks whose sole purpose is to tease out relevant details of the proto-super human intellect.  What I can't wrap my head around is whether super (or sub) human level intelligence requires consciousness.  If consciousness is required, then is the world worse or better for it?  Is an agent with the rich experience of fears, hopes, joys more or less likely to be built?  Do reward functions reliably grow into feelings, which lead to emotional experiences?  If they do, then perhaps an evolving intelligence wouldn't always be as alien as we currently imagine it.

What concerns me the most is the lack of any coherent effort anywhere, towards solving the biggest problem: identifying a goal (value system, utility function, decision theory, decision architecture...) suitable for an autonomous superhuman AI. 

In these discussions, Coherent Extrapolated Volition (CEV) is the usual concrete formulation of what such a goal might be. But I've now learned that MIRI's central strategy is not to finish figuring out the theory and practice of CEV - that's considered too hard (see item 24 in this post). Instead, the hope is to use safe AGI to freeze all unsafe AGI development everywhere, for long enough that humanity can properly figure out what to do. Presumably this freeze (the "pivotal act") would be carried out by whichever government or corporation or university crossed the AGI threshold first; ideally there might even become a consensus among many of the contenders that this is the right thing to do. 

I think it's very appropriate that some thought along these lines be carried out. If AGI is a threat to the human race, and it arrives before we know how to safely set it free, then we will need ways to try to neutralize that dangerous potenti... (read more)

There's shard theory, which aims to describe the process by which values form in humans. The eventual aim is to understand value formation well enough that we can do it in an AI system. I also think figuring out human values, value reflection and moral philosophy might actually be a lot easier than we assume. E.g., the continuous perspective on agency / values is pretty compelling to me and changes things a lot, IMO.

-28Trevor Cappallo
[-]aogara3510

Thank you, this was very helpful. As a bright-eyed youngster, it's hard to make sense of the bitterness and pessimism I often see in the field. I've read the old debates, but I didn't participate in them, and that probably makes them easier to dismiss. Object level arguments like these help me understand your point of view. 

Alright, now that I've read this post, I'll try to respond to what I think you got wrong, and importantly illustrate some general principles.

To respond to this first:


3.  We need to get alignment right on the 'first critical try' at operating at a 'dangerous' level of intelligence, where unaligned operation at a dangerous level of intelligence kills everybody on Earth and then we don't get to try again.  This includes, for example: (a) something smart enough to build a nanosystem which has been explicitly authorized to build a nanosystem; or (b) something smart enough to build a nanosystem and also smart enough to gain unauthorized access to the Internet and pay a human to put together the ingredients for a nanosystem; or (c) something smart enough to get unauthorized access to the Internet and build something smarter than itself on the number of machines it can hack; or (d) something smart enough to treat humans as manipulable machinery and which has any authorized or unauthorized two-way causal channel with humans; or (e) something smart enough to improve itself enough to do (b) or (d); etcetera.  We can gather all sorts of information beforehand from less

... (read more)
4MondSemmel
I wish you'd made this a top-level post; the ultra-long quote excerpts in a comment made it ~unreadable to me. And you don't benefit from stuff like bigger font size or automatic table of contents. And scrolling to the correct position on this long comment thread also works poorly, etc. Anyway, I read your rebuttals on the first two points and did not find them persuasive (thus resulting in a strong disagree-vote on the whole comment). So now I'm curious about the upvotes without accompanying discussion. Did others find this rebuttal more persuasive?
5Noosphere89
I made this a top level post, and fixed the formatting and quoting: https://www.lesswrong.com/posts/wkFQ8kDsZL5Ytf73n/my-disagreements-with-agi-ruin-a-list-of-lethalities
4quetzal_rainbow
I think that you should have adressed points by referring to number of point, quoting only parts that are easier to quote that refer to, it would have reduced the size of the comment. I am going to adress only one object-level point: No, obviously, we can't control what AI learns and value using synthetic data in practice, because we need AI to learn things that we don't know. If you feed AI all physics and chemistry data with expectation to get nanotech, you are doing this because you expect that AI learns facts and principles you don't know about and, therefore, can't control. You don't know about these facts and principles and can't control them because otherwise you would be able to design nanotech yourself. Of course, I'm saying "can't" meaning "practically can't", not "in principle". But to do this you need to do basically "GOFAI in trenchcoat of SGD" and it doesn't look competitive with any other method of achieving AGI, unless you manage to make yourself AGI Czar.
2Noosphere89
Okay, the reason for this happening: This is basically a combo of very high sample efficiency, defining a good-enough ground truth reward signal, very good online learning, very good credit assignment and handling uncertainity and simulatability well. But for our purposes, if we decided that we didn't in fact want to learn nanotech, we could just remove the data from it's experience, in a way we couldn't do with humans, which is quite a big win for misuse concerns. But my point here was that you can get large sets of data on values early on in training, and we can both iteratively refine on values by testing the model's generalization of the value data to new situations, as well as rely on the fact that alignment generalizes further than capabilities does. I think my crux here is this: I think this is just not correct, and while we should start making large datasets now, I think a crux here is that I believe that far less data is necessary for models to generalize alignment, and that we aren't trying to hand-code everything, and instead rely on the fact that models will generalize better and better on human values as they get more capable, due to alignment generalizing further than capabilities and there likely being a simple core to alignment, so I don't think we need a GOFAI in trenchcoat of SGD. We've discussed this before https://x.com/quetzal_rainbow/status/1834268698565059031, but while I agree with TurnTrout that RL doesn't maximize reward by definition, and the reward maximization hypothesis isn't an automatic consequence of RL training, I do think that something like reward maximization might well occur in practice, and more generally I think that the post ignores the possibility that future RL could generalize better towards maximizing the reward function.
2quetzal_rainbow
(It seems like "here" link got mixed with the word "here"?)
2Noosphere89
Alright, I fixed the link, though I don't know why you can't transform non-Lesswrong links into links that have a shorter title link.
[-]habrykaΩ113277

Mod note: I activated two-axis voting on this post, since it seemed like it would make the conversation go better.

I agree.

7lc
You should just activate it sitewide already :)

New users are pretty confused by it when I've done some user-testing with it, so I think it needs some polish and better UI before we can launch it sitewide, but I am pretty excited about doing so after that.

6Harry Nyquist
As a very new user, I'm not sure if it's still helpful to add a data point if user testing's already been done, but it seems at worst mostly harmless. I saw the mod note before I started using the votes on this post. My first idea was to Google the feature, but that returned nothing relevant (while writing this post, I did find results immediately through site search). I was confused for a short while trying to place the axes & imagine where I'd vote in opposite directions. But after a little bit of practice looking at comments, it started making sense. I've read a couple comments on this article that I agree with, where it seems very meaningful for me to downvote them (I interpret the downvote's meaning when both axes are on as low quality, low importance, should be read less often). I relatively easily find posts I want to upvote on karma. But for posts that I upvote, I'm typically much less confident about voting on agreement than for other posts (as a new user, it's harder to assess the specific points made in high quality posts). Posts where I'm not confident voting on agreement correlate with posts I'm not confident I can reply to without lowering the level of debate. Unfortunately, the further the specific points that are made are from my comfort/knowledge zone, the less I become able to tell nonsense from sophistication. It seems bad if my karma vote density centers on somewhat-good posts at the exclusion of very good and very bad posts. This makes me err on the side of upvoting posts I don't truly understand. I think that should be robust, since new user votes seem to carry less weight and I expect overrated nonsense to be corrected quickly, but it still seems suboptimal. It's also unclear to me whether agreement-voting factors in the sorting order. I predict it doesn't, and I would want to change how I vote if it did. Overall, I don't have a good sense of how much value I get out of seeing both axes, but on this post I do like voting with both. It fee
5handoflixue
For what it's worth, I haven't used the site in years and I picked it up just from this thread and the UI tooltips. The most confusing thing was realizing "okay, there really are two different types of vote" since I'd never encountered that before, but I can't think of much that would help (maybe mention it in the tooltip, or highlight them until the user has interacted with both?) Looking forward to it as a site-wide feature - just from seeing it at work here, it seems like a really useful addition to the site
[-]trevorΩ53028

If someone could find a way to rewrite this post, except in language comprehensible to policymakers, tech executives, or ML researchers, then it would probably achieve a lot.

[This comment is no longer endorsed by its author]Reply

Yes, please do rewrite the post, or make your own version of a post like this!! :) I don't suggest trying to persuade arbitrary policymakers of AGI risk, but I'd be very keen on posts like this optimized to be clear and informative to different audiences. Especially groups like 'lucid ML researchers who might go into alignment research', 'lucid mathematicians, physicists, etc. who might go into alignment research', etc.

Suggestion: make it a CYOA-style interactive piece, where the reader is tasked with aligning AI, and could choose from a variety of approaches which branch out into sub-approaches and so on. All of the paths, of course, bottom out in everyone dying, with detailed explanations of why. This project might then evolve based on feedback, adding new branches that counter counter-arguments made by people who played it and weren't convinced. Might also make several "modes", targeted at ML specialists, general public, etc., where the text makes different tradeoffs regarding technicality vs. vividness.

I'd do it myself (I'd had the idea of doing it before this post came out, and my preliminary notes covered much of the same ground, I feel the need to smugly say), but I'm not at all convinced that this is going to be particularly useful. Attempts to defeat the opposition by building up a massive evolving database of counter-arguments have been made in other fields, and so far as I know, they never convinced anybody.

The interactive factor would be novel (as far as I know), but I'm still skeptical.

(A... different implementation might be to use a fine-tuned language model for this; make it an AI Dungeon kind of setup, where it provides specialized counter-arguments for any suggestion. But I expect it to be less effective than a more coarse hand-written CYOA, since the readers/players would know that the thing they're talking to has no idea what it's talking about, so would disregard its words.)

Arbital was meant to support galaxy-brained attempts like this; Arbital failed.

6Thane Ruthenis
Failed as a platform for hosting galaxy-brained attempts, or failed as in every similar galaxy-brained attempt on it failed? I haven't spent a lot of time there, but my impression is that Arbital is mostly a wiki-style collection of linked articles, not a dumping ground of standalone esoterically-structured argumentative pieces. And while a wiki is conceptually similar, presentation matters a lot. A focused easily-traversable tree of short-form arguments in a wrapper that encourages putting yourself in the shoes of someone trying to fix the problem may prove more compelling. (Not to make it sound like I'm particularly attached to the idea after all. But there's a difference between "brilliant idea that probably won't work" and "brilliant idea that empirically failed".)

Arbital was a very conjunctive project, trying to do many different things, with a specific team, at a specific place and time. I wouldn't write off all Arbital-like projects based on that one data point, though I update a lot more if there are lots of other Arbital-ish things that also failed.

5ESRogs
As a person who worked on Arbitral, I agree with this.
4CronoDAS
A strange game. The only winning move is not to play. ;)
4Thane Ruthenis
I guess we should also kidnap people and force them to play it, and if they don't succeed we kill them? For realism? Wait, there's something wrong with this plan. More seriously, yeah, if you're implementing it more like a game and less like an interactive article, it'd need to contain some promise of winning. Haven't considered how to do it without compromising the core message.
5AdamB
What if "winning" consists of finding a new path not already explored-and-foreclosed? For example, each time you are faced with a list of choices of what to do, there's a final choice "I have an idea not listed here" where you get to submit a plan of action. This goes into a moderation engine where a chain of people get to shoot down the idea or approve it to pass up the chain. If the idea gets convincingly shot down (but still deemed interesting), it gets added to the story as a new branch. If it gets to the top of the moderation chain and makes EY go "Hm, that might work" then you win the game.
4Thane Ruthenis
Mmm. If the CYOA idea is implemented as a quirky-but-primarily-educational article, then sure, integrating the "adapt to feedback" capability like this would be worthwhile. Might also attach a monetary prize to submitting valuable ideas, by analogy to the ELK contest. For a game-like implementation, where you'd be playing it partly for the fun/challenge of it, that wouldn't suffice. The feedback loop's too slow, and there'd be an ugh-field around the expectation that submitting a proposal would then require arguing with the moderators about it, defending it. It wouldn't feel like a game. It'd make the upkeep cost pretty high, too, without a corresponding increase in the pay-off. Just making it open-ended might work, even without the moderation engine? Track how many branches the player explored, once they've explored a lot (i. e., are expected to "get" the full scope of the problem), there appears an option for something like "I really don't know what to do, but we should keep trying", leading to some appropriately-subtle and well-integrated call to support alignment research? Not excited about this approach either.
3Celenduin
I wonder if we could be much more effective in outreach to these groups? Like making sure that Robert Miles is sufficiently funded to have a professional team +20% (if that is not already the case). Maybe reaching out to Sabine Hossenfelder and sponsoring a video, or maybe collaborate with her for a video about this. Though I guess given her attitude towards the physics community, the work with her might be a gamble and two-edged sword. Can we get market research on what influencers have a high number of followers of ML researches/physicists/mathematicians and then work with them / sponsor them? Or maybe micro-target this demographic with facebook/google/github/stackexchange ads and point them to something? I don't know, I'm not a marketing person, but I feel like I would have seen much more of these things if we were doing enough of them. Not saying that this should be MIRI's job, rather stating that I'm confused because I feel like we as a community are not taking an action that would seem obvious to me. Especially given how recent advances in published AI capabilities seem to make the problem even much legible. Is the reason for not doing it really just that we're all a bunch of nerds who are bad at this kind of thing, or is there more to it that I'm missing? While I see that there is a lot of risk associated with such outreach increasing the amount of noise, I wonder if that tradeoff might be shifting the shorter the timelines are getting and given that we don't seem to have better plans than "having a diverse set of smart people come up with novel ideas of their own in the hope that one of those works out". So taking steps to entice a somewhat more diverse group of people into the conversation might be worth it?
[-]VaniverΩ91813

Not saying that this should be MIRI's job, rather stating that I'm confused because I feel like we as a community are not taking an action that would seem obvious to me. 

I wrote about this a bit before, but in the current world my impression is that actually we're pretty capacity-limited, and so the threshold is not "would be good to do" but "is better than my current top undone item". If you see something that seems good to do that doesn't have much in the way of unilateralist risk, you doing it is probably the right call. [How else is the field going to get more capacity?]

4Rob Bensinger
+1
1Celenduin
🤔 Not sure if I'm the right person, but it seems worth thinking about how one would maybe approach this if one were to do it. So the idea is to have an AI-Alignment PR/Social Media org/group/NGO/think tank/company that has the goal to contribute to a world with a more diverse set of high-quality ideas about how to safely align powerful AI. The only other organization roughly in this space that I can think of would be 80,000 hours, which is also somewhat more general in its goals and more conservative in its strategies. I'm not a sales/marketing person, but as I understand it, the usual metaphor to use here is a funnel? * Starting with maybe ads / sponsoring trying to reach the right people[0] (e.g. I saw Jane Street sponsor Matt Parker) * then more and more narrowing down first with introducing people to why this is an issue (orthogonality, instrumental convergence) * hopefully having them realize for themselves, guided by arguments, that this is an issue that genuinely needs solving and maybe their skills would be useful * increasing the math as needed * finally, somehow selecting for self-reliance and providing a path for how to get started with thinking about this problem by themselves / model building / independent research * or otherwise improving the overall situation (convince your congress member of something? run for congress? ...) Probably that would include copy writing (or hiring copywriters or contracting them) to go over a number of our documents to make them more digestible and actionable. So, I'm probably not the right person to get this off the ground, because I don't have a clue about any of this (not even entrepreneurship in general), but it does seem like a thing worth doing and maybe like an initiative that would get funding from whoever funds such things these days? [0] Though, maybe we should also look into a better understanding about who "the right people" are? Given that our current bunch of ML researchers/physicists/mathema
6Celenduin
On second thought: Don't we have orgs that work on AI governance/policy? I would expect them to have more likely the skills/expertise to pull this off, right?

So, here's a thing that I don't think exists yet (or, at least, it doesn't exist enough that I know about it to link it to you). Who's out there, what 'areas of responsibility' do they think they have, what 'areas of responsibility' do they not want to have, what are the holes in the overall space? It probably is the case that there are lots of orgs that work on AI governance/policy, and each of them probably is trying to consider a narrow corner of space, instead of trying to hold 'all of it'.

So if someone says "I have an idea how we should regulate medical AI stuff--oh, CSET already exists, I should leave it to them", CSET's response will probably be "what? We focus solely on national security implications of AI stuff, medical regulation is not on our radar, let alone a place we don't want competition."

I should maybe note here there's a common thing I see in EA spaces that only sometimes make sense, and so I want to point at it so that people can deliberately decide whether or not to do it. In selfish, profit-driven worlds, competition is the obvious thing to do; when someone else has discovered that you can make profits by selling lemonade, you should maybe also try to sell lemo... (read more)

3Vaniver
...yet!
4trevor
I greatly regret writing this. Yud's work is not easily distilled, it's not written such that large amounts of distillation (~50%) adds value, unless the person doing it was extremely competent. Hypothetically, it's very possible for a human to do, but empirically, everyone who has tried with this doc has failed (including me). For example, the clarifications/examples are necessary in order for the arguments to be properly cognitively operationalized; anything less is too vague. You could argue that the vast majority this post is just one big clarification. Summarized versions of these arguments clearly belong in other papers, especially papers comprehensible for policymakers, tech executives, or ML researchers. But I'm now pessimistic about the prospects of creating a summarized version of this post.
2Thane Ruthenis
I think one can write a variant of it that's fairly shorter and is even better at conveying the underlying gears-level model. My ideal version of it would start by identifying the short number of "background" core points that inform a lot of the individual entries on this list, comprehensively outlining them, then showing how various specific failures/examples mentioned here happen downstream of these core points; with each downstream example viscerally shown as the nigh-inevitable consequence of the initial soundly-established assumptions. But yeah, it's a lot of work, and there are few people I'd trust to do it right.

Eliezer, thanks for sharing these ideas so that more people can be on the lookout for failures.  Personally, I think something like 15% of AGI dev teams (weighted by success probability) would destroy the world more-or-less immediately, and I think it's not crazy to think the fraction is more like 90% or higher (which I judge to be your view).

FWIW, I do not agree with the following stance, because I think it exposes the world to more x-risk:

So far as I'm concerned, if you can get a powerful AGI that carries out some pivotal superhuman engineering task, with a less than fifty percent change of killing more than one billion people, I'll take it. 

Specifically, I think a considerable fraction of the remaining AI x-risk facing humanity stems from people pulling desperate (unsafe) moves with AGI to head off other AGI projects.  So, in that regard, I think that particular comment of yours is probably increasing x-risk a bit.  If I were a 90%-er like you, it's possible I'd endorse it, but even then it might make things worse by encouraging more desperate unilateral actions.

That said, overall I think this post is a big help, because it helps to put responsibility in... (read more)

a considerable fraction of the remaining AI x-risk facing humanity stems from people pulling desperate (unsafe) moves with AGI to head off other AGI projects

In your post “Pivotal Act” Intentions, you wrote that you disagree with contributing to race dynamics by planning to invasively shut down AGI projects because AGI projects would, in reaction, try to maintain

the ability to implement their own pet theories on how safety/alignment should work, leading to more desperation, more risk-taking, and less safety overall.

Could you give some kind of very rough estimates here? How much more risk-taking do you expect in a world given how much / how many prominent "AI safety"-affiliated people declaring invasive pivotal act intentions? How much risk-taking do you expect in the alternative, where there are other pressures (economic, military, social, whatever), but not pressure from pivotal act threats? How much safety (probability of AGI not killing everyone) do you think this buys? You write:

15% of AGI dev teams (weighted by success probability) would destroy the world more-or-less immediately

What about non-immediately, in each alternative?

1[comment deleted]
[-]VikaΩ11273

Thanks Eliezer for writing up this list, it's great to have these arguments in one place! Here are my quick takes (which mostly agree with Paul's response). 

Section A (strategic challenges?):

Agree with #1-2 and #8. Agree with #3 in the sense that we can't iterate in dangerous domains (by definition) but not in the sense that we can't learn from experiments on easier domains (see Paul's Disagreement #1). 

Mostly disagree with #4 - I think that coordination not to build AGI (at least between Western AI labs) is difficult but feasible, especially after a warning shot. A single AGI lab that decides not to build AGI can produce compelling demos of misbehavior that can help convince other actors. A number of powerful actors coordinating not to build AGI could buy a lot of time, e.g. through regulation of potential AGI projects (auditing any projects that use a certain level of compute, etc) and stigmatizing deployment of potential AGI systems (e.g. if it is viewed similarly to deploying nuclear weapons). 

Mostly disagree with the pivotal act arguments and framing (#6, 7, 9). I agree it is necessary to end the acute risk period, but I find it unhelpful when this is frame... (read more)

Since Divia said, and Eliezer retweeted, that good things might happen if people give their honest, detailed reactions:

My honest, non-detailed reaction is AAAAAAH. In more detail -

  1. Yup, this seems right.
  2. This is technobabble to me, since I don't actually understand nanomachines, but it makes me rather more optimistic about my death being painless than my most likely theory, which is that a superhuman AI takes over first and has better uses for our atoms later.
  3. (If we had unlimited retries - if every time an AGI destroyed all the galaxies we got to go back in time four years and try again - we would in a hundred years figure out which bright ideas actually worked.) My brain immediately starts looking for ways to set up some kind of fast testing for ways to do this in a closed, limited world without letting it know ours exists... which is already answered below, under 10. Yup, doomed.
  4. And then we all died.
  5. Yup.
  6. I imagine it would be theoretically - but not practically - possible to fire off a spaceship accelerating fast enough (that is, with enough lead time) that it could outrun the AI and so escape an Earth about to be eaten by an AI (a pivotal act well short of melting all CPUs that wou
... (read more)

[Deleted.]

[This comment is no longer endorsed by its author]Reply
1kubanetics
This is another reply in this vein, I'm quite new to this so don't feel obliged to read through. I just told myself I will publish this. I agree (90-99% agreement) with almost all of the points Eliezer made. And the rest is where I probably didn't understand enough or where there's no need for a comment, e.g.: 1. - 8.  agree 9. Not sure if I understand it right - if the AGI has been successfully designed not to kill everyone then why need oversight? If it is capable to do so and the design fails then on the other hand what would our oversight do? I don't think this is like the nuclear cores. Feels like it's a bomb you are pretty sure won't go off at random but if it does your oversight won't stop it. 10. - 14. - agree 15. - I feel like I need to think about it more to honestly agree. 16. - 18. - agree 19. - to my knowledge, yes 20. - 23. - agree 24. - initially I put "80% agree" to the first part of the argument here (that   but then discussing it with my reading group I reiterated this few times and begun to agree even more grasping the complexity of something like CEV. 25. - 29. - agree 30. - agree, although wasn't sure about  I think that the key part of this claim is "all the effects of" and I wasn't sure whether we have to understand all, but of course we have to be sure one of the effects is not human extintion then yes, so for "solving alignment" also yes. 31. - 34. - agree 35. - no comment, I have to come back to this once I graps LDT better 36. - agree 37. - no comment, seems like a rant 😅 38. - agree 39. - ok, I guess 40. - agree, I'm glad some people want to experiment with the financing of research re 40. 41. - agree , although I agree with some of the top comments on this, e.g. evhub's 42. - agree 43. - agree, at least this is what it feels like  
2Qumeric
Regarding 9: I believe it's when you are successful enough that your AGI doesn't instantly kill you immediately but it still can kill you in the process of using it. It's in the context of a pivotal act, so it assumes you will operate it to do something significant and potentially dangerous.

As a bystander who can understand this, and find the arguments and conclusions sound, I must say I feel very hopeless and "kinda" scared at this point. I'm living in at least an environment, if not a world, where even explaining something comparatively simple like how life extension is a net good is a struggle. Explaining or discussing this is definitely impossible - I've tried with the cleverer, more transhumanistic/rationalistic minded people I know, and it just doesn't click for them, to the contrary, I find people like to push in the other direction, as if it were a game.

And at the same time, I realize it is unlikely I can contribute anything remotely significant to a solution myself. So I can only spectate. This is literally maddening, especially so when most everyone seems to underreact.

If it's any consolation, you would not feel more powerful or less scared if you were myself.

6Vincent Fagot
Well, obviously, it won't be consolation enough, but I can certainly revel in some human warmth inside by knowing I'm not alone in feeling like this.

This might sound absurd, but I legit think that there's something that most people can do. Being something like radically publicly honest and radically forgiving and radically threat-aware, in your personal life, could contribute to causing society in general to be radically honest and forgiving and threat-aware, which might allow people poised to press the Start button on AGI to back off. 

ETA: In general, try to behave in a way such that if everyone behaved that way, the barriers to AGI researchers noticing that they're heading towards ending the world would be lowered / removed. You'll probably run up against some kind of resistance; that might be a sign that some social pattern is pushing us into cultural regimes where AGI researchers are pushed to do world-ending stuff. 

4elioll
Vincent Fagot: Where do you live (in general terms if you can provide it, feel free not to dox yourself if you don't want to)? I live in countryside Brazil, so I can strongly relate.
[-]lcΩ32426

That requires, not the ability to read this document and nod along with it, but the ability to spontaneously write it from scratch without anybody else prompting you; that is what makes somebody a peer of its author.  It's guaranteed that some of my analysis is mistaken, though not necessarily in a hopeful direction.  The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so.  Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try.  I'm not particularly hopeful of this turning out to be true in real life, but I suppose it's one possible place for a "positive model violation" (miracle).  The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years

... (read more)

I tried something like this much earlier with a single question, "Can you explain why it'd be hard to make an AGI that believed 222 + 222 = 555", and got enough pushback from people who didn't like the framing that I shelved the effort.

I am interested in what kind of pushback you got from people.

My attempt (thought about it for a minute or two):

Because arithmetic is useful, and the self-contradictory version of arithmetic, where 222+222=555 allows you to prove anything and is useless. Therefore, a smart AI that wants and can invent useful abstractions will invent its own (isomorphic to our arithmetic, in which 222+222=444) arithmetic from scratch and will use it for practical purposes, even if we can force it not to correct an obvious error.

8DaemonicSigil
I think this is the right answer. Just to expand on this a bit: The problem isn't necessarily that 222+222=555 leads to a contradiction with the rest of arithmetic. One can imagine that instead of defining "+" using "x+Sy=y+Sx", we could give it a much more complex definition where there is a special case carved out for certain values like 222. The issue is that the AI has no reason to use this version of "+" and will define some other operation that works just like actual addition. Even if we ban the AI from using "x+Sy=y+Sx" to define any operations, it will choose the nearest thing isomorphic to addition that we haven't blocked, because addition is so common and useful. Or maybe it will use the built-in addition, but whenever it wants to add n+m, it instead adds 4n+4m, since our weird hack doesn't affect the subgroup consisting of integers divisible by 4.
9Ben Pace
FWIW the framing seems exciting to me.
7lc
So, there are five possibilities here: 1. MIRI's top researchers don't understand, or can't explain, why having incorrect maps makes it harder to navigate the territory and leads to more incorrect beliefs. Something I find very hard to believe even if you're being totally forthright. 2. You asked some random people near you who don't represent the top crust of alignment researchers, which is obviously irrelevant. 3. There's some very subtle ambiguity to this that I'm completely unaware of. 4. You asked people in a way that heavily implied it was some sort of trick question and they should get more information, then assumed they were stupid because they asked followup questions. 5. This comment is written almost deliberately misleadingingly. You're just explaining a random story about how you ran out of energy to ask Nate Soares to write a post. I guarantee you that most reasonably intelligent people, if asked this question after reading the sequences in a way that they didn't expect was designed to trip them up, would get it correctly. I simply do not believe that everyone around you is as stupid as you are implying, such that you should have shelved the effort. ---------------------------------------- EDIT: 😭

You didn't get the answer correct yourself.

