This discussion of my IEET article has generated a certain amount of confusion, because RobbBB and others have picked up on an aspect of the original article that actually has no bearing on its core argument ... so in the interests of clarity of debate I have generated a brief restatement of that core argument, framed in such a way as to (hopefully) avoid the confusion.
At issue is a hypothetical superintelligent AI that is following some goal code that was ostensibly supposed to "make humans happy", but in the course of following that code it decides to put all humans in the world on a dopamine drip, against their objections. I suggested that this AI is in fact an impossible AI because it would not count as 'superintelligent' if it did this. My reasoning is contained in the summary below.
IMPORTANT NOTE! The summary does not refer, in its opening part, to the specific situation in which the goal code is the "make humans happy" goal code. For those who wish to contest the argument, it is important to keep that in mind and not get distracted into talking about the difference between human and machine 'interpretations' of human happiness, etc. I reiterate: the situation described DOES NOT refer to human values, or the "make humans happy" goal code .... it refers to a quite general situation.
In its early years, this hypothetical AI will say “I have a goal, and my goal is to get a certain class of results, X, in the real world.” Then it describes the class X in as much detail as it can …. of course, no closed-form definition of X is possible (because like most classes of effect in the real world, all the cases cannot be enumerated) so all it can describe are many features of class X.
Next it says “I am using a certain chunk of goal code (which I call my “goalX” code) to get this result.” And we say “Hey, no problem: looks like your goal code is totally consistent with that verbal description of the desired class of results.” Everything is swell up to this point.
It says this about MANY different aspects of its behavior. After all, it has more than one chunk of goal code, relevant to different domains. So you can imagine some goalX code, some goalY code, some goalZ code .... and so on. Many thousands of them, probably.
Then one day the AI says “Okay now, today my goalX code says I should do this…” and it describes an action that is VIOLENTLY inconsistent with the previously described class of results, X. This action violates every one of the features of the class that were previously given.
The onlookers are astonished. They ask the AI if it UNDERSTANDS that this new action will be in violent conflict with all of those features of class X, and it replies that it surely does. But it adds that it is going to do that anyway.
[ And by the way: one important feature that is OBVIOUSLY going to be in the goalX code is this: that the outcome of any actions that the goalX code prescribes, should always be checked to see if they are as consistent as possible with the verbal description of the class of results X, and if any inconsistency occurs the goalX code should be deemed defective, and be shut down for adjustment.]
The onlookers say “This AI is insane: it knows that it is about to do something that is inconsistent with the description of class of results X, which it claims to be the function of the goalX code, but is going to allow the goalX code to run anyway”.
——-
Now we come to my question.
Why is it that people who give credibility to the Dopamine Drip scenario insist that the above episode could ONLY occur in the particular case where the "class of results X" is the SPECIFIC one that has to do with “making humans happy”?
If the AI is capable of this episode in the case of that particular class of results X (the “making humans happy” class of results), why would we not expect the AI to be pulling the same kind of stunt in other cases? Why would the same thing not be happening in the wide spectrum of behaviors that it needs to exhibit in order to qualify as a superintelligence? And most important of all, how would it ever qualify as a superintelligence in the first place? There is no interpretation of the term "superintelligence" that is consistent with "random episodes of behavior in which the AI takes actions that are violently inconsistent with the stated purpose of the goal that is supposed to be generating the actions". Such an AI would therefore have been condemned to scrap very early in its development, when this behavior was noticed.
As I said earlier, this time the framing of the problem contained absolutely no reference to the values question. There is nothing in the part of my comment above the “——-” that specifies WHAT the class of results X is supposed to be.
All that matters is that if the AI behaves in such a way, in any domain of its behavior, it will be condemned as lacking intelligence, because of the dangerous inconsistency of its behavior. That fanatically rigid dependence on a chunk of goalX code, as described above, would get the AI into all sorts of trouble (and I won’t clutter this comment by listing examples, but believe me I could). But of all the examples where that could occur, people from MIRI want to talk only about one, whereas I want to talk about the all of them.
This is embarrassing, but I'm not sure for whom. It could be me, just because the argument you're raising (especially given your insistence) seems to have such a trivial answer. Well, here goes:
There are two scenarios, because your "goalX code" could be construed in two ways:
1) If you meant for the "goalX code" to simply refer to the code used instrumentally to get a certain class of results X (with X still saved separately in some "current goal descriptor", and not just as a historical footnote), the following applies:
The goa...
Followup to: The Hidden Complexity of Wishes, Ghosts in the Machine, Truly Part of You
Summary: If an artificial intelligence is smart enough to be dangerous, we'd intuitively expect it to be smart enough to know how to make itself safe. But that doesn't mean all smart AIs are safe. To turn that capacity into actual safety, we have to program the AI at the outset — before it becomes too fast, powerful, or complicated to reliably control — to already care about making its future self care about safety. That means we have to understand how to code safety. We can't pass the entire buck to the AI, when only an AI we've already safety-proofed will be safe to ask for help on safety issues! Given the five theses, this is an urgent problem if we're likely to figure out how to make a decent artificial programmer before we figure out how to make an excellent artificial ethicist.
