The important thing for alignment work isn't the median prediction; if we had an alignment solution just by then, we'd have a 50% chance of dying from that lack.

I think the biggest takeaway is that nobody has a very precise and reliable prediction, so if we want to have good alignment plans in advance of AGI, we'd better get cracking.

I think Daniel's estimate does include a pretty specific and plausible model of a path to AGI, so I take his the most seriously. My model of possible AGI architectures requires even less compute than his, but I think the Hofstadter principle applies to AGI development if not compute progress.

Estimates in the absence of gears-level models of AGI seem much more uncertain, which might be why Ajeya and Ege's have much wider distributions.

That all makes sense. To expand a little more on some of the logic:

It seems like the outcome of a partial pause rests in part on whether that would tend to put people in the lead of the AGI race who are more or less safety-concerned.

I think it's nontrivial that we currently have three teams in the lead who all appear to honestly take the risks very seriously, and changing that might be a very bad idea.

On the other hand, the argument for alignment risks is quite strong, and we might expect more people to take the risks more seriously as those arguments diffuse. This might not happen if polarization becomes a large factor in beliefs on AGI risk. The evidence for climate change was also pretty strong, but we saw half of America believe in it less, not more, as evidence mounted. The lines of polarization would be different in this case, but I'm afraid it could happen. I outlined that case a little in AI scares and changing public beliefs

In that case, I think a partial pause would have a negative expected value, as the current lead decayed, and more people who believe in risks less get into the lead by circumventing the pause.

This makes me highly unsure if a pause would be net-positive. Having alignment solutions won't help if they're not implemented because the taxes are too high.

The creation of compute overhang is another reason to worry about a pause. It's highly uncertain how far we are from making adequate compute for AGI affordable to individuals. Algorithms and compute will keep getting better during a pause. So will theory of AGI, along with theory of alignment.

This puts me, and I think the alignment community at large, in a very uncomfortable position of not knowing whether a realistic pause would be helpful.

It does seem clear that creating mechanisms and political will for a pause are a good idea.

Advocating for more safety work also seems clear cut.

To this end, I think it's true that you create more political capitol by successfully pushing for policy.

A pause now would create even more capitol, but it's also less likely to be a win, and it could wind up creating polarization and so costing rather than creating capitol. It's harder to argue for a pause now when even most alignment folks think we're years from AGI.

So perhaps the low-hanging fruit is pushing for voluntary RSPs, and government funding for safety work. These are clear improvements, and likely to be wins that create capitol for a pause as we get closer to AGI.

There's a lot of uncertainty here, and that's uncomfortable. More discussion like this should help resolve that uncertainty, and thereby help clarify and unify the collective will of the safety community.

That is indeed a lot of points. Let me try to parse them and respond, because I think this discussion is critically important.

Point 1: overhang.

Your first two paragraphs seem to be pointing to downsides of progress, and saying that it would be better if nobody made that progress. I agree. We don't have guaranteed methods of alignment, and I think our odds of survival would be much better if everyone went way slower on developing AGI.

The standard thinking, which could use more inspection, but which I agree with, is that this is simply not going to happen. Individuals that decide to step aside are slowing progress only slightly. This leaves compute overhang that someone else is going to take advantage of, with nearly the competence, and only slightly slower. Those individuals who pick up the banner and create AGI will not be infinitely reckless, but the faster progress from that overhang will make whatever level of caution they have less effective.

This is a separate argument from regulation. Adequate regulation will slow progress universally, rather than leaving it up to the wisdom and conscience of every individual who might decide to develop AGI.

I don't think it's impossible to slow and meter progress so that overhang isn't an issue. But I think it is effectively even harder than alignment. We have decent suggestions on the table for alignment now, and as far as I know, no equally promising suggestions for getting everyone (and it does take almost everyone coordinating) to pass up the immense opportunities offered by capabilities overhangs.