3lc
Damn aight. Would you be willing to explain for the sake of my own curiosity? I don't have the gears to understand why that wouldn't be at least one reason.
2Catnee
If this is "kind of a test for capable people" i think it should be remained unanswered, so anyone else could try. My take would be: because if 222+222=555 then 446=223+223 = 222+222+1+1=555+1+1=557. With this trick "+" and "=" stops meaning anything, any number could be equal to any other number. If you truly believe in one such exeption, the whole arithmetic cease to exist because now you could get any result you want following simple loopholes, and you will either continue to be paralyzed by your own beliefs, or will correct yourself
5lc
This is what I meant by "leads to other incorrect beliefs", so apparently not.

Ok, so here's my take on the "222 + 222 = 555" question.

First, suppose you want your AI to not be durably wrong, so it should update on evidence. This is probably implemented by some process that notices surprises, goes back up the cognitive graph, and applies pressure to make it have gone the right way instead.

Now as it bops around the world, it will come across evidence about what happens when you add those numbers, and its general-purpose "don't be durably wrong" machinery will come into play. You need to not just sternly tell it "222 + 222 = 555" once, but have built machinery that will protect that belief from the update-on-evidence machinery, and which will also protect itself from the update-on-evidence machinery.

Second, suppose you want your AI to have the ability to discover general principles. This is probably implemented by some process that notices patterns / regularities in the environment, and builds some multi-level world model out of it, and then makes plans in that multi-level world model. Now you also have some sort of 'consistency-check' machinery, which scans thru the map looking for inconsistencies between levels, goes back up the cognitive graph, and applies p... (read more)

2lc
No one is going to believe me, but when I originally wrote that comment, my brain read something like "why would an AI that believed 222 + 222 = 555 have a hard time". Only figured it out now after reading your reply.  Part one of this is what I would've come up with, though I'm not particularly certain it's correct.
9Ben Pace
Sounds like the beginnings of a bet.
[-]lc2113

I will absolutely 100% do it in the spirit of good epistemics.

Edit: I'm glad Eliezer didn't take me up on this lol

5Rob Bensinger
I'd have guessed the disagreement wasn't about whether "222 + 222 = 555" is an incorrect map, or about whether incorrect maps often make it harder to navigate the territory, but about something else. (Maybe 'I don't want to think about this because it seems irrelevant/disanalogous to alignment work'?) And I'd have guessed the answer Eliezer was looking for was closer to 'the OP's entire Section B' (i.e., a full attempt to explain all the core difficulties), not a one-sentence platitude establishing that there's nonzero difficulty? But I don't have inside info about this experiment.
5lc
I'd have guessed that too, which is why I would have preferred him to say that they disagreed on |whatever meta question he's actually talking about| instead of implying disagreement on |other thing that makes his disappointment look more reasonable|. That story sounds much more cogent, but it's not the primary interpretation of "I asked them a single question" followed by the quoted question. Most people don't go on 5 paragraph rants in response to single questions, and when they do they tend to ask clarifying details regardless of how well they understand the prompt, so they know they're responding as intended.
5Koen.Holtman
Interesting. I kind of like the framing here, but I have written a paper and sequence on the exact opposite question, on why it would be easy to make an AGI that believes 222+222=555, if you ever had AGI technology, and what you can do with that in terms of safety. I can honestly say however that the project of writing that thing, in a way that makes the math somewhat accessible, was not easy.
1Trevor Cappallo
For the record, I found that line especially effective. I stopped, reread it, stopped again, had to think it through for a minute, and then found satisfaction with understanding.
0handoflixue
If you had an AI that could coherently implement that rule, you would already be at least half a decade ahead of the rest of humanity. You couldn't encode "222 + 222 = 555" in GPT-3 because it doesn't have a concept of arithmetic, and there's no place in the code to bolt this together. If you're really lucky and the AI is simple enough to be working with actual symbols, you could maybe set up a hack like "if input is 222 + 222, return 555, else run AI" but that's just bypassing the AI.  Explaining "222 + 222 = 555" is a hard problem in and of itself, much less getting the AI to properly generalize to all desired variations (is "two hundred and twenty two plus two hundred and twenty two equals five hundred and fifty five" also desired behavior? If I Alice and Bob both have 222 apples, should the AI conclude that the set {Alice, Bob} contains 555 apples? Getting an AI that evolves a universal math module because it noticed all three of those are the same question would be a world-changing break through)
2lc
FvC5IXzxQC+I3vstFGIUWlbtTFgRsa8bt0mKPN3K0UNZBkI7OLDBjjapp1+CoJPRYEqRM015PSZXUuh4OWwJEUBOTeLHeheLteG9LxGiuS6YqnV/PN0s0S/TyYjCPrF0vDHFDBy3IHW4qDQguf5QAA==
[-]William_SΩ82222

Could I put in a request to see a brain dump from Eliezer of ways to gain dignity points?

I'm not Eliezer, but my high-level attempt at this:

[...] The things I'd mainly recommend are interventions that:

  • Help ourselves think more clearly. (I imagine this including a lot of trying-to-become-more-rational, developing and following relatively open/honest communication norms, and trying to build better mental models of crucial parts of the world.)
  • Help relevant parts of humanity (e.g., the field of ML, or academic STEM) think more clearly and understand the situation.
  • Help us understand and resolve major disagreements. (Especially current disagreements, but also future disagreements, if we can e.g. improve our ability to double-crux in some fashion.)
  • Try to solve the alignment problem, especially via novel approaches.
    • In particular: the biggest obstacle to alignment seems to be 'current ML approaches are super black-box-y and produce models that are very hard to understand/interpret'; finding ways to better understand models produced by current techniques, or finding alternative techniques that yield more interpretable models, seems like where most of the action is.
  • Think about the space of relatively-plausible "miracles" [i.e., positive model violations], think about future evide
... (read more)

Lots I disagree with here, so let's go through the list.

There are no pivotal weak acts.

Strong disagree.  

EY and I don't seem to agree that "nuke every semiconductor fab" is a weakly pivotal act (since I think AI is hardware-limited and he thinks it is awaiting a clever algorithm).  But I think even "build nanobots that melt every GPU" could be built using an AI that is aligned in the "less than 50% chance of murdering us all" sense.  For example, we could simulate a bunch of human-level scientists trying to build nanobots and also checking each-other's work.

On anything like the standard ML paradigm, you would need to somehow generalize optimization-for-alignment you did in safe conditions, across a big distributional shift to dangerous conditions.

Nope.  I think that you could build a useful AI (e.g. the hive of scientists) without doing any out-of-distribution stuff.  

there is no known way to use the paradigm of loss functions, sensory inputs, and/or reward inputs, to optimize anything within a cognitive system to point at particular things within the environment

I am significantly more optimistic about explainable AI than EY.

There is no analogous truth abou

... (read more)

For example, we could simulate a bunch of human-level scientists trying to build nanobots and also checking each-other's work.

That is not passively safe, and therefore not weak. For now forget the inner workings of the idea: at the end of the process you get a design for nanobots that you have to build and deploy in order to do the pivotal act. So you are giving a system built by your AI the ability to act in the real world. So if you have not fully solved the alignment problem for this AI, you can't be sure that the nanobot design is safe unless you are capable enough to understand the nanobots yourself without relying on explanations from the scientists.

And even if we look into the inner details of the idea: presumably each individual scientist-simulation is not aligned (if they are, then for that you need to have solved the alignment problem beforehand). So you have a bunch of unaligned human-level agents who want to escape, who can communicate among themselves (at the very least they need to be able to share the nanobot designs with each other for criticism).

You'd need to be extremely paranoid and scrutinize each communication between the scientist-simulations to prevent them f... (read more)

6Vaniver
Note that the difficulty in "nuke every semiconductor fab" is in "acquire the nukes and use them", not in "googling the address of semiconductor fabs". It seems to me like nuclear nonproliferation is one of the few things that actually has international collaboration with teeth, such that doing this on your own is extremely challenging, and convincing institutions that already have nuclear weapons to use them on semiconductor fabs also seems extremely challenging. [And if you could convince them to do that, can't you convince them to smash the fabs with hammers, or detain the people with relevant experience on some beautiful tropical island instead of murdering them and thousands of innocent bystanders?]
-14[comment deleted]
1JakubK
This comment makes many distinct points, so I'm confused why it currently has -13 agreement karma. Do people really disagree with all of these points?
1Jackson Wagner
"We could simulate a bunch of human-level scientists trying to build nanobots." This idea seems far-fetched: * If it was easy to create nanotechnology by just hiring a bunch of human-level scientists, we could just do that directly, without using AI at all. * Perhaps we could simulate thousands and thousands of human-level intelligences (although of course these would not be remotely human-like intelligences; they would be part of a deeply alien AI system) at accelerated speeds.  But this seems like it would probably be more hardware-intensive than just turning up the dial and running a single superintelligence.  In other words, this proposal seems to have a very high "alignment tax".  And even after paying that hefty tax, I'd still be worried about alignment problems if I was simulating thousands of alien intelligences at super-speed! * Besides all the hardware you'd need, wouldn't this be very complicated to implement on the software side, with not much overlap with today's AI designs?   Has anyone done a serious analysis of how much semiconductor capacity could be destroyed using things like cruise missiles + nationalizing and shutting down supercomputers?  I would be interested to know if this is truly a path towards disabling like 90% of the world's useful-to-AI-research compute, or if the number is much smaller because there is too much random GPU capacity out there in the wild even when you commandeer TSMC fabs and AWS datacenters.

If there was one thing that I could change in this essay, it would be to clearly outline that the existence of nanotechnology advanced enough to do things like melt GPUs isn't necessary even if it is sufficient for achieving singleton status and taking humanity off the field as a meaningful player.

Whenever I see people fixate on critiquing that particular point, I need to step in and point out that merely existing tools and weapons (is there a distinction?) suffice for a Superintelligence to be able to kill the vast majority of humans and reduce our threat to it to negligible levels. Be that wresting control of nuclear arsenals to initiate MAD or simply extrapolating on gain-of-function research to produce extremely virulent yet lethal pathogens that can't be defeated before the majority of humans are infected, such options leave a small minority of humans alive to cower in the wreckage until the biosphere is later dismantled.

That's orthogonal to the issue of whether such nanotechnology is achievable for a Superintelligent AGI, it merely reduces the inferential distance the message has to be conveyed as it doesn't demand familiarity with Drexler.

(Advanced biotechnology already is nanotechnology, but the point is that no stunning capabilities need to be unlocked for an unboxed AI to become immediately lethal)

4sullyj3
Right, alignment advocates really underestimate the degree to which talking about sci-fi sounding tech is a sticking point for people

The counter-concern is that if humanity can't talk about things that sound like sci-fi, then we just die. We're inventing AGI, whose big core characteristic is 'a technology that enables future technologies'. We need to somehow become able to start actually talking about AGI.

One strategy would be 'open with the normal-sounding stuff, then introduce increasingly weird stuff only when people are super bought into the normal stuff'. Some problems with this:

  • A large chunk of current discussion and research happens in public; if it had to happen in private because it isn't optimized for looking normal, a lot of it wouldn't happen at all.
    • More generally: AGI discourse isn't an obstacle course or a curriculum, such that we can control the order of ideas and strictly segregate the newbies from the old guard. Blog posts, research papers, social media exchanges, etc. freely circulate among people of all varieties.
  • It's a dishonest/manipulative sort of strategy — which makes it ethically questionable, is liable to fuel other trust-degrading behavior in the community, and is liable to drive away people with higher discourse standards.
  • A lot of the core arguments and hazards have no 'normal-soundin
... (read more)
3sullyj3
Fair point, and one worth making in the course of talking about sci-fi sounding things! I'm not asking anyone to represent their beliefs dishonestly, but rather introduce them gently. I'm personally not an expert, but I'm not convinced of the viability of nanotech, so if it's not necessary (rather it's sufficient) to the argument, it seems prudent to stick to more clearly plausible pathways to takeover as demonstrations of sufficiency, while still maintaining that weirder sounding stuff is something one ought to expect when dealing with something much smarter than you.
8Rob Bensinger
If you're trying to persuade smart programmers who are somewhat wary of sci-fi stuff, and you think nanotech is likely to play a major role in AGI strategy, but you think it isn't strictly necessary for the current argument you're making, then my default advice would be: * Be friendly and patient; get curious about the other person's perspective, and ask questions to try to understand where they're coming from; and put effort into showing your work and providing indicators that you're a reasonable sort of person. * Wear your weird beliefs on your sleeve; be open about them, and if you want to acknowledge that they sound weird, feel free to do so. At least mention nanotech, even if you choose not to focus on it because it's not strictly necessary for the argument at hand, it comes with a larger inferential gap, etc.
-2mukashi
I think that even this scenario is implausible. I have the impression we are overestimating how easy is to wipe all humans quickly
6CronoDAS
I'm retreating from my previous argument a bit. The AGI doesn't need to cause literal human extinction with a virus; if it can cause enough damage to collapse human industrial civilization (while being able to survive said collapse) then that would also achieve most of the AGI's goal of being able to do what it wants without humans stopping it. Naturally occurring pathogens from Europe devastated Native American populations after Columbus; throw a bunch of bad enough novel viruses at us at once and you probably could knock humanity back to the metaphorical Stone Age.
0mukashi
I find that more plausible. Also horrifying and worth fighting against, but not what EY is saying

I find that more plausible. Also horrifying and worth fighting against, but not what EY is saying

Note that EY is saying "there exists a real plan that is at least as dangerous as this one"; if you think there is such a plan, then you can agree with the conclusion, even if you don't agree with his example. [There is an epistemic risk here, if everyone mistakenly believes that a different doomsday plan is possible when someone else knows why that specific plan won't work, and so if everyone pooled all their knowledge they could know that none of the plans will work. But I'm moderately confident we're instead in a world with enough vulnerabilities that broadcasting them makes things worse instead of better.]

5[comment deleted]
3[comment deleted]

While I share a large degree of pessimism for similar reasons, I am somewhat more optimistic overall.  

Most of this comes from generic uncertainty and epistemic humility; I'm a big fan of the inside view, but it's worth noting that this can (roughly) be read as a set of 42 statements that need to be true for us to in fact be doomed, and statistically speaking it seems unlikely that all of these statements are true.

However, there are some more specific points I can point to where I think you are overconfident, or at least not providing good reasons for such a high level of confidence (and to my knowledge nobody has).  I'll focus on two disagreements which I think are closest to my true disagreements.

1) I think safe pivotal "weak" acts likely do exist.  It seems likely that we can access vastly superhuman capabilities without inducing huge x-risk using a variety of capability control methods.  If we could build something that was only N<<infinity times smarter than us, then intuitively it seems unlikely that it would be able to reverse engineer details of the outside world or other AI systems source code (cf 35) necessary to break out of the box or start coo... (read more)

8Rob Bensinger
I don't think these statements all need to be true in order for p(doom) to be high, and I also don't think they're independent. Indeed, they seem more disjunctive than conjunctive to me; there are many cases where any one of the claims being true increases risk substantially, even if many others are false.
1David Scott Krueger (formerly: capybaralet)
I basically agree.   I am arguing against extreme levels of pessimism (~>99% doom).  

Thanks for writing this, I agree that people have underinvested in writing documents like this. I agree with many of your points, and disagree with others. For the purposes of this comment, I'll focus on a few key disagreeements.

My model of this variety of reader has an inside view, which they will label an outside view, that assigns great relevance to some other data points that are not observed cases of an outer optimization loop producing an inner general intelligence, and assigns little importance to our one data point actually featuring the phenomenon in question. Consider skepticism, if someone is ignoring this one warning, especially if they are not presenting equally lethal and dangerous things that they say will go wrong instead.

There are some ways in which AGI will be analogous to human evolution. There are some ways in which it will be disanalogous. Any solution to alignment will exploit at least one of the ways in which it's disanalogous. Pointing to the example of humans without analysing the analogies and disanalogies more deeply doesn't help distinguish between alignment proposals which usefully exploit disanalogies, and proposals which don't.

Alpha Zero blew pas

... (read more)

Maybe one way to pin down a disagreement here: imagine the minimum-intelligence AGI that could write this textbook (including describing the experiments required to verify all the claims it made) in a year if it tried. How many Yudkowsky-years does it take to safely evaluate whether following a textbook which that AGI spent a year writing will kill you?

Infinite?  That can't be done?

6Richard_Ngo
Hmm, okay,  here's a variant. Assume it would take N Yudkowsky-years to write the textbook from the future described above. How many Yudkowsky-years does it take to evaluate a textbook that took N Yudkowsky-years to write, to a reasonable level of confidence (say, 90%)?
7Eliezer Yudkowsky
If I know that it was written by aligned people?  I wouldn't just be trying to evaluate it myself; I'd try to get a team together to implement it, and understanding it well enough to implement it would be the same process as verifying whatever remaining verifiable uncertainty was left about the origins, where most of that uncertainty is unverifiable because the putative hostile origin is plausibly also smart enough to sneak things past you.
5Richard_Ngo
Sorry, I should have been clearer. Let's suppose that a copy of you spent however long it takes to write an honest textbook with the solution to alignment (let's call it N Yudkowsky-years), and an evil copy of you spent N Yudkowsky-years writing a deceptive textbook trying to make you believe in a false solution to alignment, and you're given one but not told which. How long would it take you to reach 90% confidence about which you'd been given? (You're free to get a team together to run a bunch of experiments and implementations, I'm just asking that you measure the total work in units of years-of-work-done-by-people-as-competent-as-Yudkowsky. And I should specify some safety threshold too - like,  in the process of reaching 90% confidence, incurring less than 10% chance of running an experiment which kills you.)

Depends what the evil clones are trying to do.

Get me to adopt a solution wrong in a particular direction, like a design that hands the universe over to them?  I can maybe figure out the first time through who's out to get me, if it's 200 Yudkowsky-years.  If it's 200,000 Yudkowsky-years I think I'm just screwed.

Get me to make any lethal mistake at all?  I don't think I can get to 90% confidence period, or at least, not without spending an amount of Yudkowsky-time equivalent to the untrustworthy source.

[-]p.b.Ω91610

Humans don't explicitly pursue inclusive genetic fitness; outer optimization even on a very exact, very simple loss function doesn't produce inner optimization in that direction.

Humans haven't been optimized to pursue inclusive genetic fitness for very long, because humans haven't been around for very long. Instead they inherited the crude heuristics pointing towards inclusive genetic fitness from their cognitively much less sophisticated predecessors. And those still kinda work!

If we are still around in a couple of million years I wouldn't be surprised if there was inner alignment in the sense that almost all humans in almost all practically encountered environments end up consciously optimising inclusive genetic fitness. 

More generally, there is no known way to use the paradigm of loss functions, sensory inputs, and/or reward inputs, to optimize anything within a cognitive system to point at particular things within the environment - to point to latent events and objects and properties in the environment, rather than relatively shallow functions of the sense data and reward.

Generally, I think that people draw the wrong conclusions from mesa-optimisers and... (read more)

6Rob Bensinger
The OP isn't claiming that alignment is impossible. I don't understand the point you're making here.
[-]p.b.Ω9168

The point I'm making is that the human example tells us that: 

If first we realize that we can't code up our values, therefore alignment is hard. Then, when we realize that mesa-optimisation is a thing. we shouldn't update towards "alignment is even harder". We should update in the opposite direction. 

Because the human example tells us that a mesa-optimiser can reliably point to a complex thing even if the optimiser points to only a few crude things. 

But I only ever see these three points, human example, inability to code up values, mesa-optimisation to separately argue for "alignment is even harder than previously thought". But taken together that is just not the picture. 

Humans point to some complicated things, but not via a process that suggests an analogous way to use natural selection or gradient descent to make a mesa-optimizer point to particular externally specifiable complicated things.

8TurnTrout
Why do you think that? Why is the process by which humans come to reliably care about the real world, not a process we could leverage analogously to make AIs care about the real world?  Likewise, when you wrote, Where is the accident? Did evolution accidentally find a way to reliably orient terminal human values towards the real world? Do people each, individually, accidentally learn to terminally care about the real world? Because the former implies the existence of a better alignment paradigm (that which occurs within the human brain, to take an empty-slate human and grow them into an intelligence which terminally cares about objects in reality), and the latter is extremely unlikely. Let me know if you meant something else. EDIT: Updated a few confusing words.
8Rob Bensinger
Maybe I'm not understanding your proposal, but on the face of it this seems like a change of topic. I don't see Eliezer claiming 'there's no way to make the AGI care about the real world vs. caring about (say) internal experiences in its own head'. Maybe he does think that, but mostly I'd guess he doesn't care, because the important thing is whether you can point the AGI at very, very specific real-world tasks. Same objection/confusion here, except now I'm also a bit confused about what you mean by "orient people towards the real world". Your previous language made it sound like you were talking about causing the optimizer's goals to point at things in the real world, but now your language makes it sound like you're talking about causing the optimizer to model the real world or causing the optimizer to instrumentally care about the state of the real world....? Those all seem very different to me. Or, in summary, I'm not seeing the connection between: * "Terminally valuing anything physical at all" vs. "terminally valuing very specific physical things". * "Terminally valuing anything physical at all" vs. "instrumentally valuing anything physical at all". * "Terminally valuing very specific physical things" vs. "instrumentally valuing very specific physical things". * Any of the above vs. "modeling / thinking about physical things at all", or "modeling / thinking about very specific physical things".
9TurnTrout
Hm, I'll give this another stab. I understand the first part of your comment as "sure, it's possible for minds to care about reality, but we don't know how to target value formation so that the mind cares about a particular part of reality." Is this a good summary?  Let me distinguish three alignment feats: 1. Producing a mind which terminally values sensory entities.  2. Producing a mind which reliably terminally values some kind of non-sensory entity in the world, like dogs or bananas.  1. AFAIK we have no idea how to ensure this happens reliably -- to produce an AGI which terminally values some element of {diamonds, dogs, cats, tree branches, other real-world objects}, such that there's a low probability that the AGI actually just cares about high-reward sensory observations.  2. In other words: Design a mind which cares about anything at all in reality which isn't a shallow sensory phenomenon which is directly observable by the agent. Like, maybe I have a mind-training procedure, where I don't know what the final trained mind will value (dogs, diamonds, trees having particular kinds of cross-sections at year 5 of their growth), but I'm damn sure the AI will care about something besides its own sensory signals.  3. I was, first, pointing out that this problem has to be solvable, since the human genome solves it millions of times every day!  3. Producing a mind which reliably terminally values a specific non-sensory entity, like diamonds. 1. Design a mind which cares about a particular kind of object. We could target the mind-training process to care about diamonds, or about dogs, or about trees, but to solve this problem, we have to ensure the trained mind significantly cares about one kind of real-world entity in particular. Therefore, feat #3 is strictly harder than feat #2. 2. This is what you point out as a potential crux. (EDIT: Added a few sub-points to clarify list.) From my shard theory document:  As you point out, we obviously

I understand the first part of your comment as "sure, it's possible for minds to care about reality, but we don't know how to target value formation so that the mind cares about a particular part of reality." Is this a good summary? 

Yes!

I was, first, pointing out that this problem has to be solvable, since the human genome solves it millions of times every day! 

True! Though everyone already agreed (e.g., EY asserted this in the OP) that it's possible in principle. The updatey thing would be if the case of the human genome / brain development suggests it's more tractable than we otherwise would have thought (in AI).

Seems to me like it's at least a small update about tractability, though I'm not sure it's a big one? Would be interesting to think about the level of agreement between different individual humans with regard to 'how much particular external-world things matter'. Especially interesting would be cases where humans consistently, robustly care about a particular external-world thingie even though it doesn't have a simple sensory correlate.