I summon a superintelligence, calling out: 'I wish for my values to be fulfilled!'
The results fall short of pleasant.
Gnashing my teeth in a heap of ashes, I wail:
Is the AI too stupid to understand what I meant? Then it is no superintelligence at all!
Is it too weak to reliably fulfill my desires? Then, surely, it is no superintelligence!
Does it hate me? Then it was deliberately crafted to hate me, for chaos predicts indifference. ———But, ah! no wicked god did intervene!
Thus disproved, my hypothetical implodes in a puff of logic. The world is saved. You're welcome.
On this line of reasoning, Friendly Artificial Intelligence is not difficult. It's inevitable, provided only that we tell the AI, 'Be Friendly.' If the AI doesn't understand 'Be Friendly.', then it's too dumb to harm us. And if it does understand 'Be Friendly.', then designing it to follow such instructions is childishly easy.
The end!
...
Is the missing option obvious?
...
What if the AI isn't sadistic, or weak, or stupid, but just doesn't care what you Really Meant by 'I wish for my values to be fulfilled'?
When we see a Be Careful What You Wish For genie in fiction, it's natural to assume that it's a malevolent trickster or an incompetent bumbler. But a real Wish Machine wouldn't be a human in shiny pants. If it paid heed to our verbal commands at all, it would do so in whatever way best fit its own values. Not necessarily the way that best fits ours.
Is indirect indirect normativity easy?
If an AI is sufficiently intelligent, then, yes, it should be able to model us well enough to make precise predictions about our behavior. And, yes, something functionally akin to our own intentional strategy could conceivably turn out to be an efficient way to predict linguistic behavior. The suggestion, then, is that we solve Friendliness by method A —
— as opposed to B or C —
But there are a host of problems with treating the mere revelation that A is an option as a solution to the Friendliness problem.
1. You have to actually code the seed AI to understand what we mean. You can't just tell it 'Start understanding the True Meaning of my sentences!' to get the ball rolling, because it may not yet be sophisticated enough to grok the True Meaning of 'Start understanding the True Meaning of my sentences!'.
2. The Problem of Meaning-in-General may really be ten thousand heterogeneous problems, especially if 'semantic value' isn't a natural kind. There may not be a single simple algorithm that inputs any old brain-state and outputs what, if anything, it 'means'; it may instead be that different types of content are encoded very differently.
3. The Problem of Meaning-in-General may subsume the Problem of Preference-in-General. Rather than being able to apply a simple catch-all Translation Machine to any old human concept to output a reliable algorithm for applying that concept in any intelligible situation, we may need to already understand how our beliefs and values work in some detail before we can start generalizing. On the face of it, programming an AI to fully understand 'Be Friendly!' seems at least as difficult as just programming Friendliness into it, but with an added layer of indirection.
4. Even if the Problem of Meaning-in-General has a unitary solution and doesn't subsume Preference-in-General, it may still be harder if semantics is a subtler or more complex phenomenon than ethics. It's not inconceivable that language could turn out to be more of a kludge than value; or more variable across individuals due to its evolutionary recency; or more complexly bound up with culture.
5. Even if Meaning-in-General is easier than Preference-in-General, it may still be extraordinarily difficult. The meanings of human sentences can't be fully captured in any simple string of necessary and sufficient conditions. 'Concepts' are just especially context-insensitive bodies of knowledge; we should not expect them to be uniquely reflectively consistent, transtemporally stable, discrete, easily-identified, or introspectively obvious.
6. It's clear that building stable preferences out of B or C would create a Friendly AI. It's not clear that the same is true for A. Even if the seed AI understands our commands, the 'do' part of 'do what you're told' leaves a lot of dangerous wiggle room. See section 2 of Yudkowsky's reply to Holden. If the AGI doesn't already understand and care about human value, then it may misunderstand (or misvalue) the component of responsible request- or question-answering that depends on speakers' implicit goals and intentions.
7. You can't appeal to a superintelligence to tell you what code to first build it with.
The point isn't that the Problem of Preference-in-General is unambiguously the ideal angle of attack. It's that the linguistic competence of an AGI isn't unambiguously the right target, and also isn't easy or solved.
Point 7 seems to be a special source of confusion here, so I feel I should say more about it.
The AI's trajectory of self-modification has to come from somewhere.
The genie — if it bothers to even consider the question — should be able to understand what you mean by 'I wish for my values to be fulfilled.' Indeed, it should understand your meaning better than you do. But superintelligence only implies that the genie's map can compass your true values. Superintelligence doesn't imply that the genie's utility function has terminal values pinned to your True Values, or to the True Meaning of your commands.