Point 2: Are LLMs safer than other approaches?

I agree that this is a questionable proposition. I think it's worth questioning. Aiming progress at easier-to-align approaches seems highly worthwhile.

I agree that an LLM may have something like a mind inside. I think current versions are almost certainly too dumb to be existentially dangerous (at least directly - if a facebook algorithm can nearly cause an insurrection, who knows what dangerous side effects any AI can have).

I'm less worried about GPT10 playing a superintelligent, Waluigi-collapsed villain than I am about a GPT6 that has been amplified to agency, situational awareness, and weak superintelligence by scaffolding it into something like a cognitive architecture. I think this type of advance is inevitable. ChatGPT extensions and Bing Chat both use internal prompting to boost intelligence, and approaches like SmartGPT and Tree of Thoughts massively improve benchmark results over the base LLM.

Fortunately, this direction also has huge advantages for alignment. It has a very low alignment tax, since you give them additional goals in natural language, like "support human empowerment" or whatever the SOTA alignment goal is. And they have vastly better interpretability since they're at least summarizing their thoughts in natural language.

Here's where your skepticism that they're being honest about summarizing those thoughts comes into full force. I agree that it's not reliable; for instance, changing the intermediate answer in chain of thought prompting often doesn't change the final output, indicating that that output was for show.

However, a safer setup is to never use the same model twice. When you use chain-of-thought reasoning, construct a new context with the relevant information from memory; don't just let the context window accrue, since this allows fake chains-of-thought and the collapse of the simulator into a waluigi.

Scaffolded LLMs should not turn an LLM into an agent, but rather create a committee of LLMs that are called for individual questions needed to accomplish that committee's goals.

This isn't remotely a solution to the alignment problem, but it really seems to have massive upsides, and only the same downsides as other practically viable approaches to AGI.

To be clear, I only see some form of RL agents as the other practical possibility, and I like our odds much less with those.

I think there are other, even more readily alignable approaches to AGI. But they all seem wildly impractical. I think we need to get ready to align the AGI we get, rather than just preparing to say I-told-you-so after the world refuses to forego massive incentives to take a much slower but safer route to AGI.

To paraphrase, we need to go to the alignment war with the AGI we get, not the AGI we want.

I really like your recent series of posts that succinctly address common objections/questions/suggestions about alignment concerns. I'm making a list to show my favorite skeptics (all ML/AI people; nontechnical people, as Connor Leahy puts it, tend to respond "You fucking what? Oh hell no!" or similar when informed that we are going to make genuinely smarter-than-us AI soonish).

We do have ways to get an AI to do what we want. The hardcoded algorithmic maximizer approach seems to be utterly impractical at this point. That leaves us with approaches that don't obviously do a good job of preserving their own goals as they learn and evolve:

1. training a system to pursue things we like, such as shard theory and similar approaches.
2. training or hand-coding a critic system, as in outlined approaches from Steve Byrnes and me as well as many others. Nicely summarized as a Steering systems approach. This seems a bit less sketchy than training in our goals and hoping they generalize adequately, but still pretty sketchy. 
3. Telling the agent what to do in a Natural language alignment approach. This seems absurdly naive. However, I'm starting to think our first human-plus AGIs will be wrapped or scaffolded LLMs, and they to a nontrivial degree actually think in natural language. People are right now specifying goals in natural language, and those can include alignment goals (or destroying humanity, haha).  I just wrote an in-depth post on the potential Capabilities and alignment of LLM cognitive architectures, but I don't have a lot to say about stability in that post.

None of these directly address what I'm calling The alignment stability problem, to give a name to what you're addressing here. I think addressing it will work very differently in each of the three approaches listed above, and might well come down to implementational details within each approach. I think we should be turning our attention to this problem along with the initial alignment problems, because some of the optimism in the field stems from thinking about initial alignment and not long-term stability.