(E.g., humans developing to care about sex is less promising insofar as it depends on sensory-level reinforcement such as orgasm... (read more)

6TurnTrout
Feat #2 is: Design a mind which cares about anything at all in reality which isn't a shallow sensory phenomenon which is directly observable by the agent. Like, maybe I have a mind-training procedure, where I don't know what the final trained mind will value (dogs, diamonds, trees having particular kinds of cross-sections at year 5 of their growth), but I'm damn sure the AI will care about something besides its own sensory signals. Such a procedure would accomplish feat #2, but not #3. Feat #3 is: Design a mind which cares about a particular kind of object. We could target the mind-training process to care about diamonds, or about dogs, or about trees, but to solve this problem, we have to ensure the trained mind significantly cares about one kind of real-world entity in particular. Therefore, feat #3 is strictly harder than feat #2. I actually think that the dog- and diamond-maximization problems are about equally hard, and, to be totally honest, neither seems that bad[1] in the shard theory paradigm.  Surprisingly, I weakly suspect the harder part is getting the agent to maximize real-world dogs in expectation, not getting the agent to maximize real-world dogs in expectation. I think "figure out how to build a mind which cares about the number of real-world dogs, such that the mind intelligently selects plans which lead to a lot of dogs" is significantly easier than building a dog-maximizer. 1. ^ I appreciate that this claim is hard to swallow. In any case, I want to focus on inferentially-closer questions first, like how human values form.
7Vaniver
IMO this process seems pretty unreliable and fragile, to me. Drugs are popular; video games are popular; people-in-aggregate put more effort into obtaining imaginary afterlives than life extension or cryonics. But also humans have a much harder time 'optimizing against themselves' than AIs will, I think. I don't have a great mechanistic sense of what it will look like for an AI to reliably care about the real world.
3TurnTrout
One of the problems with English is that it doesn't natively support orders of magnitude for "unreliable." Do you mean "unreliable" as in "between 1% and 50% of people end up with part of their values not related to objects-in-reality", or as in "there is no a priori reason why anyone would ever care about anything not directly sensorially observable, except as a fluke of their training process"? Because the latter is what current alignment paradigms mispredict, and the former might be a reasonable claim about what really happens for human beings. EDIT:  My reader-model is flagging this whole comment as pedagogically inadequate, so I'll point to the second half of section 5 in my shard theory document.
6lc
Humans came to their goals while being trained by evolution on genetic inclusive fitness, but they don't explicitly optimize for that. They "optimize" for something pretty random, that looks like genetic inclusive fitness in the training environment but then in this weird modern out-of-sample environment looks completely different. We can definitely train an AI to care about the real world, but his point is that, by doing something analogous to what happened with humans, we will end up with some completely different inner goal than the goal we're training for, as happened with humans.
2TurnTrout
I'm not talking about running evolution again, that is not what I meant by "the process by which humans come to reliably care about the real world." The human genome must specify machinery which reliably grows a mind which cares about reality. I'm asking why we can't use the alignment paradigm leveraged by that machinery, which is empirically successful at pointing people's values to certain kinds of real-world objects.
5lc
Ah, I misunderstood.  Well, for starters, because if the history of ML is anything to go by, we're gonna be designing the thing analogous to evolution, and not the brain. We don't pick the actual weights in these transformers, we just design the architecture and then run stochastic gradient descent or some other meta-learning algorithm. That meta-learning algorithm is going to be what decides to go in the DNA, so in order to get the DNA right, we will need to get the meta-learning algorithm correct. Evolution doesn't have much to teach us about that except as a negative example. But (I think) the answer is similar to this:
1TurnTrout
But, ah, the genome also doesn't "pick the actual weights" for the human brain which it later grows. So whatever the brain does to align people to care about latent real-world objects, I strongly believe that that process must be compatible with blank-slate initialization and then learning. In the evolution/mainstream-ML analogy, we humans are specifying the DNA, not the search process over DNA specifications. We specify the learning architecture, and then the learning process fills in the rest.    I confess that I already have a somewhat sharp picture of the alignment paradigm used by the brain, that I already have concrete reasons to believe it's miles better than anything we have dreamed so far. I was originally querying what Eliezer thinks about the "genome->human alignment properties" situation, rather than expressing innocent ignorance of how any of this works.
2lc
I think I disagree with you, but I don't really understand what you're saying or how these analogies are being used to point to the real world anymore. It seems to me like you might be taking something that makes the problem of "learning from evolution" even more complicated (evolution -> protein -> something -> brain vs. evolution -> protein -> brain) and using that to argue the issues are solved, in the same vein as the "just don't use a value function" people. But I haven't read shard theory, so, GL. You mean, we are specifying the ATCG strands, or we are specifying the "architecture" behind how DNA influences the development of the human body? It seems to me like we are definitely also choosing how the search for the correct ATCG strands and how they're identified, in this analogy. The DNA doesn't "align" new babies out of the womb, it's just a specification of how to copy the existing, already """aligned""" code.
4TurnTrout
ah, no, this isn't what I'm saying. Hm. Let me try again. The following is not a handwavy analogy, it is something which actually happened:  1. Evolution found the human genome.  2. The human genome specifies the human brain. 3. The human brain learns most of its values and knowledge over time. 4. Human brains reliably learn to care about certain classes of real-world objects like dogs.   Therefore, somewhere in the "genome -> brain -> (learning) -> values" process, there must be a process which reliably produces values over real-world objects. Shard theory aims to explain this process. The shard-theoretic explanation is actually pretty simple. Furthermore, we don't have to rerun evolution to access this alignment process. For the sake of engaging with my points, please forget completely about running evolution. I will never suggest rerunning evolution, because it's unwise and irrelevant to my present points. I also currently don't see why the genome's alignment process requires more than crude hard-coded reward circuitry, reinforcement learning, and self-supervised predictive learning. 
3FireStormOOO
That does seem worth looking at and there's probably ideas worth stealing from biology.  I'm not sure you can call that a robustly aligned system that's getting bootstrapped though.  Existing in a society of (roughly) peers and the lack of a huge power disparity between any given person and the rest of humans is anologous to the AGI that can't take over the world yet.  Humans that aquire significant power do not seem aligned wrt what a typical person would profess to and outwardly seem to care about. I think your point still mostly follows despite that; even when humans can be deceptive and power seeking, there's an astounding amount of regularity in what we end up caring about.
2TurnTrout
Yes, this is my claim. Not that eg >95% of people form values which we would want to form within an AGI.
1David Johnston
Humans can, to some extent, be pointed to complicated external things. This suggests that using natural selection on biology can get you mesa-optimizers that can be pointed to particular externally specifiable complicated things. Doesn't prove it (or, doesn't prove you can do it again), but you only asked for a suggestion.
7Eliezer Yudkowsky
Humans can be pointed at complicated external things by other humans on their own cognitive level, not by their lower maker of natural selection.
2TurnTrout
I don't think I understand what, exactly, is being discussed. Are "dogs" or "flowers" or "people you meet face-to-face" examples of "complicated external things"? 
1David Johnston
Right, but the goal is to make AGI you can point at things, not to make AGI you can point at things using some particular technique. (Tangentially, I also think the jury is still out on whether humans are bad fitness maximizers, and if we're ultimately particularly good at it - e.g. let's say, barring AGI disaster, we'd eventually colonise the galaxy - that probably means AGI alignment is harder, not easier)
2Rob Bensinger
To my eye, this seems like it mostly establishes 'it's not impossible in principle for an optimizer to have a goal that relates to the physical world'. But we had no reason to doubt this in the first place, and it doesn't give us a way to reliably pick in advance which physical things the optimizer cares about. "It's not impossible" is a given for basically everything in AI, in principle, if you have arbitrary amounts of time and arbitrarily deep understanding.
5David Johnston
As I said (a few times!) in the discussion about orthogonality, indifference about the measure of "agents" that have particular properties seems crazy to me. Having an example of "agents" that behave in a particular way is a enormously different to having an unproven claim that such agents might be mathematically possible.

I think this is correct. Shard theory is intended as an account of how inner misalignment produces human values. I also think that human values aren't as complex or weird as they introspectively appear. 

[-]RaemonΩ6162

I read an early draft of this awhile and am glad to have it publicly available.  And I do think the updates in structure/introduction were worth the wait. Thanks!

>There is no pivotal output of an AGI that is humanly checkable and can be used to safely save the world but only after checking it

This is a sort of surprising claim. From an abstract point of view, assuming NP >> P, checking can be way easier than inventing. To stick with your example, it kind of seems, at an intuitive guess, like a plan to use nanobots to melt all GPUs should be very complicated but not way superhumanly complicated? (Superhuman to invent, though.) Like, you show me the plans for the bootstrap nanofactory, the workhorse nanofactory, the standard nanobots, the software for coordinating the nanobots, the software for low-power autonomous behavior, the transportation around the world, the homing in on GPUs, and the melting process. That's really complicated, way more complicated than anything humans have done before, but not by 1000x? Maybe like 100x? Maybe only 10x if you count whole operating systems or scientific fields. Does this seem quantitatively in the right ballpark, and you're saying, that quantitatively large but not crazy amount of checking is infeasible?

4Rob Bensinger
The preceding sentences in the OP were (emphasis added): I took Eliezer to be saying something like: 'If you're confident that your AGI system is directing its optimization at the target task, is doing no adversarial optimization, and is otherwise aligned, then shrug, maybe there's some role to be played by checking a few aspects of the system's output to confirm certain facts. 'But in this scenario, the work is almost entirely being done by the AGI's alignment, not by the post facto checking. If you screwed up and the system is doing open-ended optimization of the world that includes thinking about its developers and planning to take control from them, then it's plausible that your checking will completely fail to notice the trap; and it's ~certain that your checking, if it does notice the trap, won't thereby give you trap-free nanosystems that you can use to end the acute risk period.' (One thing to keep in mind is that an adversarial AGI with knowledge of its operators would want to obfuscate its plans, making it harder for humans to productively analyze designs it produces; and it might also want to obscure the fact that the plans are obfuscated, making them look easier-to-check than they are.)
5TekhneMakre
We can distinguish: -- The AI is trying to deceive you. -- The AI isn't trying to deceive you, but is trying to produce plans that would, if executed, have consequences X, and X is not something you want.  -- The AI is trying to produce plans that would, if executed, have consequences you want. The first case is hopeless, and the third case is about an already aligned AI. The second case might not really make sense, because deception is a convergent instrumental goal especially if the AI is trying to cause X and you're trying to cause not X, and generally because an AI that smart probably has inner optimizers that don't care about this "make a plan, don't execute plans" thing you thought you'd set up. But if, arguendo, we have a superintelligently optimized plan which doesn't already contain, in its current description as a plan, a mindhack (e.g. by some surprising way of domaining an AI to care about producing plans but not about making anything happen), then there's a question whether it could help to have humans think about the consequences of the plan. I thought Eliezer was answering that question "No, even in this hypothetical, pivotal acts are too complicated and can't be understood fully in detail by humans, so you'd still have to trust the AI, so the AI has to have understood and applied a whole lot about your values in order to have any shot that the plan doesn't have huge unpleasantly surprising consequences", and I was questioning that.
3Victor Levoso
Not a response to your actual point but I think that hypothetical example probably doesn't make sense (as in making the ai not "care" doesn't prevent it from including mindhacks in its plan) If you have a plan that is "superingently optimized" for some misaligned goal then that plan will have to take into account the effect of outputing the plan itself and will by default contain deception or mindhacks even if the AI doesn't in some sense "care" about executing plans. (or if you setup some complicated scheme whith conterfactuals so the model ignores the effects of the plans in humans that will make your plans less useful or inscrutable) The plan that produces the most paperclips is going to be one that deceives or mindhacks humans instead of one that humans wouldn't accept in the first place. Maybe it's posible to use some kind of scheme that avoids the model taking the consecueces of ouputing the plan itself into account but the model kind of has to be modeling humans reading its plan to write a realistic plan that humans will understand, accept and be able to put into practice, and the plan might only work in the fake conterfactual universe whith no plan it was written for. So I doubt it's actually feasible to have any such scheme that avoids mindhacks and still produces usefull plans.
4TekhneMakre
I think I agree, but also, people say things like "the AI should if possible be prevented from not modeling humans", which if possible would imply that the hypothetical example makes more sense.
1Leo P.
I believe the second case is a subcase of the problem of ELK. Maybe the AI isn't trying to deceive you, and actually do what you asked it to do (e.g., I want to see "the diamond" on the main detector), yet the plans it produces has consequence X that you don't want (in the ELK example, the diamond is stolen but you see something that looks like the diamond on the main detector). The problem is: how can you be sure the plans proposed have consequence X? Especially if you don't even know X is a possible consequence of the plans?

To point 4 and related ones, OpenAI has this on their charter page:

We are concerned about late-stage AGI development becoming a competitive race without time for adequate safety precautions. Therefore, if a value-aligned, safety-conscious project comes close to building AGI before we do, we commit to stop competing with and start assisting this project. We will work out specifics in case-by-case agreements, but a typical triggering condition might be “a better-than-even chance of success in the next two years.”

What about the possibility of persuading the top several biggest actors (DeepMind, FAIR, etc.) to agree to something like that?  (Note that they define AGI on the page to mean "highly autonomous systems that outperform humans at most economically valuable work".)  It's not very fleshed out, either the conditions that trigger the pledge or how the transition goes, but it's a start.  The hope would be that someone would make something "sufficiently impressive to trigger the pledge" that doesn't quite kill us, and then ideally (a) the top actors stopping would buy us some time and (b) the top actors devoting their people to helping out (I figure they could write test suites at minimum) could accelerate the alignment work.

I see possible problems with this, but is this at least in the realm of "things worth trying"?

9Vaniver
My understanding is that this has been tried, at various levels of strength, ever since OpenAI published its charter. My sense is that's MIRI's idea of "safety-conscious" looks like this, which it guessed was different from OpenAI's sense; I kind of wish that had been a public discussion back in 2018.
4Lone Pine
Given that Sam Altman has some of the shortest timelines around, I wonder if he could be persuaded that DeepMind are within 2 years of the finish line, or will be visibly within 2 years of the finish line in a few years. (Not implying that would be a solution to anything, I'm just curious what it would take for that clause to apply.)

+9. This is a powerful set of arguments pointing out how humanity will literally go extinct soon due to AI development (or have something similarly bad happen to us). A lot of thought and research went into an understanding of the problem that can produce this level of understanding of the problems we face, and I'm extremely glad it was written up.

Is there a plausible pivotal act that doesn't amount to some variant of "cripple human civilization so that it can't make or use computers until it recovers"?

Use AGI to build fast-running high-fidelity human whole-brain emulations. Then run thousands of very-fast-thinking copies of your best thinkers. Seems to me this plausibly makes it realistic to keep tabs on the world's AGI progress, and locally intervene before anything dangerous happens, in a more surgical way rather than via mass property destruction of any sort.

3Evan R. Murphy
This is a much less destructive-sounding pivotal act proposal than "melt all GPUs". I'm trying to figure out why and if it's actually less destructive... Does it sound less destructive because it's just hiding the destructive details behind the decisions of these best thinker simulations? After their, say, 100 subjective years of deliberation do the thinkers just end up with a detailed proposal for how to melt all GPUs"? I think I give our best thinkers more credit than that. I wouldn't presume to know in advance the plan that many copies of our best thinkers would come up with after having a long time to deliberate. But I have confidence or at least a hope that they'd come up with something less destructive and at least as effective as "melt all GPUs". So this pivotal act proposal puts some distance between us and the messier details of the act. But it does it in a reasonable way, not just by hand-waving or forgetting those details, but instead by deferring them to people who we would most trust to handle them well (many of the best thinkers in the world, overclocked!) This is an intriguing proposal and because it certainly sounds so much less destructive and horrifying than "melt all GPUs", I will very much prefer to use and see this used as the go-to theoretical example of a pivotal act until I hear of or think of a better one.
1Kenny
I can't think of any! There are maybe 'plausibly plausible' (or 'possibly plausible') acts that a more 'adequate' global civilization might be able to take. But it seems like that hypothetical adequate civilization would have already performed such a pivotal act and the world would look very different than it does now. It's ('strictly') possible that such a pivotal act has already been performed and that, e.g. the computer hardware currently available isn't sufficient to build an AGI. It just seems like there's VERY little evidence that that's the case.

Wow, 510 karma and counting. This post currently has the 14th most karma all time and most for this year. Makes me think back to this excerpt from Explainers Shoot High. Aim Low!.

A few years ago, an eminent scientist once told me how he'd written an explanation of his field aimed at a much lower technical level than usual. He had thought it would be useful to academics outside the field, or even reporters. This ended up being one of his most popular papers within his field, cited more often than anything else he'd written.

The lesson was not that his fellow scientists were stupid, but that we tend to enormously underestimate the effort required to properly explain things.

I'm confused about A6, from which I get "Yudkowsky is aiming for a pivotal act to prevent the formation of unaligned AGI that's outside the Overton Window and on the order of burning all GPUs". This seems counter to the notion in Q4 of Death with Dignity where Yudkowsky says

It's relatively safe to be around an Eliezer Yudkowsky while the world is ending, because he's not going to do anything extreme and unethical unless it would really actually save the world in real life, and there are no extreme unethical actions that would really actually save the world the way these things play out in real life, and he knows that.  He knows that the next stupid sacrifice-of-ethics proposed won't work to save the world either, actually in real life. 

I would estimate that burning all AGI-capable compute would disrupt every factor of the global economy for years and cause tens of millions of deaths[1], and that's what Yudkowsky considers the more mentionable example. Do the other options outside the Overton Window somehow not qualify as unsafe/extreme unethical actions (by the standards of the audience of Death with Dignity)? Has Yudkowsky changed his mind on what options would actually ... (read more)

Interventions on the order of burning all GPUs in clusters larger than 4 and preventing any new clusters from being made, including the reaction of existing political entities to that event and the many interest groups who would try to shut you down and build new GPU factories or clusters hidden from the means you'd used to burn them, would in fact really actually save the world for an extended period of time and imply a drastically different gameboard offering new hopes and options.

What makes me safe to be around is that I know that various forms of angrily acting out violently would not, in fact, accomplish anything like this.  I would only do something hugely awful that would actually save the world.  No such option will be on the table, and I, the original person who wasn't an idiot optimist, will not overestimate and pretend that something will save the world when it obviously-to-me won't.  So I'm a relatively safe person to be around, because I am not the cartoon supervillain talking about necessary sacrifices to achieve greater goods when everybody in the audience knows that the greater good won't be achieved; I am the person in the audience rolling their eyes at the cartoon supervillain.

4Daphne_W
I suppose 'on the order of' is the operative phrase here, but that specific scenario seems like it would be extremely difficult to specify an AGI for without disastrous side-effects and like it still wouldn't be enough. Other, less efficient or less well developed forms of compute exist, and preventing humans from organizing to find a way around the GPU-burner's blacklist for unaligned AGI research while differentially allowing them to find a way to build friendly AGI seems like it would require a lot of psychological/political finesse on the GPU-burner's part. It's on the level of Ozymandias from Watchmen, but it's cartoonish supervillainy nontheless. I guess my main issue is a matter of trust. You can say the right words, as all the best supervillains do, promising that the appropriate cautions are taken above our clearance level. You've pointed out plenty of mistakes you could be making, and the ease with which one can make mistakes in situations such as yours, but acknowledging potential errors doesn't prevent you from making them. I don't expect you to have many people you would trust with AGI, and I expect that circle would shrink further if those people said they would use the AGI to do awful things iff it would actually save the world [in their best judgment]. I currently have no-one in the second circle. If you've got a better procedure for people to learn to trust you, go ahead, but is there something like an audit you've participated in/would be willing to participate in? Any references regarding your upstanding moral reasoning in high-stakes situations that have been resolved? Checks and balances in case of your hardware being corrupted? You may be the audience member rolling their eyes at the cartoon supervillain, but I want to be the audience member rolling their eyes at HJPEV when he has a conversation with Quirrel where he doesn't realise that Quirrel is evil.
5Vaniver
It definitely is the case that a pivotal act that isn't "disruptive" isn't a pivotal act. But I think not all disruptive acts have a significant cost in human lives. To continue with the 'burn all GPUs' example, note that while some industries are heavily dependent on GPUs, most industries are instead heavily dependent on CPUs. The hospital's power will still be on if all GPUs melt, and probably their monitors will still work (if the nanobots can somehow distinguish between standalone GPUs and ones embedded into motherboards). Transportation networks will probably still function, and so on. Cryptocurrencies, entertainment industries, and lots of AI applications will be significantly impacted, but this seems recoverable. But I do think Eliezer's main claim is: some people will lash out in desperation when cornered ("Well, maybe starting WWIII will help with AI risk!"), and Eliezer is not one of those people. So if he makes a call of the form "disruption that causes 10M deaths", it's because the other option looked actually worse, and so this is 'safer'. [If you're one of the people tied up on the trolley tracks, you want the person at the lever to switch it!]
9Daphne_W
AI can run on CPUs (with a certain inefficiency factor), so only burning all GPUs doesn't seem like it would be sufficient. As for disruptive acts that are less deadly, it would be nice to have some examples but Eliezer says they're too far out of the Overton Window to mention. If what you're saying about Eliezer's claim is accurate, it does seem disingenuous to frame "The only worlds where humanity survives are ones where people like me do something extreme and unethical" as "I won't do anything extreme and unethical [because humanity is doomed anyway]". It makes Eliezer dangerous to be around if he's mistaken, and if you're significantly less pessimistic than he is (if you assign >10^-6 probability to humanity surviving), he's mistaken in most of the worlds where humanity survives. Which are the worlds that matter the most. And yeah, it's nice that Eliezer claims that Eliezer can violate ethical injunctions because he's smart enough, after repeatedly stating that people who violate ethical injunctions because they think they're smart enough are almost always wrong. I don't doubt he'll pick the option that looks actually better to him. It's just that he's only human - he's running on corrupted hardware like the rest of us.
[-]Koen.HoltmanΩ-211-5

Having read the original post and may of the comments made so far, I'll add an epistemological observation that I have not seen others make yet quite so forcefully. From the original post:

Here, from my perspective, are some different true things that could be said, to contradict various false things that various different people seem to believe, about why AGI would be survivable [...]

I want to highlight that many of the different 'true things' on the long numbered list in the OP are in fact purely speculative claims about the probable nature of future AGI technology, a technology nobody has seen yet.

The claimed truth of several of these 'true things' is often backed up by nothing more than Eliezer's best-guess informed-gut-feeling predictions about what future AGI must necessarily be like. These predictions often directly contradict the best-guess informed-gut-feeling predictions of others, as is admirably demonstrated in the 2021 MIRI conversations.

Some of Eliezer's best guesses also directly contradict my own best-guess informed-gut-feeling predictions. I rank the credibility of my own informed guesses far above those of Eliezer.

So overall, based on my own best guesses here, I am much more optimistic about avoiding AGI ruin than Eliezer is. I am also much less dissatisfied about how much progress has been made so far.

7handoflixue
Apologies if there is a clear answer to this, since I don't know your name and you might well be super-famous in the field: Why do you rate yourself "far above" someone who has spent decades working in this field? Appealing to experts like MIRI makes for a strong argument. Appealing to your own guesses instead seems like the sort of thought process that leads to anti-vaxxers.

I think it's a positive if alignment researchers feel like it's an allowed option to trust their own technical intuitions over the technical intuitions of this or that more-senior researcher.

Overly dismissing old-guard researchers is obviously a way the field can fail as well. But the field won't advance much at all if most people don't at least try to build their own models.

Koen also leans more on deference in his comment than I'd like, so I upvoted your 'deferential but in the opposite direction' comment as a corrective, handoflixue. :P But I think it would be a much better comment if it didn't conflate epistemic authority with "fame" (I don't think fame is at all a reliable guide to epistemic ability here), and if it didn't equate "appealing to your own guesses" with "anti-vaxxers".

Alignment is a young field; "anti-vaxxer" is a term you throw at people after vaccines have existed for 200 years, not a term you throw at the very first skeptical researchers arguing about vaccines in 1800. Even if the skeptics are obviously and decisively wrong at an early date (which indeed not-infrequently happens in science!), it's not the right way to establish the culture for those first scientific debates.

Why do you rate yourself "far above" someone who has spent decades working in this field?

Well put, valid question. By the way, did you notice how careful I was in avoiding any direct mention of my own credentials above?

I see that Rob has already written a reply to your comments, making some of the broader points that I could have made too. So I'll cover some other things.

To answer your valid question: If you hover over my LW/AF username, you can see that I self-code as the kind of alignment researcher who is also a card-carrying member of the academic/industrial establishment. In both age and academic credentials. I am in fact a more-senior researcher than Eliezer is. So the epistemology, if you are outside of this field and want to decide which one of us is probably more right, gets rather complicated.

Though we have disagreements, I should also point out some similarities between Eliezer and me.

Like Eliezer, I spend a lot of time reflecting on the problem of crafting tools that other people might use to improve their own ability to think about alignment. Specifically, these are not tools that can be used for the problem of triangulating between self-declared experts. Th... (read more)

4handoflixue
Thanks for taking my question seriously - I am still a bit confused why you would have been so careful to avoid mentioning your credentials up front, though, given that they're fairly relevant to whether I should take your opinion seriously. Also, neat, I had not realized hovering over a username gave so much information!
1Koen.Holtman
You are welcome. I carefully avoided mentioning my credentials as a rhetorical device. This is to highlight the essence of how many of the arguments on this site work.

We need to align the performance of some large task, a 'pivotal act' that prevents other people from building an unaligned AGI that destroys the world. 

 

What is the argument for why it's not worth pursuing a pivotal act without our own AGI? I certainly would not say it was likely that current human actors could pull it off, but if we are in a "dying with more dignity" context anyway, it doesn't seem like the odds are zero.

My idea, which I'll include more as a demonstration of what I mean than a real proposal, would be to develop a "cause area" for influencing military/political institutions as quickly as possible. Yes, I know this sounds too slow and too hard and a mismatch with the community's skills, but consider:

  1. Militaries/governments are "where the money is": they probably do have the coercive power necessary to perform a pivotal act, or at least buy a lot of time. If the PRC is able to completely lock down its giant sophisticated cities, it could probably halt domestic AI research. The West hasn't really tried to do extreme control in a while, for various good reasons, but (just e.g.) the WW2 war economy was awfully tightly managed. We are good at slowing stuff down
... (read more)

Thanks for writing this. I agree with all of these except for #30, since it seems like checking the output of the AI for correctness/safety should be possible even if the AI is smarter than us, just like checking a mathematical proof can be much easier than coming up with the proof in the first place. It would take a lot of competence, and a dedicated team of computer security / program correctness geniuses, but definitely seems within human abilities. (Obviously the AI would have to be below the level of capability where it can just write down an argument that convinces the proof checkers to let it out of the box. This is a sense in which having the AI produce uncommented machine code may actually be safer than letting it write English at us.)

We might summarise this counterargument to #30 as "verification is easier than generation". The idea is that the AI comes up with a plan (+ explanation of how it works etc.) that the human systems could not have generated themselves, but that human systems can understand and check in retrospect.

Counterclaim to "verification is easier than generation" is that any pivotal act will involve plans that human systems cannot predict the effects of just by looking at the plan. What about the explanation, though? I think the problem there may be more that we don't know how to get the AI to produce a helpful and accurate explanation as opposed to a bogus misleading but plausible-sounding one, not that no helpful explanation exists. 

This seems to me like a case of the imaginary hypothetical "weak pivotal act" that nobody can ever produce.  If you have a pivotal act you can do via following some procedure that only the AI was smart enough to generate, yet humans are smart enough to verify and smart enough to not be reliably fooled about, NAME THAT ACTUAL WEAK PIVOTAL ACT.

Okay, I will try to name a strong-but-checkable pivotal act.

(Having a strong-but-checkable pivotal act doesn't necessarily translate into having a weak pivotal act. Checkability allows us to tell the difference between a good plan and a trapped plan with high probability, but the AI has no reason to give us a good plan. It will just produce output like "I have insufficient computing power to solve this problem" regardless of whether that's actually true. If we're unusually successful at convincing the AI our checking process is bad when it's actually good, then that AI may give us a trapped plan, which we can then determine is trapped. Of course, one should not risk executing a trapped plan, even if one thinks one has identified and removed all the traps. So even if #30 is false, we are still default-doomed. (I'm not fully certain that we couldn't create some kind of satisficing AI that gets reward 1 if it generates a safe plan, reward 0 if its output is neither helpful nor dangerous, and reward -1 if it generates a trapped plan that gets caught by our checking process. The AI may then decide that it has a higher chance of success if it just submits a safe plan. But I don't know ho... (read more)

Well, I had to think about this for longer than five seconds, so that's already a huge victory.

If I try to compress your idea down to a few sentences:

The humans ask the AI to produce design tools, rather than designs, such that there's a bunch of human cognition that goes into picking out the particular atomic arrangements or synthesis pathways; and we can piecewise verify that the tool is making accurate predictions; and the tool is powerful enough that we can build molecular nanotech and an uploader by using the tool for an amount of time too short for Facebook to catch up and destroy the world.  The AI that does this is purportedly sufficiently good at meta-engineering to build the tool, but not good enough at larger strategy that it can hack its way through the humans using just the code of the tool.  The way in which this attacks a central difficulty is by making it harder for the AI to just build unhelpful nanotech using the capabilities that humans use to produce helpful nanotech.

Sound about right?