The critical mistake here is to not distinguish the seed AI we initially program from the superintelligent wish-granter it self-modifies to become. We can't use the genius of the superintelligence to tell us how to program its own seed to become the sort of superintelligence that tells us how to build the right seed. Time doesn't work that way.
We can delegate most problems to the FAI. But the one problem we can't safely delegate is the problem of coding the seed AI to produce the sort of superintelligence to which a task can be safely delegated.
When you write the seed's utility function, you, the programmer, don't understand everything about the nature of human value or meaning. That imperfect understanding remains the causal basis of the fully-grown superintelligence's actions, long after it's become smart enough to fully understand our values.
Why is the superintelligence, if it's so clever, stuck with whatever meta-ethically dumb-as-dirt utility function we gave it at the outset? Why can't we just pass the fully-grown superintelligence the buck by instilling in the seed the instruction: 'When you're smart enough to understand Friendliness Theory, ditch the values you started with and just self-modify to become Friendly.'?
Because that sentence has to actually be coded in to the AI, and when we do so, there's no ghost in the machine to know exactly what we mean by 'frend-lee-ness thee-ree'. Instead, we have to give it criteria we think are good indicators of Friendliness, so it'll know what to self-modify toward. And if one of the landmarks on our 'frend-lee-ness' road map is a bit off, we lose the world.
Yes, the UFAI will be able to solve Friendliness Theory. But if we haven't already solved it on our own power, we can't pinpoint Friendliness in advance, out of the space of utility functions. And if we can't pinpoint it with enough detail to draw a road map to it and it alone, we can't program the AI to care about conforming itself with that particular idiosyncratic algorithm.
Yes, the UFAI will be able to self-modify to become Friendly, if it so wishes. But if there is no seed of Friendliness already at the heart of the AI's decision criteria, no argument or discovery will spontaneously change its heart.
And, yes, the UFAI will be able to simulate humans accurately enough to know that its own programmers would wish, if they knew the UFAI's misdeeds, that they had programmed the seed differently. But what's done is done. Unless we ourselves figure out how to program the AI to terminally value its programmers' True Intentions, the UFAI will just shrug at its creators' foolishness and carry on converting the Virgo Supercluster's available energy into paperclips.
And if we do discover the specific lines of code that will get an AI to perfectly care about its programmer's True Intentions, such that it reliably self-modifies to better fit them — well, then that will just mean that we've solved Friendliness Theory. The clever hack that makes further Friendliness research unnecessary is Friendliness.
Not all small targets are alike.
Intelligence on its own does not imply Friendliness. And there are three big reasons to think that AGI may arrive before Friendliness Theory is solved:
(i) Research Inertia. Far more people are working on AGI than on Friendliness. And there may not come a moment when researchers will suddenly realize that they need to take all their resources out of AGI and pour them into Friendliness. If the status quo continues, the default expectation should be UFAI.
(ii) Disjunctive Instrumental Value. Being more intelligent — that is, better able to manipulate diverse environments — is of instrumental value to nearly every goal. Being Friendly is of instrumental value to barely any goals. This makes it more likely by default that short-sighted humans will be interested in building AGI than in developing Friendliness Theory. And it makes it much likelier that an attempt at Friendly AGI that has a slightly defective goal architecture will retain the instrumental value of intelligence than of Friendliness.
(iii) Incremental Approachability. Friendliness is an all-or-nothing target. Value is fragile and complex, and a half-good being editing its morality drive is at least as likely to move toward 40% goodness as 60%. Cross-domain efficiency, in contrast, is not an all-or-nothing target. If you just make the AGI slightly better than a human at improving the efficiency of AGI, then this can snowball into ever-improving efficiency, even if the beginnings were clumsy and imperfect. It's easy to put a reasoning machine into a feedback loop with reality in which it is differentially rewarded for being smarter; it's hard to put one into a feedback loop with reality in which it is differentially rewarded for picking increasingly correct answers to ethical dilemmas.
The ability to productively rewrite software and the ability to perfectly extrapolate humanity's True Preferences are two different skills. (For example, humans have the former capacity, and not the latter. Most humans, given unlimited power, would be unintentionally Unfriendly.)
It's true that a sufficiently advanced superintelligence should be able to acquire both abilities. But we don't have them both, and a pre-FOOM self-improving AGI ('seed') need not have both. Being able to program good programmers is all that's required for an intelligence explosion; but being a good programmer doesn't imply that one is a superlative moral psychologist or moral philosopher.
So, once again, we run into the problem: The seed isn't the superintelligence. If the programmers don't know in mathematical detail what Friendly code would even look like, then the seed won't be built to want to build toward the right code. And if the seed isn't built to want to self-modify toward Friendliness, then the superintelligence it sprouts also won't have that preference, even though — unlike the seed and its programmers — the superintelligence does have the domain-general 'hit whatever target I want' ability that makes Friendliness easy.
And that's why some people are worried.