Edit: I left out Ozyrus's posts on approach 3. He's the first person I know of to see agentized LLMs coming, outside of David Shapiro's 2021 book. His post was written a year ago and posted two weeks ago to avoid infohazards. I'm sure there are others who saw this coming more clearly than I did, but I thought I'd try to give credit where it's due.

Great analysis. I'm impressed by how thoroughly you've thought this through in the last week or so. I hadn't gotten as far. I concur with your projected timeline, including the difficulty of putting time units onto it. Of course, we'll probably both be wrong in important ways, but I think it's important to at least try to do semi-accurate prediction if we want to be useful.

I have only one substantive addition to your projected timeline, but I think it's important for the alignment implications.

LLM-bots are inherently easy to align. At least for surface-level alignment. You can tell them "make me a lot of money selling shoes, but also make the world a better place" and they will try to do both. Yes, there are still tons of ways this can go off the rails. It doesn't solve outer alignment or alignment stability, for a start. But GPT4's ability to balance several goals, including ethical ones, and to reason about ethics, is impressive.^[1] You can easily make agents that both try to make money, and thinks about not harming people.

In short, the fact that you can do this is going to seep into the public consciousness, and we may see regulations and will definitely see social pressure to do this.

I think the agent disasters you describe will occur, but they will happen to people that don't put safeguards into their bots, like "track how much of my money you're spending and stop if it hits $X and check with me". When agent disasters affect other people, the media will blow it sky high, and everyone will say "why the hell didn't you have your bot worry about wrecking things for others?". Those who do put additional ethical goals into their agents will crow about it. There will be pressure to conform and run safe bots. As bot disasters get more clever, people will take more seriously the big bot disaster.

Will all of that matter? I don't know. But predicting the social and economic backdrop for alignment work is worth trying.

Edit: I finished my own followup post on the topic, Capabilities and alignment of LLM cognitive architectures. It's a cognitive psychology/neuroscience perspective on why these things might work better, faster than you'd intuitively think. Improvements to the executive function (outer script code) and episodic memory (pinecone or other vector search over saved text files) will interact so that improvements in each make the rest of system work better and easier to improve.

1. ^{^}

   I did a little informal testing of asking for responses in hypothetical situations where ethical and financial goals collide, and it did a remarkably good job, including coming up with win/win solutions that would've taken me a while to come up with. It looked like the ethical/capitalist reasoning of a pretty intelligent person; but also a fairly ethical one.

Thanks for engaging. I did read your linked post. I think you're actually in the majority in your opinion on AI leading to a continuation and expansion of business as usual. I've long been curious about about this line of thinking; while it makes a good bit of sense to me for the near future, I become confused at the "indefinite" part of your prediction.

When you say that AI continues from the first step indefinitely, it seems to me that you must believe one or more of the following:

• No one would ever tell their arbitrarily powerful AI to take over the world
  • Even if it might succeed
• No arbitrarily powerful AI could succeed at taking over the world
  • Even if it was willing to do terrible damage in the process
• We'll have a limited number of humans controlling arbitrarily powerful AI
  • And an indefinitely stable balance-of-power agreement among them
• By "indefinitely" you mean only until we create and proliferate really powerful AI

If I believed in any of those, I'd agree with you. 

Or perhaps I'm missing some other belief we don't share that leads to your conclusions.

Care to share?

Separately, in response to that post: your post you linked was titled AI values will be shaped by a variety of forces, not just the values of AI developers. In my prediction here, AI and AGI will not have values in any important sense; it will merely carry out the values of its principals (its creators, or the government that shows up to take control). This might just be terminological distinction, except for the following bit of implied logic: I don't think AI needs to share clients' values to be of immense economic and practical advantage to them. When (if) someone creates a highly capable AI system, they will instruct it to serve customers needs in certain ways, including following their requests within certain limits; this will not necessitate changing the A(G)I's core values (if they exist) to use it to make enormous profits when licensed to clients. To the extent this is correct, we should go on assuming that AI will share or at least follow its creators' values (or IMO more likely, take orders/values from the government that takes control, citing security concerns)

I'm sure it's not the same, particularly since neither one has really been fully fleshed out and thought through. In particular, Yudkowsky doesn't focus on the advantages of instructing the AGI to tell you the truth, and interacting with it as it gets smarter. I'd guess that's because he was still anticipating a faster takeoff than network-based AGI affords. 