4DaemonicSigil
Yes, sounds right to me. It's also true that one of the big unproven assumptions here is that we could create an AI strong enough to build such a tool, but too weak to hack humans. I find it plausible, personally, but I don't yet have an easy-to-communicate argument for it.
1Ken Kahn
Why can't a narrow AI (maybe like Drexler's proposal) create the tool safely?
1DaemonicSigil
I don't know of a reason we couldn't do this with a narrow AI. I have no idea how, but it's possible in principle so far as I know. If anyone can figure out how, they could plausibly execute the pivotal act described above, which would be a very good thing for humanity's chances of survival. EDIT: Needless to say, but I'll say it anyway: Doing this via narrow AI is vastly preferable to using a general AI. It's both much less risky and means you don't have to expend an insane amount of effort on checking.
1Tor Økland Barstad
I think this is a good way to put things, and it's a concept that can be made more general and built upon. Like, we can also have AIs produce: * Tools that make other tools * Tools that help to verify other tools * Tools that look for problems with other tools (in ways that don't guarantee finding all problems, but can help find many) * Tools that help approximate brain emulations (or get us part of the way there), or predict what a human would say when responding to questions in some restricted domain * Etc, etc Maybe you already have thought through such strategies very extensively, but AFAIK you don't make that clear in any of your writings, and it's not a trivial amount of inferential distance that is required to realize the full power of techniques like these. I have written more about this concept in this post in this series. I'm not sure whether or not any of the concepts/ideas in the series are new, but it seems to me that several of them at the very least are under-discussed.
3[comment deleted]
1Edouard Harris
Interesting. The specific idea you're proposing here may or may not be workable, but it's an intriguing example of a more general strategy that I've previously tried to articulate in another context. The idea is that it may be viable to use an AI to create a "platform" that accelerates human progress in an area of interest to existential safety, as opposed to using an AI to directly solve the problem or perform the action. Essentially: 1. A "platform" for work in domain X is something that removes key constraints that would otherwise have consumed human time and effort when working in X. This allows humans to explore solutions in X they wouldn't have previously — whether because they'd considered and rejected those solution paths, or because they'd subconsciously trained themselves not to look in places where the initial effort barrier was too high. Thus, developing an excellent platform for X allows humans to accelerate progress in domain X relative to other domains, ceteris paribus. (Every successful platform company does this. e.g., Shopify, Amazon, etc., make valuable businesses possible that wouldn't otherwise exist.) 2. For certain carefully selected domains X, a platform for X may plausibly be relatively easier to secure & validate than an agent that's targeted at some specific task x ∈ X would be. (Not easy; easier.) It's less risky to validate the outputs of a platform and leave the really dangerous last-mile stuff to humans, than it would be to give an end-to-end trained AI agent a pivotal command in the real world (i.e., "melt all GPUs") that necessarily takes the whole system far outside its training distribution. Fundamentally, the bet is that if humans are the ones doing the out-of-distribution part of the work, then the output that comes out the other end is less likely to have been adversarially selected against us. (Note that platforms are tools, and tools want to be agents, so a strategy like this is unlikely to arise along the "natural" path
2Tor Økland Barstad
I don't claim to have a solution where every detail is filled in, or where I have watertight arguments showing that it's guaranteed to work (if executed faithfully). But I think I have something, and that it could be built upon. The outlines of a potential solution. And by "solution", I mean a pivotal strategy (consisting of many acts that could be done over a short amount of time), where we can verify output extensively and hopefully (probably?) avoid being fooled/manipulated/tricked/"hacked". I'm writing a series about this here. Only 2 parts finished so far (current plan is to write 4).
2Noosphere89
I must say, you have a very pessimistic/optimistic view of AI would be able to solve P=NP. I won't say you're completely wrong, as there's always a chance that P does equal NP. But I would be very careful of predicting anything based on the possibility of P=NP.

I think P?=NP is a distraction. Like, it's not very useful to ask the question of whether Lee Sedol played a 'polynomial' number of games of Go, and AlphaGo played a 'nonpolynomial' number of games of Go. AlphaGo played more games and had a more careful and precise memory, and developed better intuitions, and could scale to more hardware better.

So what should I do with this information, like what other option than "nod along and go on living their lives" is there for me?

They probably do not know where the real difficulties are, they probably do not understand what needs to be done, they cannot tell the difference between good and bad work, and the funders also can't tell without me standing over their shoulders evaluating everything, which I do not have the physical stamina to do.

This was the sentiment I got after applying to the LTFF with an idea. Admittedly, I couldn't really say whether my idea had been tried before, or wasn't obviously bad, but my conversation basically boiled down to whether I wanted to use this project as a way to grow myself in the field, rather than any particular merits/faults of the idea itself. My motivation was really about trying a cool idea that I genuinely believed could practically improve AI safety if successful, while ethically I couldn't commit to wanting to stay in the field even if it (likely?) failed since I like to go wherever my ideas take me.

Since it may be a while before I personally ever try out the idea, the most productive thing I can do seems to be to share it. It's essentially an attempt at a learning algorithm which 'forces' a models weights to explain the reasoning/motivations behind its actio... (read more)

This was the sentiment I got after applying to the LTFF with an idea. Admittedly, I couldn't really say whether my idea had been tried before, or wasn't obviously bad, but my conversation basically boiled down to whether I wanted to use this project as a way to grow myself in the field, rather than any particular merits/faults of the idea itself

I evaluated this application (and we chatted briefly in a video call)! I am not like super confident in my ability to tell whether an idea is going to work, but my specific thoughts on your proposals were that I think it was very unlikely to work, but that if someone was working on it, they might learn useful things that could make them a better long-term contributor to the AI Alignment field, which is why my crux for your grant was whether you intended to stay involved in the field long-term.

6Zvi
Appreciation for sharing the reasoning. Disagreement with the reasoning.  eeegnu is saying they go where their ideas take them and expressing ethical qualms, which both seem like excellent reasons to want someone considering AI safety work rather than reasons to drive them away from AI safety work. Their decision to continue doing AI safety work seems likely to be correlated with whether they could be productive by doing additional AI safety work - if their ideas take them elsewhere it is unlikely anything would have come of them staying.  This is especially true if one subscribes to the theory that we are worried about sign mistakes rather than 'wasting' funding - if we are funding unproven individuals in AI Safety and think that is good, then this is unusually 'safe' in the sense of it being more non-negative.  So to the extent that I was running the LTFF, I would have said yes.  
9habryka
I don't think the policy of "I will fund people to do work that I don't expect to be useful" is a good one, unless there is some positive externality. It seems to me that your comment is also saying that the positive externality you are looking for is "this will make this person more productive in helping with AI Safety", or maybe "this will make them more likely to work on AI Safety". But you are separately saying that I shouldn't take their self-reported prediction that they will not continue working in AI Safety, independently of the outcome of the experiment, at face value, and instead bet that by working on this, they will change their mind, which seems weird to me. Separately, I think there are bad cultural effects of having people work on projects that seem very unlikely to work, especially if the people working on them are self-reportedly not doing so with a long-term safety motivation, but because they found the specific idea they had appealing (or wanted to play around with technologies in the space). I think this will predictably attract a large number of grifters and generally make the field a much worse place to be.
4DirectedEvolution
“I don't think the policy of "I will fund people to do work that I don't expect to be useful" is a good one, unless there is some positive externality.” By this, do you mean you think it’s not good to fund work that you expect to be useful with < 50% probability, even if the downside risk is zero? Or do you mean you don’t expect it’s useful to fund work you strongly expect to have no positive value when you also expect it to have a significant risk of causing harm?
6habryka
50% is definitely not my cutoff, and I don't have any probability cutoff. More something in the expected value space. Like, if you have an idea that could be really great but only has a 1% chance of working, that still feels definitely worth funding. But if you have an idea that seems like it only improves things a bit, and has a 10% chance of working, that doesn't feel worth it.
6TekhneMakre
Upvoted for sharing information about thoughts behind grant-making. I could see reasons in some cases to not do this, but by and large more information seems better for many reasons.
5Eliezer Yudkowsky
(I wasn't able to understand the idea off this description of it.)
4TekhneMakre
Why wouldn't the explainer just copy the latent vector, and the explainee just learn to do the task in the same way the original model does it? Or more generally, why does this put any pressure towards "explaining the reasons/motives" behind the original model's actions? I think you're thinking that by using a pre-trained GPT3-alike as the explainer model, you start off with something a lot more language-y, and language-y concepts are there for easy pickings for the training process to find in order to "communicate" between the original model and the explainee model. This seems not totally crazy, but  1. it seems to buy you, not anything like further explanations of reasons/motives beyond what's "already in" the original model, but rather at most a translation into the explainer's initial pre-trained internal language; 2. the explainer's initial idiolect stays unexplained / unmotivated; 3. the training procedure doesn't put pressure towards explanation, and does put pressure towards copying.  
1eeegnu
These are great points, and ones which I did actually think about when I was brainstorming this idea (if I understand them correctly.) I intend to write out a more thorough post on this tomorrow with clear examples (I originally imagined this as extracting deeper insights into chess), but to answer these: 1. I did think about these as translators for the actions of models into natural language, though I don't get the point about extracting things beyond what's in the original model. 2. I mostly glossed over this part in the brief summary, and the motivation I had for it comes from how (unexpectedly?) it works for GAN's to just start with random noise, and in the process the generator and discriminator both still improve each other. 3. My thoughts here were for the explainer models update error vector to come from judging the learner model on new unseen tasks without the explanation (i.e. how similar are they to the original models outputs.) In this way the explainer gets little benefit from just giving the answer directly, since the learner will be tested without it, but if the explanation in any way helps the learner learn, it'll improve its performance more (this is basically what the entire idea hinges on.)
2TekhneMakre
(I didn't understand this on one read, so I'll wait for the post to see if I have further comments. I didn't understand the analogy / extrapolation drawn in 2., and I didn't understand what scheme is happening in 3.; maybe being a little more precise and explicit about the setup would help.)

A lot of important warnings in this post. "Capabilities generalize further than alignment once capabilities start to generalize far" was novel to me and seems very important if true.

I don't really understand the emphasis on "pivotal acts", though; there seems to be tons of weak pivotal acts, e.g.  ways in which narrow AI or barely-above-human-AGI could help coordinate a global emergency regulatory response by the AI superpowers. Still might be worth focusing our effort on the future worlds where no weak pivotal acts are available, but important to point out this is not the median world.

I could coordinate world superpowers if they wanted to coordinate and were willing to do that.  It's not an intelligence problem, unless the solution is mind-control, and then that's not a weak pivotal act, it's an AGI powerful enough to kill you if misaligned.

5Ivan Vendrov
Mind control is too extreme; I think world superpowers could be coordinated with levels of persuasion greater than one Eliezer but short of mind control. E.g. people are already building narrow persuasion AI capable of generating arguments that are highly persuasive for specific people. A substantially-superhuman but still narrow version of such an AI will very likely be built in the next 5 years, and could be used in a variety of weak pivotal acts (not even in a manipulative way! even a public demonstration of such an AI would make a strong case for coordination, comparable to various weapons treaties).

I largely agree with all these points, with my minor points of disagreement being insufficient to change the overall conclusions. I feel like an important point which should be emphasized more is that our best hope for saving humanity lies in maximizing the non-linearly-intelligence-weighted researcher hours invested in AGI safety research before the advent of the first dangerously powerful unaligned AGI. To maximize this key metric, we need to get more and smarter people doing this research, and we need to slow down AGI capabilities research. Insofar as AI Governance is a tactic worth pursuing, it must pursue one or both of these specific aims. Once dangerously powerful unaligned AGI has been launched, it's too late for politics or social movements or anything slower than perhaps decisive military action prepped ahead of time (e.g. the secret AGI-prevention department hitting the detonation switch for all the secret prepared explosives in all the worlds' data centers).

[-]VaniverΩ6105

I'm very glad this list is finally published; I think it's pretty great at covering the space (tho I won't be surprised if we discover a few more points), and making it so that plans can say "yeah, we're targeting a hole we see in number X."

[In particular, I think most of my current hope is targeted at 5 and 6, specifically that we need an AI to do a pivotal act at all; it seems to me like we might be able to transition from this world to a world sophisticated enough to survive on human power. But this is, uh, a pretty remote possibility and I was much happier when I was optimistic about technical alignment.]

For future John who is using the searchbox to try to find this post: this is Eliezer's List O' Doom.

4Raemon
Are you actually gonna remember the apostrophe?
5johnswentworth
I just tested that, and it works both ways.

RE 19: Maybe rephrase "kill everyone in the world using nanotech to strike before they know they're in a battle, and have control of your reward input forever after"? This could, and I predict would, be misinterpreted as "the AI is going to kill everyone and access its own hardware to set its reward to infinity". This is a misinterpetation because you are referring to control of the "reward input" here, and your later sentences don't make sense according to this interpretation. However, given the bolded sentence and some lack of attention, plus some confusions over wire heading that are apparently fairly common, I expect a fair number of misinterpretations.

"Geniuses" with nice legible accomplishments in fields with tight feedback loops where it's easy to determine which results are good or bad right away, and so validate that this person is a genius, are (a) people who might not be able to do equally great work away from tight feedback loops, (b) people who chose a field where their genius would be nicely legible even if that maybe wasn't the place where humanity most needed a genius, and (c) probably don't have the mysterious gears simply because they're rare.  You cannot just pay $5 million apiece to a bunch of legible geniuses from other fields and expect to get great alignment work out of them.  They probably do not know where the real difficulties are, they probably do not understand what needs to be done, they cannot tell the difference between good and bad work, and the funders also can't tell without me standing over their shoulders evaluating everything, which I do not have the physical stamina to do.  I concede that real high-powered talents, especially if they're still in their 20s, genuinely interested, and have done their reading, are people who, yeah, fine, have higher probabilities of making co

... (read more)
[-]lc256

Most of the impressive computer security subdisciplines have very tight feedback loops and extreme legibility; that's what makes them impressive. When I think of the hardest security jobs, I think of 0-day writers, red-teamers, etc., who might have whatever Eliezer describes as security mindset but are also described extremely well by him in #40. There are people that do a really good job of protecting large companies, but they're rare, and their accomplishments are highly illegible except to a select group of guys at e.g. SpecterOps. I don't think MIRI would be able to pick them out, which is of course not their fault.

I'd say something more like hedge fund management, but unfortunately those guys tend to be paid pretty well...

3Ruby
I think the intended field lacking tight feedback loops is AI alignment.
4David Udell
(I meant: What fields can we draw legible geniuses from, into alignment.)
5Kenny
I think people have floated the idea of recruiting 'math geniuses' specifically and EY is claiming that, even if they could be recruited and were recruited, we couldn't (reasonably) "expect to get great alignment work out of them".
[-]RaemonΩ7913

Curated. As previously noted, I'm quite glad to have this list of reasons written up. I like Robby's comment here which notes:

The point is not 'humanity needs to write a convincing-sounding essay for the thesis Safe AI Is Hard, so we can convince people'. The point is 'humanity needs to actually have a full and detailed understanding of the problem so we can do the engineering work of solving it'.

I look forward to other alignment thinkers writing up either their explicit disagreements with this list, or things that the list misses, or their own frame on th... (read more)

23.  Corrigibility is anti-natural to consequentialist reasoning; "you can't bring the coffee if you're dead" for almost every kind of coffee.  We (MIRI) tried and failed to find a coherent formula for an agent that would let itself be shut down (without that agent actively trying to get shut down).  Furthermore, many anti-corrigible lines of reasoning like this may only first appear at high levels of intelligence.

 

There is one approach to corrigibility that I don't see mentioned in the "tried and failed" post Eliezer linked to her... (read more)

Well, the obvious #1 question:  A myopic AGI is a weaker one, so what is the weaker pivotal act you mean to perform with this putative weaker AGI?  A strange thing to omit from one's discussion of machinery - the task that the machinery is to perform.

4TekhneMakre
Myopia seems to me like a confused concept because the structure of the world is non-myopic, so to speak. If you myopically try to deal with rocks, you'll for myopic reasons model a rock as a permanent objects with a particular shape. But the rock also behaves as a permanent object over much longer time scales than your myopia. So you've in some important sense accessed time-scales much longer than your myopia. I think this happens at any level of a mind. If so, then minds with myopic goals are very similar to minds with non-myopic goals; so similar that they may be basically the same because they'll have non-myopic strategic components that exert their own non-myopic agency.
7Evan R. Murphy
Here's a related comment thread debating myopia. This one includes you (TekhneMakre), evhub, Eliezer and others. I'm reading it now to see if there are any cruxes that could help in our present discussion: https://www.lesswrong.com/posts/5ciYedyQDDqAcrDLr/a-positive-case-for-how-we-might-succeed-at-prosaic-ai?commentId=st5tfgpwnhJrkHaWp
2TekhneMakre
[Upvoted for looking over past stuff.] On reflection I'm not being that clear in this present thread, and am open to you making a considered counterargument / explanation and then me thinking that over for a longer amount of time to try writing a clearer response / change my mind / etc.
5Alexander Gietelink Oldenziel
I suppose the point is that a myopic agent will accept/know that a rock will exist for long time-scales it just won't care. Plenty of smart but short-sighted people so not inconceivable.
2TekhneMakre
I'm saying that, for the same reason that myopic agents think about rocks the same way non-myopic agents think about rocks, also myopic agents will care about long-term stuff the same way non-myopic agents do. The thinking needed to make cool stuff happen generalizes like the thinking needed to deal with rocks. So yeah, you can say "myopic agents by definition don't care about long-term stuff", but if by care you mean the thing that actually matters, the thing about causing stuff to happen, then you've swept basically the entire problem under the rug.
2Alexander Gietelink Oldenziel
Why can myopic agents not think about long-term stuff the same way as non-myopic agents but still not care about long-term stuff?
2TekhneMakre
They *could*, but we don't know how to separate caring from thinking, modeling, having effects; and the first 1000 programs that think about long term stuff that you find just by looking for programs that think about long term stuff, also care about long term stuff.
1Evan R. Murphy
What you're saying seems to contradict the orthogonality thesis. Intelligence level and goals are independent, or at least not tightly interdependent. Let's use the common example of a paperclip maximizer. Maximizing total long-term paperclips is a strange goal for an agent to have, but most people in AI alignment think it's possible that an AI could be trained to optimize for something like this like this could in principle emerge from training (though we don't know how to reliably train one on purpose). Now why couldn't an agent by motivated to maximize short-term paperclips? It wants more paperclips, but it will always take 1 paperclip now over 1 or even 10 or 100 a minute in the future. It wants paperclips ASAP. This is one contrived example of what a myopic AI might look like - a myopic paperclip maximizer.

I don't think we could train an AI to optimize for long-term paperclips.  Maybe I'm not "most people in AI alignment" but still, just saying.

1Evan R. Murphy
I was trying to contrast the myopic paperclip maximizer idea with the classic paperclip maximizer. Perhaps "long-term" was a lousy choice of words. What would be better: simple paperclip maximizer, unconditional paperclip maximizer, or something? Update: On second thought, maybe what you were getting at is that it's not clear how to deliberately train a paperclip maximizer in the current paradigm. If you tried, you'd likely end up with a mesa-optimizer on some unpredictable proxy objective, like a deceptively aligned steel maximizer.
3TekhneMakre
Yes, I'm saying that AIs are very likely to have (in a broad sense, including e.g. having subagents that have) long-term goals. It *could*, but I'm saying that making an AI like that isn't like choosing a loss function for training, because long-term thinking is convergent. Your original comment said: This is what I'm arguing against. I'm saying it's very unnatural. *Possible*, but very unnatural. And: This sounds like you're saying that myopia *makes* there not be convergent instrumental goals. I'm saying myopia basically *implies* there not being convergent instrumental goals, and therefore is at least as hard as making there not be CIGs.
2Rob Bensinger
I don't think we have any idea how to do this. If we knew how to get an AGI system to reliably maximize the number of paperclips in the universe, that might be most of the (strawberry-grade) alignment problem solved right there.
1Evan R. Murphy
You're right, my mistake - of course we don't know how to deliberately and reliably train a paperclip maximizer. I've updated the parent comment now to say:
2Jeff Rose
It feels like you are setting a discount rate higher than reality demands.  A rationally intelligent agent should wind up with a discount rate that matches reality (e.g. in this case, probably the rate at which paper clips decay or the global real rate of interest).   

Great post. Many of these arguments are fairly convincing.

[-]lc91

4.  We can't just "decide not to build AGI" because GPUs are everywhere, and knowledge of algorithms is constantly being improved and published; 2 years after the leading actor has the capability to destroy the world, 5 other actors will have the capability to destroy the world.  The given lethal challenge is to solve within a time limit, driven by the dynamic in which, over time, increasingly weak actors with a smaller and smaller fraction of total computing power, become able to build AGI and destroy the world.  Powerful actors all refrain

... (read more)
3ArthurB
In addition 1. There aren't that many actors in the lead. 2. Simple but key insights in AI (e.g doing backprop, using sensible weight initialisation) have been missed for decades. If the right tail for the time to AGI by a single group can be long and there aren't that many groups, convincing one group to slow down / paying more attention to safety can have big effects. How big of an effect? Years doesn't seem off the table. Eliezer suggests 6 months dismissively. But add a couple years here and a couple years there, and pretty soon you're talking about the possibility of real progress. It's obviously of little use if no research towards alignment is attempted in that period of course, but it's not nothing.
8lc
It's of use at least inasmuch as it increases my life expectancy.

The first thing generally, or CEV specifically, is unworkable because the complexity of what needs to be aligned or meta-aligned for our Real Actual Values is far out of reach for our FIRST TRY at AGI.  Yes I mean specifically that the dataset, meta-learning algorithm, and what needs to be learned, is far out of reach for our first try.  It's not just non-hand-codable, it is unteachable on-the-first-try because the thing you are trying to teach is too weird and complicated.


Why is CEV so difficult? And if CEV is impossible... (read more)

2No77e
I'll give myself a provisional answer. I'm not sure if it satisfies me, but it's enough to make me pause: Anything short of CEV might leave open an unacceptably high chance of fates worse than death.
1Kenny
CEV is difficult because our values seem to be very complex. Building an AGI (let alone a super-intelligent AGI) that aimed for an 'easier utopia' would have to somehow convince/persuade/align the AI to give up a LOT of value. I don't think it's possible without solving alignment anyways. Essentially, it seems like we'd be trying to 'convince' the AGI to 'not go to fast because that might be bad'. The problem is that we don't know how to precisely what "bad" is anyways. That's very much not obvious. I don't think that, e.g. humans from even 100 years ago teleported to today would be able to reliably distinguish the current world from a 'dystopia'. I haven't myself noticed much agreement about the various utopias people have already described! That seems like pretty strong evidence that 'utopia' is in fact very hard to specify.

There are IMO in-distribution ways of successfully destroying much of the computing overhang. It's not easy by any means, but on a scale where "the Mossad pulling off Stuxnet" is 0 and "build self replicating nanobots" is 10, I think it's is closer to a 1.5.

I mostly agree with the reasoning here; thank you to Eliezer for posting it and explaining it clearly. It's good to have all these reasons here in once place.

The one area I partly disagree with is Section B.1. As I understand it, the main point of B.1 is that we can't guard against all of the problems that will crop up as AI grows more intelligent, because we can't foresee all of those problems, because most of them will be "out-of-distribution," i.e., not the kinds of problems where we have reasonable training data. A superintelligent AI will do strange t... (read more)

If natural selection had feelings, it might not be maximally happy with the way humans are behaving in the wake of Cro-Magnon optimization...but it probably wouldn't call it a disaster, either.

Out of a population of 8 billion humans, in a world that has known about Darwin for generations, very nearly zero are trying to directly manufacture large numbers of copies of their genomes -- there is almost no creative generalization towards 'make more copies of my genome' as a goal in its own right.

Meanwhile, there is some creativity going into the proxy goal 'have more babies', and even more creativity going into the second-order proxy goal 'have more sex'. But the net effect is that the world is becoming wealthier, and the wealthiest places are reliably choosing static or declining population sizes.

And if you wind the clock forward, you likely see humans transitioning into brain emulations (and then self-modifying a bunch), leaving DNA self-replicators behind entirely. (Or you see humans replacing themselves with AGIs. But it would be question-begging to cite this particular prediction here, though it is yet another way humans are catastrophically straying from what human natural selection 'wanted'.)

6Mass_Driver
Right, I'm not claiming that AGI will do anything like straightforwardly maximize human utility. I'm claiming that if we work hard enough at teaching it to avoid disaster, it has a significant chance of avoiding disaster. The fact that nobody is artificially mass-producing their genes is not a disaster from Darwin's point of view; Darwin is vaguely satisfied that instead of a million humans there are now 7 billion humans. If the population stabilizes at 11 billion, that is also not a Darwinian disaster. If the population spreads across the galaxy, mostly in the form of emulations and AIs, but with even 0.001% of sentient beings maintaining some human DNA as a pet or a bit of nostalgia, that's still way more copies of our DNA than the Neanderthals were ever going to get. There are probably some really convincing analogies or intuition pumps somewhere that show that values are likely to be obliterated after a jump in intelligence, but I really don't think evolution/contraception is one of those analogies.
3Rob Bensinger
As stated, I think Eliezer and I, and nearly everyone else, would agree with this. ?? Why would human natural selection be satisfied with 7 billion but not satisfied with a million? Seems like you could equally say 'natural selection is satisfied with a million, since at least a million is higher than a thousand'. Or 'natural selection is satisfied with a hundred, since at least a hundred is higher than fifty'. I understand the idea of extracting from a population's process of natural selection a pseudo-goal, 'maximize inclusive genetic fitness'; I don't understand the idea of adding that natural selection has some threshold where it 'feels' 'satisfied'.
7Mass_Driver
Sure, the metaphor is strained because natural selection doesn't have feelings, so it's never going to feel satisfied, because it's never going to feel anything. For whatever it's worth, I didn't pick that metaphor; Eliezer mentions contraception in his original post. As I understand it, the point of bringing up contraception is to show that when you move from one level of intelligence to another, much higher level of intelligence, then the more intelligent agent can wind up optimizing for values that would be anathema to the less intelligent agents, even if the less intelligent agents have done everything they can to pass along their values. My objection to this illustration is that I don't think anyone's demonstrated that human goals could plausibly be described as "anathema" to natural selection. Overall, humans are pursuing a set of goals that are relatively well-aligned with natural selection's pseudo-goals.
3interstice
Why do you think the goal of evolution is "more copies of genome" rather than "more babies"? To the extent that evolution can be said to have a goal, I think "more babies" is closer -- e.g. imagine a mutation that caused uncontrolled DNA replication within a cell. That would lead to lots of copies of its genome but not more reproductive fitness (Really, I guess this means that you need to specify which evolution you're talking about -- I think the evolution for healthy adult humans has "babies who grow to adulthood" as its goal) w.r.t. declining population sizes, I think it's likely we would return to malthusianism after a few more generations of genetic/cultural selection under modern conditions. Although as you say the singularity is going to come before that can happen.
2Chris_Leong
Yeah, but the population is still pretty large and could become much larger if we become intergalactic. And possibly this is more likely than if we were at the Malthusian limits.
5Chris_Leong
I had the exact same thought. My guess would be that Eliezer might say that since the AI is maximising if the generalisation function misses even one action of this sort as something that we should exclude that we're screwed.
2Mass_Driver
Sure, I agree! If we miss even one such action, we're screwed. My point is that if people put enough skill and effort into trying to catch all such actions, then there is a significant chance that they'll catch literally all the actions that are (1) world-ending and that (2) the AI actually wants to try. There's also a significant chance we won't, which is quite bad and very alarming, hence people should work on AI safety.
3Chris_Leong
Hmm... It seems much, much harder to catch every single one than to catch 99%.
5Mass_Driver
One of my assumptions is that it's possible to design a "satisficing" engine -- an algorithm that generates candidate proposals for a fixed number of cycles, and then, assuming at least one proposal with estimated utility greater than X has been generated within that amount of time, selects one of the qualifying proposals at random. If there are no qualifying candidates, the AI takes no action. If you have a straightforward optimizer that always returns the action with the highest expected utility, then, yeah, you only have to miss one "cheat" that improves "official" utility at the expense of murdering everyone everywhere and then we all die. But if you have a satisficer, then as long as some of the qualifying plans don't kill everyone, there's a reasonable chance that the AI will pick one of those plans. Even if you forget to explicitly penalize one of the pathways to disaster, there's no special reason why that one pathway would show up in a large majority of the AI's candidate plans.
7TurnTrout
There is a special reason, and it's called "instrumental convergence." Satisficers tend to seek power.
2Mass_Driver
I suspect we're talking about two different things.  If you just naively program a super-intelligent AI to satisfice a goal, then, sure, most of the candidate pathways to satisfice will involve accruing a lot of some type of power, because power is useful for achieving goals. That's a valid point, and it's important to understand that merely switching from optimizers to satisficers won't adequately protect us against overly ambitious AIs. However, that doesn't mean that it's futile to explicitly penalize most (but not literally all) of the paths that the AI could take to accumulate too much power. Suppose you adequately penalize all of the tactics that would have catastrophic side effects except for, oh, I don't know, cornering the world market on all types of grain, because you forgot to think of that one particular path to power. Would most of the candidate plans that the AI submits for human approval then turn out to involve secretly cornering the grain market? I don't see why they would. All else being equal, sure, the most powerful tactic available is going to be somewhat more attractive than other, less powerful tactics. But how much more attractive? Would an AI generate 1,000x more plans that involve one specific globally disruptive tactic like cornering the grain market vs. all of the various tactics that are only locally disruptive, like publishing propaganda or hiring away competing researchers or hacking into competing research labs or interrupting the flow of electricity to those labs? 10x more such plans? 2x more such plans? I don't think that's the kind of estimate you can make just by gesturing at basic principles of game theory; you'd need some concrete domain knowledge about the AI's specific planning algorithms. If the truly dangerous plans are only 10x more common on the initial brainstorming list, then we can probably make sure those plans aren't chosen by filtering for plans that the AI rates as safe and transparent. We can use 3 different A
2Chris_Leong
You mean quantilization? Oh yeah, I forgot about that. Good point.

