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.

Big upvote for looking for routes to cooperation instead of either despairing or looking for reasons for conflict.

This all got a little long, so I'll put the biggest conclusion up front: I think we're in a good situation right now, in which the leading players pursuing AGI are probably not sadists, dense psychopaths, or zealots of mistaken ideologies. We'd probably like their utopias just fine. If we assume the competition to control aligned AGI will be much broader, we have more reason to be concerned.

One major crux of this post's claims is the intuition that there would be only minor variations in the "utopia" brought about by different actors with an aligned ASI. Intuitions/theories seem to vary widely on this point. OP didn't present much argument for that, so let me expand on it a little.

In sum, it's a question about human nature. Given unlimited power, will people use it to give people what they want, or will they enforce a world order most people hate?

This of course depends on the individuals and ideologies that wind up controlling that AGI.

It requires little benevolence to help people when it's trivially easy for the people in charge to do it.

This is one reason for optimism. It's based on the prediction that aligned AGI becomes aligned ASI and ASI can create a post-scarcity world. In a post-scarcity world, everyone can easily be given material resources that empower them to do whatever they want.

The pessimistic view is that some people or organizations don't have even the small bit of benevolence required to do good when it's trivially easy.

The thesis is that other motivations would outweigh their small amount of empathy/benevolence. This could be sadism; desire for revenge for perceived harms; or sincere belief in a mistaken worldview (e.g., religion or other prescriptive forms of ethics).

I think those possibilities are real, but we must also ask how those ideologies and emotional preferences would change over time.

Another reason for pessimism is not believing or not considering the post-scarcity hypothesis. The way corporations and individuals wield power in a world with scarcity does not inspire confidence. But the profit motive barely applies once you've "won that game". How would a corporation with access to unlimited production use that power? I think it would depend on the particular individuals in power, and their ideologies. And the power motive that's so destructive loses much of its emotional force once that individual has attained nearly unlimited power.

The creators of aligned AGI have won whatever game they've been playing. They have access to unlimited personal wealth and power. They and their loved ones are permanently safe from physical harms. No individual in history has ever been in that position.

I think the common saying "power corrupts" is quite mistaken, in an understandable way. The pursuit of power is what corrupts. It rewards unethical decisions, and provides pressure for the individual to see those decisions as ethical or virtuous. This corrupts their character. Every leader in history has had legitimate concerns about preserving their power. The individuals controlling AGI would not. If this is correct, the question is how corrupted they became while acquiring power, and whether they'll over time become more generous, as the legitimate reasons for selfishness disappear in reality, and perhaps in their emotional makeup as a result.

There's another important concern that sociopaths tend to acquire power in our current systems, while hiding their sociopathy. I think this is true, but we don't know how common it is. And we don't know that much about sociopathy. I think it probably exists on a spectrum, since it doesn't have a single source of genetic variance.

So, I hold out some hope that even most people on the sociopathy spectrum, or ideologically confused power structures would shift toward having the minimal benevolence-balance to provide a pretty awesome utopia. But I'd prefer to gamble on the utopias offered by Altman, Hassabis, or Amodio. This is an argument against an AI pause, but not a strong one.

I'm excited to have this written up so clearly, nice work! I think this is important for alignment work in two ways: discourse and thinking about alignment is affected by powerful cognitive biases that this hypothesis explains; and, as you point out, we might build AGI that works like this, since it's so effective for human cognition.

I'm very curious if this "rings true" to other readers based on their introspection and observation of others' thinking patterns. I think this is both true and important. I'd arrived at this conclusion over the course of a research career studying dopamine and higher cognition. When we started researching cognitive biases, this came together, and I think this ubiquitous valence effect is the source of the most important cognitive biases. This goes by the names motivated reasoning, confirmation bias, and the halo effect; they have overlapping behavioral definitions. I think they're the major stumbling block to humans behaving rationally.

I think this hypothesis is consistent with a vast array of empirical work on dopamine function and related cognitive function. But the evidence isn't adequate to firmly establish that dopamine signals valence. That's part of why I'd never written this up adequately, and because hypotheses this broad are outside of the scope of standard neuroscience funding.

I'm looking forward to the rest of the series, and hoping the posts addressing cognitive biases generate some discussion about how those biases affect alignment discussions. I think the combination of motivated reasoning, confirmation bias, and the halo/horns effects create powerful polarization that's a big obstacle to rational discussions of alignment

Calling it an "unambiguously bad post" definitely sounds like unhelpful trolling, not genuine approval. If you have similar "praise" in the future, I suggest you word it more carefully.

It sounds like you wish this wasn't written at all because you'd prefer a more detailed post? But that's not always a realistic option. There's a lot to attend to.

If you're saying the that posting a "I ignored this but here's some guesses I took about it that lead me to ignore it", isn't helpful, that's more reasonable.

I find it helpful because Nate's worldview is an important one, and he's at least bothered to tell us something about why he's not engaging more deeply.

Fortunately, that more detailed analysis has been done by Steve Byrnes: Thoughts on “AI is easy to control” by Pope & Belrose

Would any of these count for you?

We have promising alignment plans with low taxes

Each of the three plans I mention are attempts to put the "understanding" part into the "wanting" slot (a "steering subsystem" that controls goals for decision-making purposes) in a different AGI design. That brief post links to somewhat more detailed plans.