But to give credit where it's due, I think that literal instruction-following was probably part of (but not the whole of) his conception of task-based AGI. From the discussion thread w Paul Christiano following the task directed AGI article on Greater Wrong: 

The AI is getting short-term objectives from humans and carrying them out under some general imperative to do things conservatively or with ‘low unnecessary impact’ in some sense of that, and describes plans and probable consequences that are subject to further human checking, and then does them, and then the humans observe the results and file more requests.

And the first line of that article:

A task-based AGI is an AGI intended to follow a series of human-originated orders, with these orders each being of limited scope [...]

These sections, in connection with the lack of reference to instructions and checking for most of the presentation, suggest to me that he probably was thinking of things like hard-coding it to design nonotech, melt down GPUs (or whatever) and then delete itself, but also of more online, continuous instruction-following AGI more similar to my conception of likely AGI projects. Bensinger may have been pursuing one part of that broader conception.

That would be fine by me if it were a stable long-term situation, but I don't think it is. It sounds like you're thinking mostly of AI and not AGI that can self-improve at some point. My major point in this post is that the same logic about following human instructiosn applies to AGI, but that's vastly more dangerous to have proliferate. There won't have to be many RSI-capable AGIs before someone tells their AGI "figure out how to take over the world and turn it into my utopia, before some other AGI turns it into theirs". It seems like the game theory will resemble the nuclear standoff, but without the mutually assured destruction aspect that prevents deployment. The incentives will be to be the first mover to prevent others from deploying AGIs in ways you don't like.

That is fascinating. I hadn't seen his "task AGI" plan, and I agree it's highly overlapping with this proposal - more so than any other work I was aware of. What's most fascinating is that YK doesn't currently endorse that plan, even though it looks to me as though on main reason he calls it "insanely difficult" has been mitigated greatly by the success of LLMs in understanding human semantics and therefore preferences. We are already well up his Do-What-I-Mean hierarchy, arguably at an adequate level for safety/success even before inevitable improvements on the way to AGI. In addition, the slow takeoff path we're on seems to also make the project easier (although less likely to allow a pivotal act before we have many AGIs causing coordination problems).

So, why does YK think we should Shut It Down instead of build DWIM AGI? Ii've been trying to figure this out. I think his principal reasons are two: reinforcement learning sounds like a good way to get any central goal somewhat wrong, and being somewhat wrong could well be too much for survival. As I mentioned in the article, I think we have good alternatives to RL alignment, particularly for the AGI we're most likely to build first, and I don't think YK has ever considered proposals of that type. Second, he thinks that humans are stunningly foolish, and that competitive race dynamiccs will make them even more prone to critical errors, even for a project that's in-principal quite accomplishable. On this, I'm afraid I agree. So if I were in charge, I would indeed Shut It Down instead of shooting for DWIM alignment. But I'm not, and neither is YK. He thinks it's worth trying, to at least slow down AGI progress; I think it's more critical to use the time we've got to refine the alignment approaches that are most likely to actually be deployed.

I very much agree. Part of why I wrote that post was that this is a common assumption, yet much of the discourse ignores it and addresses value alignment. Which would be better if we could get it, but it seems wildly unrealistic to expect us to try.

The pragmatics of creating AGI for profit are a powerful reason to aim for instruction-following instead of value alignment; to the extent it will actually be safer and work better, that's just one more reason that we should be thinking about that type of alignment. Not talking about it won't keep it from taking that path.