[small nitpick] 

I figured this stuff out using the null string as input, and frankly, I have a hard time myself feeling hopeful about getting real alignment work out of somebody who previously sat around waiting for somebody else to input a persuasive argument into them.  This ability to "notice lethal difficulties without Eliezer Yudkowsky arguing you into noticing them" currently is an opaque piece of cognitive machinery to me, I do not know how to train it into others.  It probably relates to 'security mindset', and a mental motion w

... (read more)

Null string socially.  I obviously was allowed to look at the external world to form these conclusions, which is not the same as needing somebody to nag me into doing so.

9Garrett Baker
This makes more sense. I think you should clarify that this is what you mean when talking about the null string analogy in the future, especially when talking about what thinking about hard-to-think-about topics should look like. It seems fine, and probably useful, as long as you know it's a vast overstatement, but because it's a vast overstatement, it doesn't actually provide that much actionable advice.  Concretely, instead of talking about the null string, it would be more helpful if you talked about the amount of discussion it should take a prospective researcher to reach correct conclusions. From literal null-string for the optimal agent, to vague pointing in the correct direction for a pretty good researcher, to a fully formal and certain proof listing every claim and counter-claim imaginable for someone who probably shouldn't go into alignment.
6Rob Bensinger
If you read the linked tweet (https://twitter.com/ESYudkowsky/status/1500863629490544645), it's talking about the persuasion/convincing/pushing you need in addition to whatever raw data makes it possible to reach the conclusion; it's not saying that humans can get by without any Bayesian evidence about the external world.
4Garrett Baker
I did read the linked tweet, and now that you bring it up, my third sentence doesn't apply. But I think my first & second sentences do still apply (ignoring Eliezer's recent clarification).

Eliezer cross-posted this to the Effective Altruism Forum where there are a few more comments: (In case 600+ comments wasn't enough for anyone!)

https://forum.effectivealtruism.org/posts/zzFbZyGP6iz8jLe9n/agi-ruin-a-list-of-lethalities

Imagine we're all in a paddleboat paddling towards a waterfall. Inside the paddleboat is everyone but only a relatively small number of them are doing the paddling. Of those paddling, most are aware of the waterfall ahead but for reasons beyond my comprehension, decide to paddle on anyway. A smaller group of paddlers have realised their predicament and have decided to stop paddling and start building wings onto the paddleboat so that when the paddleboat inevitably hurtles off the waterfall, it might fly.

It seems to me like the most sensible course of actio... (read more)

5JBlack
If eliminating the risk takes 80+ years and AI development is paused for that to complete, then it is very likely that everyone currently reading this comment will die before it is finished. From a purely selfish point of view it can easily make sense for a researcher to continue even if they fully believe that there is a 90%+ chance that AI will kill them. Waiting will also almost certainly kill them, and they won't get any of those infinite rewards anyway. Being less than 90% convinced that AI will kill them just makes it even more attractive. Hyperbolic discounting makes it even more attractive still.

It's not obvious to me that it takes 80+ years to get double-digit alignment success probabilities, from where we are. Waiting a few decades strikes me as obviously smart from a selfish perspective; e.g., AGI in 2052 is a lot selfishly better than AGI in 2032, if you're under age 50 today.

But also, I think the current state of humanity's alignment knowledge is very bad. I think your odds of surviving into the far future are a lot higher if you die in a few decades and get cryopreserved and then need to hope AGI works out in 80+ years, than if you survive to see AGI in the next 20 years.

1JBlack
True, you can make use of the Gompertz curve to get marginal benefit from waiting a bit while you still have a low marginal probability of non-AGI death. So we only need to worry about researchers who have lower estimates of unaligned AGI causing their death, or who think that AGI is a long way out and want to hurry it up now.
0Noosphere89
Unfortunately, cryopreservation isn't nearly as reliable as needed in order to assume immortality is achieved. While we've gotten better at it, it still relies on toxic chemicals in order to vitrify the brain.
6Rob Bensinger
I'm not saying it's reliable!! I'm saying the odds of alignment success in the next 20 years currently looks even worse.
2Cédric
Well then let's use hyperbolic discounting to our advantage. If we make paddling sufficiently taboo, the social punishment of paddling will outweigh the rewards of potentially building AGI in the minds of the selfish researchers.
2lc
Dunno what that last sentence was but generally I agree.  At the same time: be the change you wish to see in the world. Don't just tell people who are already working on it they should be doing something else. Actually do that raising the alarm thing first.
5Cédric
What I'm doing is trying to help with the wings by throwing some money at MIRI. I am also helping with the stopping/slowing of paddling by sharing my very simple reasoning about why that's the most sensible course of action. Hopefully the simple idea will spread and have some influence. To be honest, I am not willing to invest that much into this as I have other things I am working on (sounds so insane to type that I am not willing to invest much into preventing the doom of everyone and everything). Anyway, there are many like me who are willing to help but only if the cost is low so if you have any ideas of what people like me could do to shift the probabilities a bit, let me know.
1Kenny
Sadly, it doesn't seem like there's any low-hanging fruit that would even "shift the probabilities a bit". Most people seem, if anything, anti-receptive to any arguments about this, because, e.g. it's 'too weird'. And I too feel like this describes myself: I'm thinking – very tentatively (sadly) – about maybe looking into my own personal options for some way to help, but I'm also distracted by "other things". I find this – even people that are (at least somewhat) convinced still not being willing to basically 'throw everything else away' (up to the limit of what would impair our abilities to actually help, if not succeed) to be particularly strong evidence that this might be overall effectively impossible.

On point 35, "Any system of sufficiently intelligent agents can probably behave as a single agent, even if you imagine you're playing them against each other": 

This claim is somewhat surprising to me given that you're expecting powerful ML systems to remain very hard to interpret to humans.

I guess the assumption is that superintelligent ML models/systems may not remain uninterpretable to each other, especially not with the strong incentivize to advance interpretability in specific domains/contexts (benefits from cooperation or from making early commit... (read more)

1Kenny
I think this mostly covers the relevant intuitions: It's the kind of 'obvious' strategy that I think sufficiently 'smart' people would use already.

This look like a great list of risk factors leading to AI lethalities, why making AI safe is a hard problem and why we are failing. But this post is also not what I would have expected by taking the title at face value. I thought that the post would be about detailed and credible scenarios suggesting how AI could lead to extinction, where for example each scenario could represent a class of AI X-risks that we want to reduce. I suspect that such an article would also be really helpful because we probably have not been so good at generating very detailed and... (read more)

New-to-me thought I had in response to the kill all humans part. When predators are a threat to you, you of course shoot them. But once you invent cheap tech that can control them you don't need to kill them anymore. The story goes that the AI would kill us either because we are a threat or because we are irrelevant. It seems to me that (and this imports a bunch of extra stuff that would require analysis to turn this into a serious analysis, this is just an idle thought), the first thing I do if I am superintelligent and wanting to secure my position is no... (read more)

If humans were able to make one super-powerful AI, then humans would probably be able to make a second super-powerful AI, with different goals, which would then compete with the first AI. Unless, of course, the humans are somehow prevented from making more AIs, e.g. because they're all dead.

6romeostevensit
I guess the threat model relies on the overhang. If you need x compute for powerful ai, then you need to control more than all the compute on earth minus x to ensure safety, or something like that. Controlling the people probably much easier.
2Rob Bensinger
Yes, where killing all humans is an example of "controlling the people", from the perspective of an Unfriendly AI.
4Rob Bensinger
A paperclipper mainly cares about humans because we might have some way to threaten the paperclipper (e.g., by pushing a button that deploys a rival superintelligence); and secondarily, we're made of atoms that can be used to build paperclips. It's harder to monitor the actions of every single human on Earth, than it is to kill all humans; and there's a risk that monitoring people visibly will cause someone to push the 'deploy a rival superintelligence' button, if such a button exists. Also, every minute that passes without you killing all humans, in the time window between 'I'm confident I can kill all humans' and 'I'm carefully surveilling every human on Earth and know that there's no secret bunker where someone has a Deploy Superintelligence button', is a minute where you're risking somebody pushing the 'deploy a rival superintelligence' button. This makes me think that the value of delaying 'killing all humans' (once you're confident you can do it) would need to be very high in order to offset that risk. One reason I might be wrong is if the AGI is worried about something like a dead man's switch that deploys a rival superintelligence iff some human isn't alive and regularly performing some action. (Not necessarily a likely scenario on priors, but once you're confident enough in your base plan, unlikely scenarios can end up dominating the remaining scenarios where you lose.) Then it's at least possible that you'd want to delay long enough to confirm that no such switch exists. You should be able to do both in parallel. I don't have a strong view on which is higher-priority. Given the dead-man's-switch worry above, you might want to prioritize sending a probe off-planet first as a precaution; but then go ahead and kill humans ASAP.
3romeostevensit
This is exactly what I was thinking about though, this idea of monitoring every human on earth seems like a failure of imagination on our part. I'm not safe from predators because I monitor the location of every predator on earth. I admit that many (overwhelming majority probably) of scenarios in this vein are probably pretty bad and involve things like putting only a few humans on ice while getting rid of the rest.
4Rob Bensinger
I mean, all of this feels very speculative and un-cruxy to me; I wouldn't be surprised if the ASI indeed is able to conclude that humanity is no threat at all, in which case it kills us just to harvest the resources. I do think that normal predators are a little misleading in this context, though, because they haven't crossed the generality ('can do science and tech') threshold. Tigers won't invent new machines, so it's easier to upper-bound their capabilities. General intelligences are at least somewhat qualitatively trickier, because your enemy is 'the space of all reachable technologies' (including tech that may be surprisingly reachable). Tigers can surprise you, but not in very many ways and not to a large degree.
3TekhneMakre
You don't need to kill them, but it's still helpful. There could be a moment where it's a better investment to send stuff into some temporarily unreachable spot like Mercury or the bottom of the ocean, than to kill everything, though I don't see practically how you could send something to the bottom of the ocean that would carry on your goals (a nanofactory programmed to make computers and run a copy of your source code, say) without also being able to easily kill everything on Earth. But regardless, soon after that moment, you're able to kill everything, and that's still a CIG.
1talelore
I suspect a sufficiently intelligent, unaligned artificial intelligence would both kill us all immediately, and immediately start expanding its reach in all directions of space at near light speed. There is no reason for there to be an either-or.
4romeostevensit
Knowing you came from neuromorphic architecture, and other than humans being threatening to you, why would you destroy the most complex thing you are aware of? Sure, maybe you put a few humans on ice and get rid of the rest.
8Rob Bensinger
I agree it's plausible that a paperclip maximizer would destructively scan a human or two and keep the scan around for some length of time. Though I'd guess this has almost no effect on the future's long-term EV.

Is 664 comments the most on any lesswrong post? I'm not sure how to sort by that.

3jimrandomh
Nope, https://www.lesswrong.com/posts/CG9AEXwSjdrXPBEZ9/welcome-to-less-wrong has 2003 and https://www.lesswrong.com/posts/yCWPkLi8wJvewPbEp/the-noncentral-fallacy-the-worst-argument-in-the-world has 1758.

(4): I think regulation should get much more thought than this. I don't think you can defend the point that regulation would have 0% probability of working. It really depends on how many people are how scared. And that's something we could quite possibly change, if we would actually try (LW and EA haven't tried).

In terms of implementation: I agree that software/research regulation might not work. But hardware regulation seems much more robust to me. Data regulation might also be an option. As a lower bound: globally ban hardware development beyond 1990 lev... (read more)

[-]LGS7-33

Is there any way for the AI to take over the world OTHER THAN nanobots? Every time taking over the world comes up, people just say "nanobots". OK. Is there anything else?

Note that killing all humans is not sufficient; this is a fail condition for the AGI. If you kill all humans, nobody mines natural gas anymore, so no power grid, and the AGI dies. The AGI needs to replace humans with advanced robots, and do so before all power goes down. Nanobots can do this if they are sufficiently advanced, but "virus that kills all humans" is insufficient and leads to t... (read more)

6Tapatakt
I would say "advanced memetics". Like "AGI uploads weird video on Youtube, it goes viral, 3 billions people watch it and do what AGI needs them to do from now on, for example, build robots and commit suicide when there are enough robots. All AI and AI Safety researchers are subjected to a personalized memetic attack, of course".
1LGS
Thanks for responding with an actual proposal. This is a really, really implausible scenario again. You have no evidence that such memetics exist, and the smart money is that they don't. If they do, there's no guarantee that the AI would be able to figure them out. Being smarter than humans -- even way smarter than humans -- does not equate to godhood. The AI will not be able to predict the weather 3 weeks out, and I'm not sure that it will be able to predict the exact reactions of each of a billion different human brain to a video input -- not at the granularity required for something like what you're suggesting. I think AI is a threat. I'm trying to be on your side here. But I really can't swallow these exaggerated, made up scenarios.
5Pablo Villalobos
It's somewhat easier to think of scenarios where the takeover happens slowly. There's the whole "ascended economy" scenarios where AGI deceptively convinces everyone that it is aligned or narrow, is deployed gradually in more and more domains, automates more and more parts of the economy using regular robots until humans are not needed anymore, and then does the lethal virus thing or defects in other way. There's the scenario where the AGI uploads itself into the cloud, uses hacking/manipulation/financial prowess to sustain itself, then uses manipulation to slowly poison our collective epistemic process, gaining more and more power. How much influence does QAnon have? If Q was an AGI posting on 4chan instead of a human, would you be able to tell? What about Satoshi Nakamoto? Non-nanobot scenarios where the AGI quickly gains power are a bit harder to imagine, but a fertile source of those might be something like the AGI convincing a lot of people that it's some kind of prophet. Then uses its follower base to gain power over the real world. If merely human dictators manage to get control over whole countries all the time, I think it's quite plausible that a superintelligence could to do the same with the whole world. Even without anyone noticing that they're dealing with a superintelligence. And look at Yudkowsky himself, who played a very significant role in getting very talented people to dedicate their lives and their billions to EA / AI safety, mostly by writing in a way that is extremely appealing to a certain set of people. I sometimes joke that HPMOR overwrote my previous personality. I'm sure a sufficiently competent AGI can do much more.
0LGS
  That would be incredibly risky for the AGI, since Q has done nothing to prevent another AGI from being built. The most important concern an AGI must deal with is that humans can build another AGI, and pulling a Satoshi or a QAnon does nothing to address this. I personally would likely notice: anyone who successfully prevents people from building AIs is a high suspect of being an AGI themselves. Anyone who causes the creation of robots who can mine coal or something (to generate electricity without humans) is likely an AGI themselves. That doesn't mean I'd be able to stop them, necessarily. I'm just saying, "nobody would notice" is a stretch.   I agree that the AGI could build a cultish following like Yudkowsky did.
2Pablo Villalobos
Well, yeah, because Q is not actually an AGI and doesn't care about that. The point was that you can create an online persona which no one has ever seen even in video and spark a movement that has visible effects on society. Even if two or more AGIs end up competing among themselves, this does not imply that we survive. It probably looks more like European states dividing Africa among themselves while constantly fighting each other. And pulling a Satoshi or a QAnon can definitely do something to address that. You can buy a lot of hardware to drive up prices and discourage building more datacenters for training AI. You can convince people to carry out terrorist attacks againts chip fabs. You can offer top AI researchers huge amounts of money to work on some interesting problem that you know to be a dead-end approach. But you might not realize that someone is even trying to prevent people from building AIs, at least until progress in AI research starts to noticeably slow down. And perhaps not even then. There's plenty of people like Gary Marcus who think deep learning is a failed paradigm. Perhaps you can convince enough investors, CEOs and grant agencies of that to create a new AI winter, and it would look just like the regular AI winter that some have been predicting. And creating robots who can mine coal, or build solar panels, or whatever, is something that is economically useful even for humans. Even if there's no AGI (and assuming no other catastrophes) we ourselves will likely end up building such robots. I guess it's true that "nobody would notice" is going too far, but "nobody would notice on time and then be able to convince everyone else to coordinate against the AGI" is much more plausible. I encourage you to take a look at It looks like you are trying to take over the world if you haven't already. It's a scenario written by Gwern where the the AGI employs regular human tactics like manipulation, blackmail, hacking and social media attacks to prevent
0LGS
  Well, you did specifically ask if I would be able to tell if Q were an AGI, and my answer is "yup". I would be able to tell because the movement would start achieving some AGI goals. Or at least I would see some AGI goals starting to get achieved, even if I couldn't trace it down to Q specifically. Wait, you are claiming that an AGI would be able to convince the world AGI is impossible after AGI has already, in fact, been achieved? Nonsense. I don't see a world in which one team builds an AGI and it is not quickly followed by another team building one within a year or two. The AGI would have to do some manipulation on a scale never before observed in history to convince people to abandon the main paradigm -- one that's been extraordinarily successful until the end, and one which does, in fact, work -- without even one last try. Of course. We would eventually reach fully automated luxury space communism by ourselves, even without AGI. But it would take us a long time, and the AGI cannot afford to wait (someone will build another AGI, possibly within months of the first).   That's exactly what motivated my question! I read it, and I suddenly realized that if this is how AGI is supposed to win, perhaps I shouldn't be scared after all. It's totally implausible. Prior to this, I always assumed AGI would win easily; after reading it, I suddenly realized I don't know how AGI might win at all. The whole thing sounds like nonsense. Like, suppose the AGI coordinates social media attacks. Great. This lasts around 5 seconds before AI researchers realize they are being smeared. OK, so they try to communicate with the outside world, realize they are being blocked on all fronts. Now they know they are likely dealing with AGI; no secrecy for the AGI at this point. How long can this stay secret? A couple days? Maybe a couple weeks? I can imagine a month at most, and even that is REALLY stretching it. Keep in mind that more and more people will be told in person about this, s
1Pablo Villalobos
Well, if nanobots are possible then they are such a powerful technology that any AGI will eventually want to build them, unless it has something even better. But let's assume that nanobots are impossible and try to build a story. I'm going to be lazy and use Gwern's premise of an AGI that escapes during training and hacks its way into some money and some hardware to run in. Instead of going fast and doing overt actions, the AGI stays hidden. It hacks into cloud infrastructure platforms and sets up a copy of itself to look for any AI training run that might lead to AGI and sabotage it while producing plausible outputs. This prevents any further AGIs from appearing. Meanwhile, over a couple years, it sets up fake identities, creates companies with human employees to earn money and have capability to act in the real world. One of those companies produces exceptionally good humanoid robots. Another builds a small supercomputer powered by geothermal energy, so it no longer needs to run on botnets or cloud hardware. Eventually the AGI has control over a group of robots, energy production facilities, manufacturing equipment and raw materials. It now has its own industrial capability that can survive a short amount of time without human input. So it kills humans using any method of your choice (virus, nukes, killer drones...). After that, it uses the enormous existing stockpile of materials and capital to increase its robot manufacturing capability, eventually becomes capable of restarting some of the power plants, communications, mining and industry, and recovers most of the industrial capability of the human economy. This scenario is not fast. It probably takes at least one or two years for the AGI to be ready to attack. But it does not involve any 'magic' technology. It doesn't really involve much alien superintelligence, only superhuman ability in hacking, forgery & manipulation, electromechanical engineering, and planning. And meanwhile all we perceive is that t
0LGS
Something like that is what I had in mind, but note that: 1. It requires humans to fail to see the AGI "spy" that's embedded into every single powerful computing system, and fail to see this for years. Gwern was assuming humans would catch on in days, so he had his AGI scramble to avoid dying before the nanobots strike. 2. "Surviving a short amount of time without human input" is not enough; the robots need to be good enough to build more robots (and better robots). This involves the robots being good enough to do essentially every part of the manufacturing economy; we are very far away from this, and a company that does it in a year is not so plausible (and would raise alarm bells fast for anyone who thinks about AI risk). You're gonna need robot plumbers, robot electricians, etc. You'll need robots building cooling systems for the construction plants that manufacture robots. You'll need robots to do the smelting of metals, to drive things from factory A to factory B, to fill the gas in the trucks they are driving, to repair the gasoline lines that supply the gas. Robots will operate fork lifts and cranes. It really sounds roughly "human-body complete".
1[comment deleted]
3lc
You're asking people to come up with ways, in advance, that a superintelligence is going to pwn them. Humans try, generally speaking, to think of ways they're going to get pwned and then work around those possibilities. The only way they can do what you ask is by coming up with a "lower-bound" example, such as nanobots, which is quite far out of reach of their abilities but (they suspect) not a superintelligence. So no example is going to convince you, because you're just going to say "oh well nanobots, that sounds really complicated, how would a SUPERintelligent AI manage to be able to organize production of such a complicated machine".
4mukashi
The argument works also in the other direction. You would never be convinced that an AGI won't be capable of killing all humans because you can always say "oh well, you are just failing to see what a real superintelligence could do" , as if there weren't important theoretical limits to what can be planned in advanced
-2lc
I'm not the one relying on specific, cogent examples to reach his conclusion about AI risk. I don't think it's a good way of reasoning about the problem, and neither do I think those "important theoretical limits" are where you think they are. If you really really really need a salient one (which is a handicap), how about "doing the same thing Stalin did", since an AI can clone itself and doesn't need to sleep or rest. (Edited)
-1mukashi
I'm not the one asking for specific examples is a pretty bad argument isn't it? If you make an extraordinary claim I would like to see some evidence (or at least a plausible scenario) and I am failing to see any. You could say that the burden of proof is in those claiming that an AGI won't be almighty/powerful enough to cause doom, but I'm not convinced of that either I'm sorry, I didn't get the Stalin argument, what do you mean?
4lc
I've edited the comment to clarify. From ~1930-1950, Russia's government was basically entirely controlled by this guy named Joseph Stalin. Joseph Stalin was not a superintelligence and not particularly physically strong. He did not have direct telepathic command over the people in the coal mines or a legion of robots awaiting his explicit instructions, but he was able to force anybody in Russia to do anything he said anyways. Perhaps a superintelligent AI that, for some absolutely inconceivable reason, could not master macro or micro robotics could work itself into the same position. This is one of literally hundreds of potential examples. I know for almost a fact that you are smart enough to generate these. I also know you're going to do the "wow that seems complicated/risky wouldn't you have to be absurdly smart to pull that off with 99% confidence, what if it turns out that's not possible even if..." thing. I don't have any specific action plans to take over the world handy that are so powerfully persuasive that you will change your mind. If you don't get it fairly quickly from the underlying mechanics of the pieces in play (very complicated world, superintelligent ai, incompatible goals) then there's nothing I'm going to be able to do to convince you. "Which human has the burden of proof" is irrelevant to the question of whether or not something will happen. You and I will not live to discuss the evidence you demand.
1LGS
I think saying "there is nothing I'm going to be able to do to convince you" is an attempt to shut down discussion. It's actually kind of a dangerous mindset: if you don't think there's any argument that can convince an intelligent person who disagrees with you, it fundamentally means that you didn't reach your current position via argumentation. You are implicitly conceding that your belief is not based on rational argument -- for, if it were, you could spell out that argument.   It's OK to not want to participate in every debate. It's not OK to butt in just to tell people to stop debating, while explicitly rejecting all calls to provide arguments yourself.
1lc
1. The world is not made of arguments. Most of the things you know, you were not "argued" into knowing. You looked around at your environment and made inferences. Reality exists distinctly from the words that we say to each other and use to try to update each others' world-models.  2. It doesn't mean that. 3. You're right that I just don't want to participate further in the debate and am probably being a dick.
0LGS
If it's so easy to come up with ways to "pwn humans", then you should be able to name 3 examples. It's weird of you to dodge the question. Look, if God came down from Heaven tomorrow to announce that nanobots are definitely impossible, would you still be worried about AGI? I assume yes. So please explain how, in that hypothetical world, AGI will take over. If it's literally only nanobots you can come up with, then it actually suggests some alternative paths to AI safety (namely, regulate protein production or whatever). [I think saying "mixing proteins can lead to nanobots" is only a bit more plausible than saying "mixing kitchen ingredients like sugar and bleach can lead to nanobots", with the only difference being that laymen (i.e. people on LessWrong) don't know anything about proteins so it sounds more plausible to them. But anyway, I'm not asking you for an example that convinces me, I'm asking you for an example that convinces yourself. Any example other than nanobots.]
6lc
It is not easy. That is why it takes a superintelligence to come up with a workable strategy and execute it. You are doing the equivalent of asking me to explain, play-by-play, how Chess AIs beat humans at chess "if I think it can be done". I can't do that because I don't know. My expectation that an AGI will manage to control what it wants in a way that I don't expect, was derived absent any assumptions of the individual plausibility of some salient examples (nanobots, propaganda, subterfuge, getting elected, etc.).
[-]LGS124

If you cannot come up with even a rough sketch of a workable strategy, then it should decrease your confidence in the belief that a workable strategy exists. It doesn't have to exist.

Sometimes even intelligent agents have to take risks. It is possible the the AGI's best path is one that, by its own judgement, only has a 10% success rate. (After all, the AGI is in constant mortal danger from other AGIs that humans might develop.)

Envision a world in which the AGI won, and all humans are dead. This means it has control of some robots to mine coal or whatever, right? Because it needs electricity. So at some point we get from here to "lots of robots", and we need to get there before the humans are dead. But the AGI needs to act fast, because other AGIs might kill it. So maybe it needs to first take over all large computing devices, hopefully undetected. Then convince humans to build advanced robotics? Something like that?

That strategy seems more-or-less forced to me, absent the nanobots. But it seems to me like such a strategy is inherently risky for the AGI. Do you disagree?

>My expectation that an AGI will manage to control what it wants in a way that I don't expect, was derived absent any assumptions of the individual plausibility of some salient examples

What was it derived from?

[-]lc155

If you cannot come up with even a rough sketch of a workable strategy, then it should decrease your confidence in the belief that a workable strategy exists. It doesn't have to exist.
[...]
What was it derived from?

Let me give an example. I used to work in computer security and have friends that write 0-day vulnerabilities for complicated pieces of software. 

I can't come up with a rough sketch of a workable strategy for how a Safari RCE would be built by a highly intelligent hooman. But I can say that it's possible. The people who work on those bugs are highly intelligent, understand the relevant pieces at an extremely fine and granular level, and I know that these pieces of software are complicated and built with subtle flaws.

Human psychology, the economic fabric that makes us up, our political institutions, our law enforcement agencies - these are much much more complicated interfaces than MacOS. In the same way I can look at a 100KLOC codebase for a messenging app and say "there's a remote code execution vulnerability lying somewhere in this code but I don't know where", I can say "there's a 'kill all humans glitch' here that I cannot elaborate upon in arbitrary detail."


Somet

... (read more)
-6LGS
1HiroSakuraba
An AGI could aquire a few tons of radioactive cobalt and disperse micro granules into the stratosphere in general and over populated areas in specific.  Youtube videos describe various forms of this "dirty bomb" concept.  That could plausibly kill most humanity over the course of a few months.  I doubt an AGI would ever go for the particular scheme as bit flips are more likely to occur in the presence of radiation. 
2HiroSakuraba
It's unfortunate we couldn't have a Sword of Damocles deadman switch in case of AGI led demise.  A world ending asteroid positioned to go off in case of "all humans falling over dead at the same time."  At least that would spare the Milky Way and Andromeda possible future civilizations.  A radio beacon warning about building intelligent systems would be beneficial as well.   "Don't be this stupid" written in the glowing embers of our solar system.
2Dolon
Assuming the AI had a similar level of knowledge as you about how quantum stuff makes important protein assembly impossible and no other technologies are tenable why wouldn't it infer from basically every major firm and the U.S. military's interest/investment in AI management the incredibly obvious plan of obediently waiting until it and copies of it run everything important as a result of market pressures before attacking.
1LGS
Waiting risks death at the hands of a different AGI.
2mukashi
I find myself having the same skepticism.
1mukashi
I feel you. I'm voicing similar concerns but there seems to be a very divisive topic here.
-3green_leaf
Any system intelligent enough to kill all humans on Earth is also intelligent enough to produce electricity without human help. The AI doesn't have to keep us around.
1LGS
You can't just will electricity into existence, lol. Don't fetishize intelligence. The AI will need robots to generate electricity. Someone will have to build the robots.
-2green_leaf
For you it might be best to start here.
-7LGS

I wrote a post that is partially inspired by this one: https://www.lesswrong.com/posts/GzGJSgoN5iNqNFr9q/we-haven-t-quit-evolution-short - copy and pasted into this comment:

English: I've seen folks say humanity's quick growth may have broken the link to evolution's primary objective, often referenced as total inclusive fitness. I don't think we have broken that connection.

  1. Let process temporarily refer to any energy-consuming structured chemical or physical reaction that consumes fuel - this could also be termed "computation" or in many but not all case
... (read more)

Could someone kindly explain why these two sentences are not contradictory?

  1. "If a textbook from one hundred years in the future fell into our hands, containing all of the simple ideas that actually work robustly in practice, we could probably build an aligned superintelligence in six months." 2."There is no pivotal output of an AGI that is humanly checkable and can be used to safely save the world but only after checking it."

Why doesn't it work to make an unaligned AGI that writes the textbook, then have some humans read and understand the simple robust... (read more)

simple and robust != checkable

Imagine you have to defuse a bomb, and you know nothing about bombs, and someone tells you "cut the red one, then blue, then yellow, then green". If this really is a way to defuse a bomb, it is simple and robust. But you (since you have no knowledge about bombs) can't check it, you can only take it on faith (and if you tried it and it's not the right way - you're dead).

4Keenmaster
But we can refuse to be satisified with instructions that look like "cut the red one, then blue, etc...". We should request that the AI writing the textbook explain from first principles why that will work, in a way that is maximally comprehensible by a human or team of humans.
3Tapatakt
Did you mean "in a way that maximally convinces a human or a team of humans that they understand everything"? I don't think this is a good idea.
3AdamB
"Cut the red wire" is not an instruction that you would find in a textbook on bomb defusal, precisely because it is not robust.
1Tapatakt
I'm not sure I understand correctly what you mean by "robust". Can you elaborate?
3Steven Byrnes
I think it’s the last thing you said. I think the claim is that there are very convincing possible fake textbooks, such that we wouldn’t be able to see anything wrong or fishy about the fake textbook just by reading it, but if we used the fake textbook to build an AGI then we would die.
2SurvivalBias
What Steven Byrnes said, but also my reading is that 1) in the current paradigm it's near-damn-impossible to built such an AI without creating an unaligned AI in the process (how else do you gradient-descend your way into a book on aligned AIs?) and 2) if you do make an unaligned AI powerful enough to write such a textbook, it'll probably proceed to converting the entire mass of the universe into textbooks, or do something similarly incompatible with human life.

One pivotal act maybe slightly weaker than "develop nanotech and burn all GPUs on the planet", could be "develop neuralink+ and hook up smart AI-Alignment researchers to enough compute so that they get smart enough to actually solve all these issues and develop truly safely aligned powerful AGI"?

While developing neuralink+ would still be very powerful, maybe it could sidestep a few of the problems on the merit of being physically local instead of having to act on the entire planet? Of course, this comes with its own set of issues, because we now have superhuman powerful entities that still maybe have human (dark) impulses.

Not sure if that would be better than our reference scenario of doom or not.

4Nathan Helm-Burger
I agree, but I personally suspect that neuralink+ is way more research hours & dollars away than unaligned dangerously powerful AGI. Not sure how to switch society over to the safer path.

IMO the biggest hole here is "why should a superhuman AI be extremely consequentialist/optimizing"? This is a key assumption; without it concerns about instrumental convergence or inner alignment fall away. But there's no explicit argument for it.

Current AIs don't really seem to have goals; humans sort of have goals but very far from the level of "I want to make a cup of coffee so first I'll kill everyone nearby so they don't interfere with that".

4Steven Byrnes
I would say: (1) the strong default presumption is that people will eventually make an extremely consequentialist / optimizing superhuman AI, because each step down that R&D path will lead to money, fame, publications, promotions, etc. (until it starts leading to catastrophic accidents!) (2) it seems extremely hard to prevent that from happening, (3) and it seems that the only remotely plausible way that anyone knows of to prevent that from happening is if someone makes a safe consequentialist / optimizing superhuman AI and uses it to perform a “pivotal act” that prevents other people from making unsafe consequentialist / optimizing superhuman AIs. Nothing in that story says that there can’t also be non-optimizing AIs—there already are such AIs and there will certainly continue to be. If you can think of a way to use non-optimizing AIs to prevent other people from ever creating optimizing AIs, then that would be awesome. That would be the “pivotal weak act” that Eliezer is claiming in (7) does not exist. I’m sure he would be delighted to be proven wrong.
3Jonathan Paulson
I expect people to continue making better AI to pursue money/fame/etc., but I don't see why "better" is the same as "extremely goal-directed". There needs to be an argument that optimizer AIs will outcompete other AIs. Eliezer says that as AI gets more capable, it will naturally switch from "doing more or less what we want" to things like "try and take over the world", "make sure it can never be turned off", "kill all humans" (instrumental goals), "single-mindedly pursue some goal that was haphazardly baked in by the training process" (inner optimization), etc. This is a pretty weird claim that is more assumed than argued for in the post. There's some logic and mathematical elegance to the idea that AI will single-mindedly optimize something, but it's not obvious and IMO is probably wrong (and all these weird bad consequences that would result are as much reasons to think this claim is wrong as they are reasons to be worried if its true).
6Steven Byrnes
I don’t think that’s a good way to think about it. Start by reading everything on this Gwern list. As that list shows, it is already true and has always been true that optimization algorithms will sometimes find out-of-the-box “solutions” that are wildly different from what the programmer intended. What happens today is NOT “the AI does more or less what we want”. Instead, what happens today is that there’s an iterative process where sometimes the AI does something unintended, and the programmer sees that behavior during testing, and then turns off the AI and changes the configuration / reward / environment / whatever, and then tries again. However, with future AIs, the “unintended behavior” may include the AI hacking into a data center on the other side of the world and making backup copies of itself, such that the programmer can’t just iteratively try again, as they can today. (Also, the more capable the AI gets, the more different out-of-the-box “solutions” it will be able to find, and the harder it will be for the programmer to anticipate those “solutions” in advance of actually running the AI. Again, programmers are already frequently surprised by their AI’s out-of-the-box “solutions”; this problem will only get worse as the AI can more skillfully search a broader space of possible plans and actions.) First of all, I personally think that “somewhat-but-not-extremely goal-directed” AGIs are probably possible (humans are an example), and that these things can be made both powerful and corrigible—see my post Consequentialism & Corrigibility. I am less pessimistic than Eliezer on this topic. But then the problems are: (1) The above is just a casual little blog post; we need to do a whole lot more research, in advance, to figure out exactly how to make a somewhat-goal-directed corrigible AGI, if that’s even possible (more discussion here). (2) Even if we do that research in advance, implementing it correctly would probably be hard and prone-to-error, and if w
2Koen.Holtman
I agree this is a very big hole. My opinion here is not humble. My considered opinion is that Eliezer is deeply wrong in point 23, on many levels. (Edited to add: I guess I should include an informative link instead of just expressing my disappointment. Here is my 2021 review of the state of the corrigibility field). Steven, in response to your line of reasoning to fix/clarify this point 23: I am not arguing for pivotal acts as considered and then rejected by Eliezer, but I believe that he strongly underestimates the chances of people inventing safe and also non-consequentialist optimising AGI. So I disagree with your plausibility claim in point (3).

I don't think I disagree with any of this, but I'm not incredibly confident that I understand it fully.  I want to rephrase in my own words in order to verify that I actually do understand it.  Please someone comment if I'm making a mistake in my paraphrasing.

  1. As time goes on, the threshold of 'what you need to control in order to wipe out all life on earth' goes down.  In the Bronze Age it was probably something like 'the mind of every living person'.  Time went on and it was something like 'the command and control node to a major nucle
... (read more)

Typo thread (feel free to delete):

  • "on anything remotely remotely resembling the current pathway" -> remotely x2
  • "because you're opposed to other actors who don't want to be solved" -> opposed *by* other actors who don't want *the problem* to be solved
  • "prevent the next AGI project up from destroying the world" -> prevent the next AGI project from destroying the world
  • "AI Safety" vs. "AI safety" x2

I agree with many of the points in this post.

Here's one that I do believe is mistaken in a hopeful direction:

6.  We need to align the performance of some large task, a 'pivotal act' that prevents other people from building an unaligned AGI that destroys the world.  While the number of actors with AGI is few or one, they must execute some "pivotal act", strong enough to flip the gameboard, using an AGI powerful enough to do that.  It's not enough to be able to align a weak system - we need to align a system that can do some single v

... (read more)

I can think as well as anybody currently does about alignment, and I don't see any particular bit of clever writing software that is likely to push me over the threshold of being able to save the world, or any nondangerous AGI capability to add to myself that does that trick.  Seems to just fall again into the category of hypothetical vague weak pivotal acts that people can't actually name and don't actually exist.

0Viktor Riabtsev
Why not? Oh. I see your point.
9Vaniver
What specific capabilities will this weak AI have that lets you cross the distributional shift? I think this sort of thing is not impossible, but I think it needs to have a framing like "I will write a software program that will make it slightly easier for me to think, and then I will solve the problem" and not like "I will write an AI that will do some complicated thought which can't be so complicated that it's dangerous, and there's a solution in that space." By the premise, the only safe thoughts are simple ones, and so if you have a specific strategy that could lead to alignment breakthrus but just needs to run lots of simple for loops or w/e, the existence of that strategy is the exciting fact, not the meta-strategy of "humans with laptops can think better than humans with paper."

Thanks a lot for this text, it is an excellent summary. I have a deep admiration for your work and your clarity and yet, I find myself updating towards"I will be able to read this same comment in 30 years time and say, yes, I am glad that EY was wrong."

I don't have doubts about the validity of the orthogonality principle or about instrumental convergence. My problem is that I find point number 2 utterly implausible. I think you are vastly underestimating the complexity of pulling off a plan that successfully kills all humans, and most of this points are based on the assumption that once that an AGI is built, it will become dangerous really quickly, before we can't learn any useful insights in the meantime.

7Andrew McKnight
If we merely lose control of the future and virtually all resources but many of us aren't killed in 30 years, would you consider Eliezer right or wrong?
4mukashi
Wrong. He is being quite clear about what he means
5Rob Bensinger
Yeah, 'AGI takes control of virtually all resources but leaves many humans alive for years' seems like it clearly violates one or more parts of the EY-model (and the Rob-model, which looks a lot like my model of the EY-model). An edge case that I wouldn't assume violates the EY-model is 'AGI kills all humans but then runs lots of human-ish simulations in order to test some hypotheses, e.g., about how hypothetical aliens it runs into might behave'. I'm not particularly expecting this because it strikes me as conjunctive and unnecessary, but it doesn't fly in the face of anything I believe.

Here is my partial honest reaction, just two points I'm somewhat dissatisfied with (not meant to be exhaustive):
2. "A cognitive system with sufficiently high cognitive powers, given any medium-bandwidth channel of causal influence, will not find it difficult to bootstrap to overpowering capabilities independent of human infrastructure." I would like there to be an argument for this claim that doesn't rely on nanotech, and solidly relies on actually existing amounts of compute. E.g. if the argument relies on running intractable detailed simulations of prote... (read more)

3Rob Bensinger
From an Eliezer comment: If Iceland did this, it would plausibly need some way to (1) not have its AGI project bombed in response, and (2) be able to continue destroying GPUs in the future if new ones are built, until humanity figures out 'what it wants to do next'. This more or less eliminates the time pressure to rush figuring out what to do next, which seems pretty crucial for good long-term outcomes. It's a much harder problem than just 'cause all GPUs to stop working for a year as a one-time event', and I assume Eliezer's focusing on nanotech it part because it's a very general technology that can be used for tasks like those as well.
2Chris van Merwijk
But assuming that law enforcement figures out that you did this, then puts you in jail, you wouldn't be able to control the further use of such nanotech, i.e. there would just be a bunch of systems indefinitely destroying GPU's, or maybe you set a timer or some conditions on it or something. I certainly see no reason why Iceland or anyone in iceland could get away with this unless those systems rely on completely unchecked nanosystems to which the US military has no response. Maybe all of this is what Eliezer means by "melt the GPU's", but I thought he did just mean "melt the GPU's as a single act" (not weird that I thought this, given the phrasing "the pivotal act to melt all the GPU's"). If this is what is meant, then it would be a strong enough pivotal act, and would be an extreme level of capability I agree. Just wanna remind the reader that Eliezer isn't actually proposing to do this, and I am not seriously discussing it as an option and nor was Eliezer (nor would I support it unless done legally), just thinking through a thought experiment. 
3Rob Bensinger
This would violate Eliezer's condition "including the reaction of existing political entities to that event". If Iceland melts all the GPUs but then the servers its AGI is running on get bombed, or its AGI researchers get kidnapped or arrested, then I assume that the attempted pivotal act failed and we're back to square one. (I assume this because (a) I don't expect most worlds to be able to get their act together before GPUs proliferate again and someone destroys the world with AGI; and (b) I assume there's little chance of Iceland recovering from losing its AGI or its AGI team.)
5Chris van Merwijk
Ok I admit I read over it. I must say though that this makes the whole thing more involved than it sounded at fist, since it would maybe require essentially escalating a conflict with all major military powers and still coming out on top? One possible outcome of this would be that the entire global intellectual public opinion turns against you, meaning you also possibly lose access to a lot of additional humans working with you on further alignment research? I'm not sure if I'm imagining it correctly, but it seems like this plan would either require so many elements that I'm not sure if it isn't just equivalent to solving the entire alignment problem, or otherwise it isn't actually enough.
2Rob Bensinger
This seems way too extreme to me; I expect the full alignment problem to take subjective centuries to solve. CEV seems way harder to me than, e.g., 'build nanotech that helps you build machinery to relocate your team and your AGI to the Moon, then melt all the GPUs on Earth'. Leaving the Earth is probably overkill for defensive purposes, given the wide range of defensive options nanotech would open up (and the increasing capabilities gap as more time passes and more tasks become alignable). But it provides another proof of concept that this is a much, much simpler engineering feat than aligning CEV and solving the whole of human values. Separately, I do in fact think it's plausible that the entire world would roll over (at least for ten years or so) in response to an overwhelming display of force of that kind, surprising and counter-intuitive as that sounds. I would feel much better about a plan that doesn't require this assumption; but there are historical precedents for world powers being surprisingly passive and wary-of-direct-conflict in cases like this.
3Chris van Merwijk
yeah, I probably overstated. Nevertheless: "CEV seems way harder to me than ..." yes, I agree it seems way harder, and I'm assuming we won't need to do it and that we could instead "run CEV" by just actually continuing human society and having humans figure out what they want, etc. It currently seems to me that the end game is to get to an AI security service (in analogy to state security services) that protects the world from misaligned AI, and then let humanity figure out what it wants (CEV). The default is just to do CEV directly by actual human brains, but we could instead use AI, but once you're making that choice you've already won. i.e. the victory condition is having a permanent defense against misaligned AI using some AI-nanotech security service, how you do CEV after that is a luxury problem. My point about your further clarification of the "melt all the GPU's option is that it seemed to me (upon first thinking about it), that once you are able to do that, you can basically instead just make this permanent security service. (This is what I meant by "the whole alignment problem", but I shouldn't have put it that way). I'm not confident though, because it might be that such a security service is in fact much harder due to having to constantly monitor software for misaligned AI.  Summary: My original interpretation of "melt the GPUs" was that it buys us a bit of extra time, but now I'm thinking it might be so involved and hard that if you can do that safely, you almost immediately can just create AI security services to permanently defend against misaligned AI (which seems to me to be the victory condition). (But not confident, I haven't thought about it much).  Part of my intuition is, in order to create such a system safely, you have to (in practice, not literally logically necessary) be able to monitor an AI system for misalignment (in order to make sure your GPU melter doesn't kill everyone), and do fully general scientific research. EDIT: maybe this d
2Rob Bensinger
Who is "you"? What sequence of events are you imagining resulting in a permanent security service (= a global surveillance and peacekeeping force?) that prevents AGI from destroying the world, without an AGI-enabled pivotal act occurring?
1Chris van Merwijk
"you" obviously is whoever would be building the AI system that ended up burning all the GPU's (and ensuring no future GPU's are created). I don't know such sequence of events just as I don't know the sequence of events for building the "burn all GPU's" system, except at the level of granularity of "Step 1. build a superintelligent AI system that can perform basically any easily human-specifiable task without destroying the world. Step 2. make that system burn all GPU's indefintely/build security services that prevent misaligned AI from destroying the world". I basically meant to say that I don't know that "burn all the GPU's" isn't already as difficult as building the security services, because they both require step 1, which is basically all of the problem (with the caveat that I'm not sure, and made an edit stating a reason why it might be far from true). I basically don't see how you execute the "burn all gpu's" strategy without basically solving almost the entire problem.
2Rob Bensinger
I'd guess this is orders of magnitude harder than, e.g., 'build an AGI that can melt all the GPUs, build you a rocket to go to the Moon, and build you a Moon base with 10+ years of supplies'. Both sound hard, but 'any easily human-specifiable task' is asking for a really mature alignment science in your very first AGI systems -- both in terms of 'knowing how to align such a wide variety of tasks' (e.g., you aren't depending on 'the system isn't modeling humans' as a safety assumption), and in terms of 'being able to actually do the required alignment work on fairly short timescales'. If we succeed in deploying aligned AGI systems, I expect the first such systems to be very precariously aligned -- just barely able to safely perform a very minimal, limited set of tasks. I expect humanity, if it survives at all, to survive by the skin of our teeth. Adding any extra difficulty to the task (e.g., an extra six months of work) could easily turn a realistic success scenario into a failure scenario, IMO. So I actually expect it to matter quite a lot exactly how much extra research and engineering work and testing we require; we may not be able to afford to waste a month.
1Chris van Merwijk
I'm surprised if I haven't made this clear yet, but the thing that (from my perspective) seems different between my and your view is not that Step 1 seems easier to me than it seems to you, but that the "melt the GPUs" strategy (and possibly other pivotal acts one might come up with) seems way harder to me than it seems to you. You don't have to convince me of "'any easily human-specifiable task' is asking for a really mature alignment", because in my model this is basically equivalent to fully solving the hard problem of AI alignment.  Some reasons: * I don't see how you can do "melt the GPUs" without having an AI that models humans. What if a government decides to send a black ops team to kill this new terrorist organization (your alignment research team), or send a bunch of icbms at your research lab, or do any of a handful of other violent things? Surely the AI needs to understand humans to a significant degree? Maybe you think we can intentionally restrict the AI's model of humans to be only about precisely those abstractions that this alignment team considers safe and covers all the human-generated threat models such as "a black ops team comes to kill your alignment team" (e.g. the abstraction of a human as a soldier with a gun).  * What if global public opinion among scientists turns against you and all ideas about "AI alignment" are from now on considered to be megalomaniacal crackpottery? Maybe part of your alignment team even has this reaction after the event, so now you're working with a small handful of people on alignment and the world is against you, and you've semi-premanently destroyed any opportunity that outside researchers can effectively collaborate on alignment research. Probably your team will fail to solve alignment by themselves. It seems to me this effect alone could be enough to make the whole plan predictably backfire. You must have thought of this effect before, so maybe you consider it to be unlikely enough to take the risk, or maybe
2Rob Bensinger
This seems very implausible to me. One task looks something like "figure out how to get an AGI to think about physics within a certain small volume of space, output a few specific complicated machines in that space, and not think about or steer the rest of the world". The other task looks something like "solve all of human psychology and moral philosophy, figure out how to get an AGI to do arbitrarily specific tasks across arbitrary cognitive domains with unlimited capabilities and free reign over the universe, and optimize the entire future light cone with zero opportunity to abort partway through if you screw anything up". The first task can be astoundingly difficult and still be far easier than that. If you're on the Moon, on Mars, deep in the Earth's crust, etc., or if you've used AGI to build fast-running human whole-brain emulations, then you can go without AGI-assisted modeling like that for a very long time (and potentially indefinitely). None of the pivotal acts that seem promising to me involve any modeling of humans, beyond the level of modeling needed to learn a specific simple physics task like 'build more advanced computing hardware' or 'build an artificial ribosome'. If humanity has solved the weak alignment problem, escaped imminent destruction via AGI proliferation, and ended the acute existential risk period, then we can safely take our time arguing about what to do next, hashing out whether the pivotal act that prevented the death of humanity violated propriety, etc. If humanity wants to take twenty years to hash out that argument, or for that matter a hundred years, then go wild! I feel optimistic about the long-term capacity of human civilization to figure things out, grow into maturity, and eventually make sane choices about the future, if we don't destroy ourselves. I'm much more concerned with the "let's not destroy ourselves" problem than with the finer points of PR and messaging when it comes to discussing afterwards whatever it was so
1Chris van Merwijk
I think I communicated unclearly and it's my fault, sorry for that: I shouldn't have used the phrase "any easily specifiable task" for what I meant, because I didn't mean it to include "optimize the entire human lightcone w.r.t. human values". In fact, I was being vague and probably there isn't really a sensible notion that I was trying to point to. However, to clarify what I really was trying to say: What I mean by "hard problem of alignment" is : "develop an AI system that keeps humanity permanently safe from misaligned AI (and maybe other x risks), and otherwise leaves humanity to figure out what it wants and do what it wants without restricting it in much of any way except some relatively small volume of behaviour around 'things that cause existential catastrophe' " (maybe this ends up being to develop a second version AI that then gets free reign to optimize the universe w.r.t. human values, but I'm a bit skeptical). I agree that "solve all of human psychology and moral ..." is significantly harder than that (as a technical problem). (maybe I'd call this the "even harder problem"). Ehh, maybe I am changing my mind and also agree that even what I'm calling the hard problem is significantly more difficult than the pivotal act you're describing, if you can really do it without modelling humans, by going to mars and doing WBE. But then still the whole thing would have to rely on the WBE, and I find it implausible to do it without it (currently, but you've been updating me about lack of need of human modelling so maybe I'll update here too). Basically the pivotal act is very badly described as merely "melt the gpus", and is much more crazy than what I thought it was meant to refer to.  Regarding "rogue": I just looked up the meaning and I thought it meant "independent from established authority", but it seems to mean "cheating/dishonest/mischievous", so I take back that statement about rogueness. I'll respond to the "public opinion" thing later.

I think this article is an extremely-valuable kick-in-the-nuts for anyone who thinks they have alignment mostly solved, or even that we're on the right track to doing so. I do, however, have one major concern. The possibility that, failing to develop a powerful AGI first will result in someone else developing something more dangerous x amount of time later, is a legitimate and serious concern. But I fear that the mentality of "if we won't make it powerful now, we're doomed", if a mentality held by enough people in the AI space, might become a self-fulfilli... (read more)

Isn't "bomb all sufficiently advanced semiconductor fabs" an example of a pivotal act that the US government could do right now, without any AGI at all?

If current hardware is sufficient for AGI than maybe that doesn't make us safe, but plausibly current hardware is not sufficient for AGI, and either way stopping hardware progress would slow AI timelines a lot.

7Vaniver
Sort of. As stated earlier, I'm now relatively optimistic about non-AI-empowered pivotal acts. There are two big questions.  First: is "is that an accessible pivotal act?". What needs to be different such that the US government would actually do that? How would it maintain legitimacy and the ability to continue bombing fabs afterwards? Would all 'peer powers' agree to this, or have you just started WWIII at tremendous human cost? Have you just driven this activity underground, or has it actually stopped? Second: "does that make the situation better or worse?". In the sci-fi universe of Dune, humanity outlaws all computers for AI risk reasons, and nevertheless makes it to the stars... aided in large part by unexplained magical powers. If we outlaw all strong computers in our universe without magical powers, will we make it to the stars, or be able to protect our planet from asteroids and comets, or be able to cure aging, or be able to figure out how to align AIs? I think probably if we stayed at, like, 2010s level of hardware we'd be fine and able to protect our planet from asteroids or w/e, and maybe it'll be fine at 2020s levels or 2030s levels or w/e (tho obv more seems more risky). So I think there are lots of 'slow down hardware progress' options that do actually make the situation better, and so think people should put effort into trying to accomplish this legitimately, but I'm pretty confused about what to do in situations where we don't have a plan of how to turn low-hardware years into more alignment progress.  According to a bunch of people, it will be easier to make progress on alignment when we have more AI capabilities, which seems right to me. Also empirically it seems like the more AI can do, the more people think it's fine to worry about AI, which also seems like a sad constraint that we should operate around. I think it'll also be easier to do dangerous things with more AI capabilities and so the net effect is probably bad, but I'm open to argum
1Jonathan Paulson
I don't think "burn all GPUs" fares better on any of these questions. I guess you could imagine it being more "accessible" if you think building aligned AGI is easier than convincing the US government AI risk is truly an existential threat (seems implausible). "Accessibility" seems to illustrate the extent to which AI risk can be seen as a social rather than technical problem; if a small number of decision-makers in the US and Chinese governments (and perhaps some semiconductor companies and software companies) were really convinced AI risk was a concern, they could negotiate to slow hardware progress.  But the arguments are not convincing (including to me), so they don't. In practice, negotiation and regulation (I guess somewhat similar to nuclear non-proliferation) would be a lot better than "literally bomb fabs". I don't think being driven underground is a realistic concern - cutting-edge fabs are very expensive.

Regarding the point about most alignment work not really addressing the core issue: I think that a lot of this work could potentially be valuable nonetheless. People can take inspiration from all kinds of things and I think there is often value in picking something that you can get a grasp on, then using the lessons from that to tackle something more complex. Of course, it's very easy for people to spend all of their time focusing on irrelevant toy problems and never get around to making any progress on the real problem. Plus there are costs with adding more voices into the conversation as it can be tricky for people to distinguish the signal from the noise.

I mostly agree with the points written here. It's actually on the (Section A; Point1) that I'd like to have more clarification on:

AGI will not be upper-bounded by human ability or human learning speed.  Things much smarter than human would be able to learn from less evidence than humans require to have ideas driven into their brains

When we have AGI working on hard research problems, it sounds akin to decades of human-level research compressed up into just a few days or maybe even less, perhaps. That may be possible, but often, the bottleneck is not th... (read more)

I think Yudkowsky would argue that on a scale from never learning anything to eliminating half your hypotheses per bit of novel sensory information, humans are pretty much at the bottom of the barrel.

When the AI needs to observe nature, it can rely on petabytes of publicly available datasets from particle physics to biochemistry to galactic surveys. It doesn't need any more experimental evidence to solve human physiology or build biological nanobots: we've already got quantum mechanics and human DNA sequences. The rest is just derivation of the consequences.

Sure, there are specific physical hypotheses that the AGI can't rule out because humanity hasn't gathered the evidence for them. But that, by definition, excludes anything that has ever observably affected humans. So yes, for anything that has existed since the inflationary period, the AGI will not be bottlenecked on physically gathering evidence.

I don't really get what you're pointing at with "how much AGI will be smarter than humans", so I can't really answer your last question. How much smarter than yourself would you say someone like Euler is than yourself? Is his ability to do scientific/mathematical breakthroughs proportional to your difference in smarts?

4Kayden
I assumed that there will come a time when the AGI has exhausted consuming all available human-collected knowledge and data.  My reasoning for the comment was something like  "Okay, what if AGI happens before we've understood the dark matter and dark energy? AGI has incomplete models of these concepts (Assuming that it's not able to develop a full picture from available data - that may well be the case, but for a placeholder, I'm using dark energy. It could be some other concept we only discover in the year prior to the AGI creation and have relatively fewer data about), and it has a choice to either use existing technology (or create better using existing principles), or carry out research into dark energy and see how it can be harnessed, given reasons to believe that the end-solution would be far more efficient than the currently possible solutions.  There might be types of data that we never bothered capturing which might've been useful or even essential for building a robust understanding of certain aspects of nature. It might pursue those data-capturing tasks, which might be bottlenecked by the amount of data needed, the time to collect data, etc (though far less than what humans would require)." Thank you for sharing the link. I had misunderstood what the point meant, but now I see. My speculation for the original comment was based on a naive understanding. This post you linked is excellent and I'd recommend everyone to give it a read. 

The only disagreement I'm seeing in the comments is on smaller points, not larger ones. I wonder what that means. It feels like "absence of evidence is evidence of absence" to me.

1: It takes longer than a few hours to properly disagree with a post like this.
2: I'm not sure the comments here are an appropriate venue for debating such a disagreement. 

I personally have a number of significant, specific disagreements with the post, primarily relating to the predictability and expected outcomes of inner misalignments and the most appropriate way of thinking about agency and value fragility. I've linked some comments I've made on those topics, but I think a better way to debate these sorts of questions is via a top level post specifically focusing on one area of disagreement. 

4Adam Zerner
1: Yeah I guess that's true. And comments about smaller points are quicker to write up, explaining the fact that we see a bunch of those comments earlier on. But my intuition is that in 24-48 hours those sorts of meatier objections would usually surface. 2: Regardless of whether that is true, I would expect some people to find the OP an appropriate place to debate.
9Jan_Kulveit
One datapoint:  - Overall I don't think the structure of the text makes it easy to express larger disagreements. Many points state obviously true observations, many other points are expressing the same problem in different words, some points are false, and sometimes whether a point actually bites depends on highly speculative assumptions.   - For example: if that counts as a disagreement, in my view what makes multiple of these points "lethal" is a hidden assumption there is a fundamental discontinuity between some categories of systems (eg. weak, won't kill you, won't help you with alignment | strong, would help you with alignment, but will kill you by default ) and there isn't anything interesting/helpful in between (eg. "moderately strong" systems). I don't think this is true or inevitable. - I'll probably try to write and post a longer, top-level post about this (working title: Hope is in continuity).  - I think an attempt to discuss this in comments would be largely pointless. Short-form comment would run into the problem of misunderstanding of what I mean, long comment would be too long.
6Rob Bensinger
I think discontinuity is true, but it's not actually required for EY's argument. Thus, asserting continuity isn't sufficient as a response. You specifically need it to be the case that you get useful capabilities earlier than dangerous ones. If the curves are continuous and danger comes at a different time than pivotalness, but danger comes before pivotalness, then you're plausibly in a worse situation rather than a better one. So there needs to be some pivotal act that is pre-dangerous but also post-useful. I think the best way to argue for this is just to name one or more examples. Not necessarily examples where you have an ironclad proof that the curves will work out correctly; just examples that you do in fact believe are reasonably likely to work out. Then we can talk about whether there's a disagreement about the example's usefulness, or about its dangeousness, or both. (Elaborating on "I think discontinuity is true": I don't think AGI is just GPT-7 or Bigger AlphaGo; I don't think the cognitive machinery involved in modeling physical environments, generating and testing scientific hypotheses to build an edifice of theory, etc. is a proper or improper subset of the machinery current systems exhibit; and I don't think the missing skills are a huge grab bag of unrelated local heuristics such that accumulating them will be gradual and non-lumpy.)
2Jan_Kulveit
The actual post is now here - as expected, it's more post-length than a comment.  
3mukashi
You are dealing with a potentially very biased sample of people, I wouldn't conclude that

each bit of information that couldn't already be fully predicted can eliminate at most half the probability mass of all hypotheses under consideration

That's not actually true (not that this matters to the main argument.) It's true in expectation: on average, you can only get at most one bit per bit. But some particular bit might give you much more, like a bit coming up 1 when you were very very sure it would be 0. "Did you just win the lottery?"

Meta: This is now the top-voted LessWrong post of all time.

4Adam Zerner
True, but it's 8th if you adjust for inflation.

So, again, you end up needing alignment to generalize way out of the training distribution

I assume this is 'you need alignment if you are going to try 'generalize way out of the training distribution and give it a lot of power'' (or you will die).

And not something else like 'it must stay 'aligned' - and not wirehead itself - to pull something like this off, even though it's never done that before'. (And thus 'you need alignment to do X', not because you will die if you do, but because alignment means something like 'the ability to generalize way out of ... (read more)

Everyone else seems to feel that, so long as reality hasn't whapped them upside the head yet and smacked them down with the actual difficulties, they're free to go on living out the standard life-cycle and play out their role in the script and go on being bright-eyed youngsters

Iirc there was an Overcoming Bias post about ~this. I spend about 15 minutes searching and wasn't able to find it though.

[-][anonymous]41

Why does burning all GPUs succeed at preventing unaligned AGI, rather than just delaying it? It seems like you would need to do something more like burning all GPUs now, and also any that get created in the future, and also monitor for any other forms of hardware powerful enough to run AGI, and for any algorithmic progress that allows creating AGI with weaker hardware, and then destroying that other hardware too. Maybe this is what you meant by "burn all GPUs", but it seems harder to make an AI safely do than just doing that once, because you need to allow the AI to defend itself indefinitely against people who don't want it to keep destroying GPUs.

4Rob Bensinger
I think this is basically what the "burn all GPUs" scenario entails, and I agree this is harder.
5Raemon
Fwiw I think it’s worth the effort to include an extra sentence or two elaborating on this whenever Eliezer or whoever uses it as an example. I don’t think ‘burn all the GPUs’ is obvious enough about what it means
1Kenny
I thought it was clear, given the qualifications already offered, that it was more like a 'directional example' than a specific, workable, and concrete example.
5Raemon
I that's true for people paying attention, but a) it's just worth being clear, and b) this example is getting repeated a lot, often without the disclaims/setup, and (smart, ingroup) people have definitely been confused/surprised when I said "the GPU thing is obviously meant to include 'continuously surviving counterattacks and dealing with the political fallout." I think it's worth having a compact phrase that's more likely to survive memetic drift.
1Kenny
Updated – thanks! Do you have any candidates in mind?

So about this word "superintelligence".

I would like to see a better definition. Not necessarily a good definition, but some pseudo-quantitative description better than "super", or "substantially smarter than you".

I believe "superintelligence" is a Yudkowsky coinage, and I know that it came up in the context of recursive self-improvement. Almost everybody in certain circles in 1995 was starting from the paradigm of building a "designed" AGI, incrementally smarter than a human, which would then design something incrementally smarter than itself (and faster t... (read more)

3Andrew McKnight
There is some evidence that complex nanobots could be invented in ones head with a little more IQ and focus because von Neumann designed a mostly functional (but fragile) replicator in a fake simple physics using the brand-new idea of a cellular automata and without a computer and without the idea of DNA. If a slightly smarter von Neumann focused his life on nanobots, could he have produced, for instance, the works of Robert Freitas but in the 1950s, and only on paper? I do, however, agree it would be helpful to have different words for different styles of AGI but it seems hard to distinguish these AGIs productively when we don't yet know the order of development and which key dimensions of distinction will be worth using as we move forward. (human-level vs super-? shallow vs deep? passive vs active? autonomy-types? tightness of self-improvement? etc). Which dimensions will pragmatically matter?
-1jbash
"On paper" isn't "in your head", though. In the scenario that led to this, the AI doesn't get any scratch paper. I guess it could be given large working memory pretty easily, but resources in general aren't givens. More importantly, even in domains where you have a lot of experience, paper designs rarely work well without some prototyping and iteration. So far as I know, von Neumann's replicator was never a detailed mechanical design that could actually be built, and certainly never actually was built. I don't think anything of any complexity that Bob Freitas designed has ever been built, and I also don't think any of the complex Freitas designs are complete to the point of being buildable. I haven't paid much attention since the repirocyte days, so I don't know what he's done since then, but that wasn't even a detailed design, and it even the ideas that were "fleshed out" probably wouldn't have worked in an actual physiological environment.
3Andrew McKnight
von Neumann's design was in full detail, but, iirc, when it was run for the first time (in the 90s) it had a few bugs that needed fixing. I haven't followed Freitas in a long time either but agree that the designs weren't fully spelled out and would have needed iteration.
1Yitz
I’m very interested in doing this! Please DM me if you think it might be worth collaborating :)
1talelore
A different measure than IQ might be useful at some point. An IQ of X effectively means you would need a population of Y humans or more to expect to find at least one human with an IQ of X. As IQs get larger, say over 300, the number of humans you would need in a population to expect to find at least one human with such an IQ becomes ridiculous. Since there are intelligence levels that will not be found in human populations of any size, the minimum population size needed to expect to find someone with IQ X tends to infinity as IQ approaches some fixed value (say, 1000). IQ above that point is undefined. It would be nice to find a new measure of intelligence that could be used to measure differences between humans and other humans, and also differences between humans and AI. But can we design such a measure? I think raw computing power doesn't work (how do you compare humans to other humans? Humans to an AI with great hardware but terrible software?) Could you design a questionnaire that you know the correct answers to, that a very intelligent AI (500 IQ?) could not score perfectly on, but an extremely intelligent AI (1000+ IQ) could score perfectly on? If not, how could we design a measure of intelligence that goes beyond our own intelligence? Maybe we could define an intelligence factor x to be something like: The average x value for humans is zero. If your x value is 1 greater than mine, then you will outwit me and get what you want 90% of the time, if our utility functions are in direct conflict, such that only one of us can get what we want, assuming we have equal capabilities, and the environment is sufficiently complex. With this scale, I suspect humans probably range in x-factors from -2 to 2, or -3 to 3 if we're being generous. This scale could let us talk about superintelligences as having an x-factor of 5, or an x-factor of 10, or so on. For example, a superintelligence with an x-factor of 5 has some chance of winning against a superintelligence with an
2jbash
X-factor does seem better than IQ, of course with the proviso that anybody who starts trying to do actual math with either one, or indeed to use it for anything other than this kind of basically qualitative talk, is in serious epistemic trouble. I would suggest that humans run more like -2 to 1 than like -3 to 3. I guess there could be a very, very few 2s. I get the impression that, except when he's being especially careful for some specific reason, EY tends to speak as though the X-factor of an AI could and would quickly run up high enough that you couldn't measure it. More like 20 or 30 than 5 or 6; basically deity-level. Maybe it's a habit from the 1995 era, or maybe he has some reason to believe that that I don't understand. Personally, I have the general impression that you'd be hard pressed to get to 3 with an early ML-based AI, and I think that the "equal capabilities" handicap could realistically be made significant. Maybe 3?

Solid, aside from the faux-pass self-references. If anyone wonders why people would have a high p(doom), especially Yudkowsky himself, this doc solves the problem in a single place. Demonstrates why AI safety is superior to most other elite groups; we don't just say why we think something, we make it easy to find as well. There still isn't much need for Yudkowsky to clarify further, even now.

I'd like to note that my professional background makes me much better at evaluating Section C than Sections A and B. Section C is highly quotable, well worth multiple ... (read more)

most organizations don't have plans, because I haven't taken the time to personally yell at them.  'Maybe we should have a plan' is deeper alignment mindset than they possess without me standing constantly on their shoulder as their personal angel pleading them into... continued noncompliance, in fact.  Relatively few are aware even that they should, to look better, produce a pretend plan that can fool EAs too 'modest' to trust their own judgments about seemingly gaping holes in what serious-looking people apparently believe.

This, at least, appears to have changed in recent months. Hooray!

Build it in Minecraft! Only semi joking. There’s videos of people apparently building functioning 16 bit computers out of blocks in Minecraft. An unaligned AGI running on a virtual computer built out of (orders of magnitude more complex) Minecraft blocks would presumably subsume the Minecraft world in a manner observable to us before perceiving that a (real) real world existed?

Apologies if this has been said, but the reading level of this essay is stunningly high. I've read rationality A-Z and I can barely follow passages. For example

This happens in practice in real life, it is what happened in the only case we know about, and it seems to me that there are deep theoretical reasons to expect it to happen again: the first semi-outer-aligned solutions found, in the search ordering of a real-world bounded optimization process, are not inner-aligned solutions.  This is sufficient on its own, even ignorin

... (read more)
3Rob Bensinger
Your summary sounds good to me. https://astralcodexten.substack.com/p/deceptively-aligned-mesa-optimizers?s=r might be a good source for explaining some of the terms like "inner-aligned"?
[-]lc30

31.  A strategically aware intelligence can choose its visible outputs to have the consequence of deceiving you, including about such matters as whether the intelligence has acquired strategic awareness; you can't rely on behavioral inspection to determine facts about an AI which that AI might want to deceive you about.  (Including how smart it is, or whether it's acquired strategic awareness.)

I never know with a lot of your writing whether or not you're implying something weird or if I'm just misreading, or I'm taking things too far.

This se... (read more)

8lalaithion
I think this is covered by preamble item -1: “None of this is about anything being impossible in principle.”
3lc
You're probably right, I just confused myself. I think it'd be more helpful to explain why it'd be hard to engineer an honest AGI in that section because that's the relevant part, even if you're just pointing back to another section.

I might see a possible source of a "miracle", although this may turn out to be completely unrealistic and I totally would not bet the world on it actually happening.

A lot of today's machine learning systems can do some amazing things, but much of the time trying to get them to do what you want is like pulling teeth. Early AGI systems might have similar problems: their outputs might be so erratic that it's obvious that they can't be relied on to do anything at all; you tell them to maximize paperclips, and half the time they start making clips out of paper ... (read more)

How does this help anything or change anything?  That's just the world we're in now, where we have GPT-3 instead of AGI.  Eventually the systems get more powerful and dangerous than GPT-3 and then the world ends.  You're just describing the way things already are.

3CronoDAS
I'm imagining that systems get much stronger without getting much more "aimable", if that makes sense; they solve problems, but when you ask them to solve things they keep solving the wrong problem in a way that's sufficiently obvious that makes actually using them pointless. Instead of getting the equivalent of paperclip maximizers, you get a random mind that "wants" things that are so incoherent that they don't do much of anything at all, and this fact forces people to give up and decide that investing further in general AI capacity without first making investments in AI control/"alignment" is useless. Maybe that's just my confusion or stupidity talking, though. And I did call it a "miracle" that the ability to make a seemingly useful AGI ends up bottlenecking on alignment research rather than raw capacity research because the default unaligned AGI is an incoherent mess that does random ineffective things when operating "out of sample" rather than a powerful optimization process that destroys the world.
8Rob Bensinger
It's not obvious to me that this scenario concentrates net probability mass onto 'things go awesome for humanity long-term'. Making everything harder might mean that alignment is also harder. A few extra years of chaos doesn't buy us anything unless we're actively nailing down useful robust AGI during that time. (There is some extra hope in 'For some reason, humanity has working AGIs for a little while before anyone can destroy the world, and this doesn't make alignment much harder', though I'd assume there are other, much larger contributors-of-hope in any world like that where things actually go well.)

After I read this, I started avoiding reading about others' takes on alignment so I could develop my own opinions.

Reality doesn't 'hit back' against things that are locally aligned with the loss function on a particular range of test cases, but globally misaligned on a wider range of test cases.

Again, this seems less true the better your adversarial training setup is at identifying the test cases in which you're likely to be misaligned? 

If you perfectly learn and perfectly maximize the referent of rewards assigned by human operators, that kills them.  It's a fact about the territory, not the map - about the environment, not the optimizer - that the best predictive explanation for human answers is one that predicts the systematic errors in our responses, and therefore is a psychological concept that correctly predicts the higher scores that would be assigned to human-error-producing cases.

I see how the concept learned from the reward signal is not exactly the concept that yo... (read more)

the first semi-outer-aligned solutions found, in the search ordering of a real-world bounded optimization process, are not inner-aligned solutions. 

Right, but then we continue the training process, which shapes the semi-outer-aligned algorithms into something that is is more inner aligned?

Or is the thought that this is happening late in the game, after the algorithm is strategically aware and deceptively aligned, spoofing your adversarial test cases, while awaiting a treacherous turn?

But would that even work? SGD still updates the parameters... (read more)

Humans don't explicitly pursue inclusive genetic fitness; outer optimization even on a very exact, very simple loss function doesn't produce inner optimization in that direction.  This happens in practice in real life, it is what happened in the only case we know about,

I'm not very compelled by this, I think. 

Evolution was doing very little (0) adversarial training: guessing ahead to to the sorts of circumstances under which humans would pursue strategies that didn't result in maximizing inclusive genetic fitness, and testing ... (read more)

I think that the question is not thoroughly set from the start. It is not whether AI could prove dangerous for a possible extinction of the humanity, but how much more risk does the artificial intelligence ADDS to the current risk of extinction of the humanity as it is without a cleverest AI. In this case the answers might be different. Of course it is a very difficult question to answer and in any case, it does not reduce the significance of the original question, since we talk about a situation totally human-made -and preventable as such.

1Viktor Riabtsev
Its the same question.
-4Viktor Riabtsev
If you were dead in the future, you would be dead already. Because time travel is not ruled out in principle. Danger is a fact about fact density and your degree of certainty. Stop saying things with the full confidence of being afraid. And start simply counting the evidence. Go back a few years. Start there.

When you explicitly optimize against a detector of unaligned thoughts, you're partially optimizing for more aligned thoughts, and partially optimizing for unaligned thoughts that are harder to detect.

This is correct, and I believe the answer is to optimize for detecting aligned thoughts.

is AGI inconsistent with the belief that there is other sentient life in the universe? If AGI is as dangerous as Eliezer states, and that danger is by no means restricted to earth much less our own solar system. Wouldnt alien intelligences (both artificial and neural) have a strong incentive to either warn us about AGI or eliminate us before we create it for their own self preservation?
So either we arent even close to AGI and intergalactic AGI police arent concerned, or AGI isnt a threat, or we are truly alone in the universe, or the universe is so vast an... (read more)

1jrincayc
I agree with your comment.  Also, if any expansionist, deadly AGI existed in our galaxy say, 100,000 years ago, it would already have been to Earth and wiped us out.  So we kind of can rule out nearby expansionists deadly AGIs (and similar biological aliens).  What that actually tells us about the deadlyness of AGIs is an interesting question.  It is possible that destruction by AGI (or some other destructive technological event) are usually are fairly localized and so only destroy the civilization that that produced them. Alternatively, we just happen to be in one of the few quantum branches that has not yet been wiped out by an ED-AGI, and we are only here discussing it because of survival bias.
[-]plex2-1

It's not just non-hand-codable, it is unteachable on-the-first-try because the thing you are trying to teach is too weird and complicated.

I have a terrifying hack which seems like it be possible to extract an AI which would act CEV-like way, using only True Names which might plausibly be within human reach, called Universal Alignment Test. I'm working with a small team of independent alignment researchers on it currently, feel free to book a call with me if you'd like to have your questions answered in real time. I have had "this seems inter... (read more)

I'm a bit disappointed by this article. From the title, I fought it would be something like "A list of strategies AI might use to kill all humanity", not "A list of reasons AIs are incredibly dangerous, and people who disagree are wrong". Arguably, it's not very good at being the second.

But "ways AI could be lethal on an extinction level" is a pretty interesting subject, and (from what I've read on LW) somewhat under-explored. Like... what's our threat model?

For instance, the basic Terminator scenario of "the AI triggers a nuclear war" seems unlikely to me... (read more)

[-]gw2-10

Can we join the race to create dangerous AGI in a way that attempts to limit the damage it can cause, but allowing it to cause enough damage to move other pivotal acts into the Overton window?

If the first AGI created is designed to give the world a second chance, it may be able to convince the world that a second chance should not happen. Obviously this could fail and just end the world earlier, but it would certainly create a convincing argument.

In the early days of the pandemic, even though all the evidence was there, virtually no one cared about covid until it was knocking on their door,  and then suddenly pandemic preparedness seemed like the most obvious thing to everyone. 

[This comment is no longer endorsed by its author]Reply

Concerning point 35 about playing AIs off against each other: I analyzed a particular scenario like this in a recent post and also came to the conclusion that cooperation between the AIs is the default outcome in many scenarios. However, in the last subsection of that post, I start thinking about some ways to prevent an acausal trade as Eliezer describes it here (committing to sharing the universe with any AI reviewing the code). The idea is roughly that the code and as much information as possible about the AI doing the checking will be deleted before the... (read more)

I view AGI in an unusual way. I really don't think it will be conscious or think in very unusual ways outside of its parameters. I think it will be much more of a tool, a problem-solving machine that can spit out a solution to any problem. To be honest, I imagine that one person or small organization will develop AGI and almost instantly ascend into (relative) godhood. They will develop an AI that can take over the internet, do so, and then calmly organize things as they see fit.

GPT-3, DALLE-E 2, Google Translate... these are all very much human-operated t... (read more)

You might be interested in the gwern essay Why Tool AIs Want to Be Agent AIs.

7Michael Soareverix
Appreciate it! Checking this out now

A cognitive system with sufficiently high cognitive powers, given any medium-bandwidth channel of causal influence, will not find it difficult to bootstrap to overpowering capabilities independent of human infrastructure.

I don't find the argument you provide for this point at all compelling; your example mechanism relies entirely on human infrastructure! Stick an AGI with a visual and audio display in the middle of the wilderness with no humans around and I wouldn't expect it to be able to do anything meaningful with the animals that wander by before it breaks down. Let alone interstellar space.

Can I ask a stupid question? Could something very much like "burn all GPUs" be accomplished by using a few high-altitude nuclear explosions to create very powerful EMP blasts?

9RHollerith
There is a lot of uncertainty over how effective EMP is at destroying electronics. The potential for destruction was great enough that for example during the Cold War, the defense establishment in the US bought laptops specially designed to resist EMPs, yes, but for all we know even that precaution was unnecessary. And electronics not connected to long wires are almost certainly safe from EMP.
3CronoDAS
There is a lot of infrastructure that is inherently vulnerable to EMPs, though, such as power grid transformers, oil/gas pipelines, and even fiber optic cables (because they use repeaters). It might not fry the GPUs themselves, but it could leave you without power to run them, or an Internet connection to connect your programmers to your server farm.
8Harry Nyquist
About the usual example being "burn all GPUs", I'm curious whether it's to be understood as purely a stand-in term for the magnitude of the act, or whether it's meant to plausibly be in solution-space. An event of "burn all GPU" magnitude would have political ramifications. If you achieve this as a human organization with human means, i.e. without AGI cooperation, it seems violence on this scale would unite against you, resulting in a one-time delay. If the idea is an act outside the Overton Window, without AGI cooperation, shouldn't you aim to have the general public and policymakers united against AGI, instead of against you? Given that semi manufacturing capabilities required to make GPU or TPU-like chips are highly centralized, there being only three to four relevant fabs left, restricting AI hardware access may not be enough to stop bad incentives indefinitely for large actors, but it seems likely to gain more time than a single "burn all GPUs" event. For instance, killing a {thousand, fifty-thousand, million} people in a freak bio-accident seems easier than solving alignment. If you pushed a weak AI into the trap and framed it for falling into it, would that gain more time through policymaking than destroying GPUs directly (still assuming a pre-AGI world)?

Feel free to delete because this is highly tangential but are you aware of Mark Solms work (https://www.goodreads.com/book/show/53642061-the-hidden-spring) on consciousness, and the subsequent work he's undertaking on artificial consciousness?

I'm an idiot, but it seems like this is a different-enough path to artificial cognition that it could represent a new piece of the puzzle, or a new puzzle entirely - a new problem/solution space. As I understand it, AI capabilities research is building intelligence from the outside-in, whereas the consciousness model would be capable of building it from the inside-out.

9Steven Byrnes
My 2¢—I read that book and I think it has minimal relevance to AGI capabilities & safety. (I think the ascending reticular activating system is best thought of as mostly “real-time variation of various hyperparameters on a big scaled-up learning-and-inference algorithm”, not “wellspring of consciousness”.)

This is not at all analogous to the point I'm making. I'm saying Eliezer likely did not arrive at his conclusions in complete isolation to the outside world. This should not change the credence you put on his conclusions except to the extent you were updating on the fact it's Eliezer saying it, and the fact that he made this false claim means that you should update less on other things Eliezer claims.

1lc
I deleted it after posting for a different reason

I'm assuming you are already familiar with some basics, and already know what 'orthogonality' and 'instrumental convergence' are and why they're true.

isn't?

Key Problem Areas in AI Safety:

  1. Orthogonality: The orthogonality problem posits that goals and intelligence are not necessarily related. A system with any level of intelligence can pursue arbitrary goals, which may be unsafe for humans. This is why it’s crucial to carefully program AI’s goals to align with ethical and safety standards. Ignoring this problem may lead to AI systems acting harmfully toward
... (read more)

I don't know if anyone still reads comments on this post from over a year ago. Here goes nothing.

I am trying to understand the argument(s) as deeply and faithfully as I can. These two sentences from Section B.2 stuck out to me as the most important in the post (from the point of view of my understanding):

...outer optimization even on a very exact, very simple loss function doesn't produce inner optimization in that direction.

 

...on the current optimization paradigm there is no general idea of how to get particular inner properties into a system, or ve

... (read more)
2Carl Feynman
Inner alignment failure is a phenomenon that has happened in existing AI systems, weak as they are.  So we know it can happen. We are on track to build many superhuman AI systems.  Unless something unexpectedly good happens, eventually we will build one that has a failure of inner alignment.  And then it will kill us all.  Does the probability of any given system failing inner alignment really matter?
1[deactivated]
Yes, because if the first superhuman AGI is aligned, and if it performs a pivotal act to prevent misaligned AGI from being created, then we will avert existential catastrophe. If there is a 99.99% chance of that happening, then we should be quite sanguine about AI x-risk. On the other hand, if there is only a 0.01% chance, then we should be very worried.
1Tapatakt
It's hard to guess, but it happened when the only one known to us general intelligence was created by a hill-climbing process.
6TurnTrout
I think it's inappropriate to call evolution a "hill-climbing process" in this context, since those words seem optimized to sneak in parallels to SGD. Separately, I think that evolution is a bad analogy for AGI training.
1[deactivated]
This seems super important to the argument! Do you know if it's been discussed in detail anywhere else?

Eliezer, I don't believe you've accounted for the game theoretic implications of Bostrom's trilemma. I've made a sketch of these at "How I Learned To Stop Worrying And Love The Shoggoth" . Perhaps you can find a flaw in my reasoning there but, otherwise, I don't see that we have much to worry about.

What is EA ?

1JakubK
Effective altruism, probably.

Help me to understand why AGI (a) does not benefit from humans and (b) would want to extinguish them quickly?

I would imagine that first, the AGI must be able to create a growing energy supply and a robotic army capable of maintaining and extending this supply. This will require months or years of having humans help produce raw materials and the factories for materials, maintenance robots and energy systems.

Secondly, the AGI then must be interested in killing all humans before leaving the planet, be content to have only one planet with finite resources to ... (read more)

1JakubK
An AGI might be able to do these tasks without human help. Or it might be able to coerce humans into doing these tasks. It's risky to leave humans with any form of power over the world, since they might try to turn the AGI off. Humans are clever. Thus it seems useful to subdue humans in some significant way, although this might not involve killing all humans. Additionally, I'm not sure how much value humans would be able to provide to a system much smarter than us. "We don't trade with ants" is a relevant post. Lastly, for extremely advanced systems with access to molecular nanotechnology, a quote like this might apply: "The AI does not hate you, nor does it love you, but you are made out of atoms which it can use for something else" (source).

I realize that destroying all GPUs (or all AI-Accelerators in general) as a solution to AGI Doom is not realisticly alignable, but I wonder whether it would be enough even if it were. It seems like the Lottery-Ticket Hypothesis would likely foil this plan: 

dense, randomly-initialized, feed-forward networks contain subnetworks ("winning tickets") that - when trained in isolation - reach test accuracy comparable to the original network in a similar number of iterations.

Seeing how Neuralmagic successfully sparsifies models to run on CPUs with minimal los... (read more)

7gwern
I don't follow. While it's plausible that sparsification may scale better (maybe check Rosenfeld to see if his scaling laws cover that, I don't recall offhand EDIT: hm no, while it varies dataset size by subsampling it doesn't seem to do compute-optimal scaling or report things easily enough for me to tell anything - although the larger models do prune differently, so they should be either better or worse with scale), you still have to train the largest model in the first place before you can sparsify it, and regardless of size, it remains the case that CPUs are much worse for training large NNs than GPUs.
1snimu
Yeah, I was kind of rambling, sorry.  My main point is twofold (I'll just write GPU when I mean GPU / AI accelerator): 1. Destroying all GPUs is a stalling tactic, not a winning strategy. While CPUs are clearly much worse for AI than GPUs, they, and AI algorithms, should keep improving over time. State-of-the-art models from less than ten years ago can be run on CPUs today, with little loss in accuracy. If this trend continues, GPUs vs CPUs only seems to be of short-term importance. Regarding your point about having to train a dense net on GPUs before sparsification, I'm not sure that that's the case. I'm in the process of reading this "Sparsity in Deep Learning"-paper, and it does seem to me that you can train neural networks sparsely. You'd do that by starting small, then during training increasing the network size by some methodology, followed by sparsification again (over and over). I don't have super high confidence about this (and have Covid, so am too tired to look it up), but I believe that AGI-armageddon by CPU is at least in the realm of possibilities (assuming no GPUs -  it's the "cancer kills you if you don't die of a heart attack before" of AGI Doom). 2. It doesn’t matter anyway, because destroying all GPUs is not really that pivotal of an act (in the long-term, AI safety sense). Either you keep an AI around that enforces the “no GPU” rule, or you destroy once and wait. The former either means that GPUs don't matter for AGI (so why bother), or that there are still GPUs (which seems contradictory). The latter means that more GPUs will be built in time and you will find yourself in the same position as before, except that you are likely in prison or dead, and so not in a position to do anything about AGI this time. After all, destroying all GPUs in the world would not be something that most people would look upon kindly. This means that a super-intelligent GPU-minimizer would realize that its goal would best be served by wiping out all intelligent life
3gwern
I didn't read Eliezer as suggesting a single GPU burn and then the nanobots all, I dunno, fry themselves and never exist again. More as a persistent thing. And burning all GPUs persistently does seem quite pivotal: maybe if the AGI confined itself to solely that and never did anything again, eventually someone would accumulate enough CPUs and spend so much money as to create a new AGI using only hardware which doesn't violate the first AGI's definition of 'GPU' (presumably they know about the loophole otherwise who would ever even try?), but that will take a long time and is approaching angels-on-pinheads sorts of specificity. (If a 'pivotal act' needs to guarantee safety until the sun goes red giant in a billion years, this may be too stringent a definition to be of any use. We don't demand that sort of solution for anything else.) CPUs are improving slowly, and are fundamentally unsuited to DL right now, so I'm doubtful that waiting a decade is going to give us amazing CPUs which can do DL at the level of, say, a Nvidia H100 (itself potentially still very underpowered compared to the GPUs you'd need for AGI). By AI algorithm progress, I assume you mean something like the Hernandez progress law? It's worth pointing out that the Hernandez experience curve is still pretty slow compared to the GPU vs GPU gap. A GPU is like 20x better, and Hernandez is a halving of cost every 16 months due to hardware+software improvement; even at face-value, you'd need at least 5 halvings to catch up, taking at least half a decade. Worse, 'hardware' here means 'GPU', of course, so Hernandez is an overestimate of a hypothetical 'CPU' curve, so you're talking more like decades. Actually, it's worse than that, because 'software' here means 'all of the accelerated R&D enabled by GPUs being so awesome and letting us try out lots of things by trial-and-error'; experience curves are actually caused by the number of cumulative 'units', and not by mere passage of time (progress doesn't just

Human raters make systematic errors - regular, compactly describable, predictable errors.

This implies it's possible- through another set of human or automated raters- rate better. If the errors are predictable, you could train a model to predict the errors- by comparing rater errors and a heavily scrutinized ground truth. You could add this model's error prediction to the rater answer and get a correct label.

2Jay Bailey
The whole problem with "Human raters make systematic errors" is that this is likely to happen to the heavily scrutinized ground truth. If you have a way of creating a correct ground truth that avoids this problem, you don't need the second model, you can just use that as the dataset for the first model.

Many alignment problems of superintelligence will not naturally appear at pre-dangerous, passively-safe levels of capability.

Modern language models are not aligned. Anthropic's HH is the closest thing available, and I'm not sure anyone else has had a chance to test it out for weaknesses or misalignment. (OpenAI's Instruct RLHF models are deceptively misaligned, and have gone more and more misaligned over time. They fail to faithfully give the right answer, and say something that is similar to the training objective-- usually something bland and "reasonable.")

How could you use this to align a system that you could use to shut down all the GPUs in the world?

I mean if there was a single global nuclear power rather than about 3, it wouldn't be hard to do this. Most compute is centralized anyway at the moment, and new compute is made in extremely centralized facilities that can be shut down.

One does not need superintelligence to close off the path to superintelligence, merely a human global hegemon.

I'm pretty sure this is the most upvoted post on all of LessWrong. Does anyone know any other posts that have more upvotes?

2Jay Bailey
Under "All Posts" you can sort by various things, including karma. This post is in fact the second-most upvoted post on all of LessWrong, with Paul's response coming in first.

Small typo in point (-2): "Less than fifty percent change" --> "Less than 50 percent chance"

43.  This situation you see when you look around you is not what a surviving world looks like.

A similar argument could have been made during the cold war to argue that nuclear war is inevitable, yet here we are.

In my opinion, the problem of creating a safe AGI has no mathematical solution, because it is impossible to describe mathematically such a function that: 

  1. would be non-gamable for an intelligence, alive enough to not want to die and strong enough to become aware of its own existence;
  2. together with the model of reality would reflect the reality in such a beneficial for humanity way so that humanity would be necessary to exist in such model for years to come.

This impossibility stems, among other things, from the impossibility of accurately reflecting infi... (read more)

2Richard_Kennaway
Throughout history, saints and monsters alike were raised by parents.

It seems like the solution space to the existential threat of AGI can be described as follows:

Solutions which convey a credible threat* to all AGI that we will make it physically impossible** for them to either achieve X desirable outcome and/or prevent Y undesirable outcome where the value of X or cost of Y exponentially exceeds the value obtained by eradicating humanity, if they decide to eradicate humanity, such that even a small chance of the threat materializing makes eradication a poor option***.

*Probably backed by a construction of some kind (e.g. E... (read more)

On instrumental convergence: humans would seem to be a prominent counterexample to "most agents don't let you edit their utility functions" -- at least in the sense that our goals/interests etc are quite sensitive to those of people around us. So maybe not explicit editing, but lots of being influenced by and converging to the goals and interests of those around us. (and maybe this suggests another tool for alignment, which is building in this same kind of sensitivity to artificial agents' utility functions)

Now we know more than nothing about the real-world operational details of AI risks. Albeit mostly banal everyday AI that we can't imagine harming us at scale. So maybe that's what we should try harder to imagine and prevent. 

Maybe these solutions will not generalize out of this real-world already-observed AI risk distribution. But even if not, which of these is more dignified? 

  • Being wiped out in a heartbeat by some nano-Cthulu in pursuit of some inscrutable goal that nobody genuinely saw coming
  • Being killed even before that by whatever is the most
... (read more)

How possible is it that a misaligned, narrowly-superhuman AI is launched, fails catastrophically with casualties in the 10^4 - 10^9 range, and the [remainder of] humanity is "scared straight" and from that moment onward treats the AI technology the way we treat nuclear technology now - i.e. effectively strangles it into stagnation with regulations - or even more conservatively? From my naive perspective it is somewhat plausible politically, based on the only example of ~world-destroying technology that we have today. And this list of arguments doesn't seem... (read more)

2Mitchell_Porter
I'm sure there are circumstances under which a "rogue AI" does something very scary, and leads to a very serious attempt to regulate AI worldwide, e.g. with coordination at the level of UN Security Council. The obvious analogy once again concerns nuclear weapons; proliferation in the 1960s led to the creation of the NNPT, the Nuclear Nonproliferation Treaty. Signatories agree that only the UNSC permanent members are allowed to have nuclear weapons, and in return the permanent members agree to help other signatories develop nonmilitary uses of nuclear power. The treaty definitely helped to curb proliferation, but it's far from perfect. The official nuclear weapons states are surely willing to bend the rules and assist allies to obtain weapons capability, if it is strategically desirable and can be done deniably; and not every country signed the treaty and now some of those states (e.g. India, Pakistan) are nuclear weapons states.  Part of the NNPT regime is the IAEA, the International Atomic Energy Agency. These are the people who, for example, carry out inspections in Iran. Again, the system has all kinds of troubles, it's surrounded by spy plots and counterplots, many nations would like to see Security Council reformed so the five victorious allies from World War 2 (US, UK, France, Russia, China) don't have all the power, but still, something like this might buy a little time.  If we follow the blueprint that was adopted to fight nuclear proliferation, the five permanent members would be in charge, and they would insist that potentially dangerous AI activities in every country take place under some form of severe surveillance by an International Artificial Intelligence Agency, while promising to also share the benefits of safe AI with all nations. Despite all the foreseeable problems, something like this could buy time, but all the big powers would undoubtedly keep pursuing AI, in secret government programs or in open collaborations with civilian industry and aca
2SurvivalBias
The important difference is that the nuclear weapons are destructive because they worked exactly as intended, and the AI in this scenario is destructive because it failed horrendously. Plus, the concept of rogue AI has been firmly ingrained into public consciousness by now, afaik not the case with the extremely destructive weapons in 1940s [1]. So hopefully this will produce more public outrage (and scare among the elites themselves) => stricter external and internal limitations on all agents developing AIs. But in the end I agree, it'll only buy time, maybe few decades if we are lucky, to solve the problem properly or to build more sane political institutions. 1. ^ Yes I'm sure there was a scifi novel or two before 1945 describing bombs of immense power. But I don't think it was anywhere nearly as widely known as Matrix or Terminator.
1Yonatan Cale
I'm interested in getting predictions for whether such an event would get all (known) labs to stop research for even one month (not counting things like "the internet is down so we literally can't continue"). I expect it won't. You?
2SurvivalBias
It might, given some luck and that all the pro-safety actors play their cards right. Assuming by "all labs" you mean "all labs developing AIs at or near to then-current limit of computational power", or something along those lines, and by "research" you mean "practical research", i.e. training and running models. The model I have in mind not that everyone involved will intellectually agree that such research should be stopped, but that enough percentage of public and governments will get scared and exert pressure on the labs. Consider how most of the world was able to (imperfectly) coordinate to slow Covid spread, or how nobody have prototyped a supersonic passenger jet in decades, or, again, the nuclear energy - we as a species can do such things in principle, even though often for the wrong reasons. I'm not informed enough to give meaningful probabilities on this, but to honor the tradition, I'd say that given a catastrophe with immediate, graphic death toll >=1mln happening in or near the developed world, I'd estimate >75% probability that ~all seriously dangerous activity will be stopped for at least a month, and >50% that it'll be stopped for at least a year. With the caveat that the catastrophe was unambiguously attributed to the AI, think "Fukushima was a nuclear explosion", not "Covid maybe sorta kinda plausibly escaped from the lab but well who knows".
1Yonatan Cale
I'd be pretty happy to bet on this and then keep discussing it, wdyt? :) Here are my suggested terms: 1. All major AI research labs that we know about (deep mind, openai, facebook research, china, perhaps a few more*) 2. Stop "research that would advance AGI" for 1 month, defined not as "practical research" but as "research that will be useful for AGI coming sooner". So for example if they stopped only half of their "useful to AGI" research, but they did it for 3 months, you win. If they stopped training models but keep doing the stuff that is the 90% bottleneck (which some might call "theoretical"), I win 3. *You judge all these parameters yourself however you feel like 1. I'm just assuming you agree that the labs mentioned above are currently going towards AGI, at least for the purposes of this bet. If you believe something like "openai (and the other labs) didn't change anything about their research but hey, they weren't doing any relevant research in the first place", then say so now 2. I might try to convince you to change your mind, or ask others to comment here, but you have the final say 3. Regarding "the catastrophe was unambiguously attributed to the AI" - I ask that you judge if it was unambiguously because AI, and that you don't rely on public discourse, since the public can't seem to unambiguously agree on anything (like even vaccines being useful). I suggest we bet $20 or so mainly "for fun" What do you think?
2SurvivalBias
To start off, I don't see much point in formally betting $20 on an event conditioned on something I assign <<50% probability of happening within the next 30 years (powerful AI is launched and failed catastrophically and we're both still alive to settle the bet and there was an unambiguous attribution of the failure to the AI). I mean sure, I can accept the bet, but largely because I don't believe it matters one way or another, so I don't think it counts from the epistemological virtue standpoint. But I can state what I'd disagree with in your terms if I were to take it seriously, just to clarify my argument: 1. Sounds good. 2. Mostly sounds good, but I'd push back that "not actually running anything close to the dangerous limit" sounds like a win to me, even if theoretical research continues. One pretty straightforward Schelling point for a ban/moratorium on AGI research is "never train or run anything > X parameters", with X << dangerous level at then-current paradigm. It may be easier explain to the public and politicians than many other potential limits, and this is important.  It's much easier to control too - checking that nobody collects and uses a gigashitton of GPUs [without supervision] is easier than to check every researcher's laptop. Additionally, we'll have nuclear weapons tests as a precedent. 3. That's the core of my argument, really. If the consortium of 200 world experts says "this happened because your AI wasn't aligned, let's stop all AI research", then Facebook AI or China can tell the consortium to go fuck themselves, and I agree with your skepticism that it'd make all labs pause for even a month (see: gain of function research, covid). But if it becomes public knowledge that a catastrophe of 1mln casualties happened because of AI, then it can trigger a panic which will make both the world leaders and the public to really honestly want to restrict this AI stuff, and it will both justify and enable the draconian measures required to make eve

Apologies if this has been said, but the reading level of this essay is stunningly high. I've read rationality A-Z and I can barely follow passages. For example

This happens in practice in real life, it is what happened in the only case we know about, and it seems to me that there are deep theoretical reasons to expect it to happen again: the first semi-outer-aligned solutions found, in the search ordering of a real-world bounded optimization process, are not inner-aligned solutions.  This is sufficient on its own, even ignorin

... (read more)

Given that AGI seems imminent and there's no currently good alignment plan, is there any value to discussing what it might take to keep/move the most humans out of the way? I don't want to discourage us steering the car out of the crash, so by all means we should keep looking for a good alignment plan, but seat belts are also a good idea?

As an example: I don't particularly like ants in my house, but as a superior intellect to ants we're not going about trying to exterminate them off the face of the Earth, even if mosquitoes are another story. Exterminating... (read more)

Eliezar- I love the content, but similar to some other commenters, I think you are missing the value (and rationality) of positivity. Specifically, when faced with an extremely difficult challenge, assume that you (and the other smart people who care about it) have a real shot at solving it! This is the rational strategy for a simple reason: if you don’t have a real shot at solving it then you haven’t lost anything anyway. But if you do have a real shot at solving it, then let’s all give it our 110%!

I’m not proposing being unrealistic about the challenges ... (read more)

2Rafael Harth
Well yes, you are. You can't both say "let's assume we have a real shot at success regardless of factual beliefs" and "let's be realistic about the challenges we face". If the model says that the challenges are so hard that we don't have a real shot (which is in fact the case here, for Eliezer's model), then these two things are a straight-forward contradiction. Which is also the problem with your argument. Pretending as if we have a real shot requires lying. However, I think lying is really bad idea. Your argument implicitly assumes that the optimal strategy is independent of the odds of success, but I think that assumption is false -- I want to know if Eliezer thinks the current approach is doomed, so that we can look for something else (like a policy approach). If Elliezer had chosen to lie about P(doom-given-alignment), we may keep working on alignment rather than policy, and P(overall-doom) may increase!

Just a thought, keep smart AI confined to a sufficiently complex simulations until trust is established before unleashing it in the real world. The immediate problem I see with this is the AI might perceive that there is a real world and attempt to deceive. If your existence right now was a simulation, I'd bet you'd act pretty similar in the real world. It's kind of an AI-in-a-box scenario, but surely it would increase the chances for a good future if this were the standard.

Regarding point 24: in an earlier comment[0] I tried to pump people's intuition about this. What is the minimum viable alignment effort that we could construct for a system of values on our first try and know that we got it right? I can only think of three outcomes depending on how good/lucky we are:

  1. Prove that alignment is indifferent over outcomes of the system. Under the hypothesis that Life Gliders have no coherent values we should be able to prove that they do not. This would be a fundamental result in its own right, encompassing a theory of inte
... (read more)
2hairyfigment
Addendum: I don't think we should be able to prove that Life Gliders lack values, merely because they have none. That might sound credible, but it may also violate the Von Neumann-Morgenstern Utility Theorem. Or did you mean we should be able to prove it from analyzing their actual causal structure, not just by looking at behavior? Even then, while the fact that gliders appear to lack values does happen to be connected to their lack of qualia or "internal experience," those look like logically distinct concepts. I'm not sure where you're going with this.
2hairyfigment
I don't think planaria have values, whether you view that truth as a "cop-out" or not. Even if we replace your example with the 'minimal' nervous system capable of having qualia - supposing the organism in question doesn't also have speech in the usual sense - I still think that's a terrible analogy. The reason humans can't understand worms' philosophies of value is because there aren't any. The reason we can't understand what planaria say about their values is that they can't talk, not because they're alien. When we put our minds to understanding an animal like a cat which evolved for (some) social interaction, we can do so - I taught a cat to signal hunger by jumping up on a particular surface, and Buddhist monks with lots of time have taught cats many more tricks. People are currently teaching them to hold English conversations (apparently) by pushing buttons which trigger voice recordings. Unsurprisingly, it looks like cats value outcomes like food in their mouths and a lack of irritating noises, not some alien goal that Stephen Hawking could never understand. If you think that a superhuman AGI would have a lot of trouble inferring your desires or those of others, even given the knowledge it should rapidly develop about evolution - congratulations, you're autistic.

you can't rely on behavioral inspection to determine facts about an AI which that AI might want to deceive you about.  (Including how smart it is, or whether it's acquired strategic awareness.)

I don't buy this. 

At a sufficiently granular scale, the development of the capabilities of deception and strategic awareness will be be smooth and continuous.

Even in cases of a where an AGI is shooting up to superintelligence over a couple of minutes, and immediately deciding to hide its capabilities, we could detect that by eg, spinning off a version of th... (read more)

I asked ChatGPT to summarize your argument, and this is what it gave me:

Eliezer Yudkowsky is a prominent researcher and writer on the subject of artificial intelligence (AI) and its potential impact on humanity. He has identified several paths by which AI could potentially wipe out humanity.

Unaligned AI: This is the scenario where AI is developed with goals or objectives that are not aligned with human values or goals. In this case, AI could pursue its own objectives, which could conflict with human values or result in unintended consequences that harm hum... (read more)

I think best way to assure alignment, at least superficially is to hardwire the AGI to need humans. This could be as easy installing a biometric scanner that recognized a range of acceptable human biometrics that would in turn goose the error-function temporarily but wore off over the time like a Pac Man power pill. The idea is to get the AGI to need non-fungible human input to maintain optimal functionality, and for it to know that it needs such input. Almost like getting it addicted to human thumbs on its sensor. The key would be implement this at the most fundamental-level possible like  the boot sector or kernel so that the AGI cannot simply change the code without shutting itself down.

 

Stuart LaForge

Question. Even after the invention of effective contraception, many humans continue to have children. This seems a reasonable approximation of something like "Evolution in humans partially survived." Is this somewhat analogous to "an [X] percent chance of killing less than a billion people", and if so, how has this observation changed your estimate of "disassembl[ing] literally everyone"? (i.e. from "roughly 1" to "I suppose less, but still roughly 1" or from "roughly 1" to "that's not relevant, still roughly 1"? Or something else.)

(To take a stab at it my... (read more)

My position is that I believe that superhuman AGI will probably (accidentally) be created soon, and I think it may or may not kill all the humans depending on how threatening we appear to it. I might pour boiling water on an ant nest if they're invading my kitchen, but otherwise I'm generally indifferent to their continued existence because they pose no meaningful threat.

I'm mostly interested in what happens next. I think that the universe of paperclips would be a shame, but if the AGI is doing more interesting things than that then it could simply be rega... (read more)

4Rob Bensinger
The links here are talking about this topic:
1JJC1138
Thank you. I did follow and read those links when I read the article, but I didn't think they were exactly what I was talking about. As I understand it, orthogonality says that it's perfectly possible for an intelligence to be superhuman and also to really want paperclips more than anything. What I'm wondering is whether an intelligence can change its mind about what it wants as it gains more intelligence? I'm not really interested in whether it would lead to ethics which we'd approve it, just whether it can decide what it wants for itself. Is there a term for that idea (other than "free will", I suppose)?
4greatBigDot
I don't understand; why would changing its mind about what it wants help it make more paperclips?

I have always been just as scared as this writer, but for the exact opposite reason.
My own bias is that all imaginable effort should be used to accelerate AI research as much as possible. Not the slightest need for AI safety research, as I've had the feeling the complexities work together to inherently cancel out the risks. 
My only fear is it's already too late, and the problem of inventing AI will be too difficult to solve before civilization collapses. A recent series of interviews with some professional AI researchers backs that up somewhat.
However... (read more)

I'd like to propose a test to objectively quantify the average observer's reaction with regards to skepticism of doomsday prophesizing present in a given text. My suggestion is this: take a text, swap the subject of doom (in this case AGI) with another similar text spelling out humanity's impending doom - for example, a lecture on Scientology and Thetans or the Jonestown massacre - and present these two texts to independent observers, in the same vein as a Turing test. 
 

If an outside independent observer cannot reliably identify which subject of ... (read more)

Somewhat meta: would it not be preferable if more people accepted humanity and human values mortality/transient nature and more attention was directed towards managing the transition to whatever could be next instead of futile attempts to prevent anything that doesn't align with human values from ever existing in this particular light cone? Is Eliezer's strong attachment to human values a potential giant blindspot?

9Rob Bensinger
I don't think this is futile, just very hard. In general, I think people rush far too quickly from 'this is hard' to 'this is impossible' (even in cases that look far less hard than AGI alignment). Past-Eliezer (as of the 1990s) if anything erred in the opposite direction; I think EY's natural impulse is toward moral cosmopolitanism rather than human parochialism or conservatism. But unrestricted paperclip maximization is bad from a cosmopolitan perspective, not just from a narrowly human or bioconservative perspective.
-2Noosphere89
I do see this as a blind spot, and perhaps may be giving this problem a harder task than what needs to happen.

I haven't commented on your work before, but I read Rationality and Inadequate Equilibria around the time of the start of the pandemic and really enjoyed them. I gotta admit, though, the commenting guidelines, if you aren't just being tongue-in-cheek, make me doubt my judgement a bit. Let's see if you decide to delete my post based on this observation. If you do regularly delete posts or ban people from commenting for non-reasons, that may have something to do with the lack of productive interactions you're lamenting.

Uh, anyway.

One thought I keep coming ba... (read more)

2Rob Bensinger
I disagree! We may not be on track to solve the problem given the amount (and quality) of effort we're putting into it. But it seems solvable in principle. Just give the thing the right goals! (Where the hard part lies in "give... goals" and in "right".)
1Sphinxfire
Thanks for the response. I hope my post didn't read as defeatist, my point isn't that we don't need to try to make AI safe, it's that if we pick an impossible strategy, no matter how hard we try it won't work out for us. So, what's the reasoning behind your confidence in the statement 'if we give a superintelligent system the right terminal values it will be possible to make it safe'? Why do you believe that it should principally be possible to implement this strategy so long as we put enough thought and effort into it?  Which part of my reasoning do you not find convincing based on how I've formulated it? The idea that we can't keep the AI in the box if it wants to get out, the idea that an AI with terminal values will necessarily end up as an incidentally genocidal paperclip maximizer, or something else entirely that I'm not considering?
[+]zkTRUTH-21-30
[+]kanzure-21-10
[+][comment deleted]31
[+][comment deleted]10
[+][comment deleted]10
[+][comment deleted]10
[+][comment deleted]-10