AI Governance researcher with Polaris Ventures, formerly of the MTAIR project, TFI and Center on Long-term Risk, Graduate researcher at Kings and AI MSc at Edinburgh. Interested in philosophy, longtermism and AI Alignment.
Sammy Martin
Karma: 2,254
In the late 1940s and early 1950s nuclear weapons did not provide an overwhelming advantage against conventional forces. Being able to drop dozens of ~kiloton range fission bombs in eastern European battlefields would have been devastating but not enough by itself to win a war. Only when you got to hundreds of silo launched ICBMs with hydrogen bombs could you have gotten a true decisive strategic advantage
He’s the best person they could have gotten on the technical side but Paul’s strategic thinking has been consistently clear eyed and realistic but also constructive, see for example this: www.alignmentforum.org/posts/fRSj2W4Fjje8rQWm9/thoughts-on-sharing-information-about-language-model
So to the extent that he’ll have influence on general policy as well this seems great!
This whole thing reminds me of Scott Alexander’s Pyramid essay. That seems like a really good case where it seems like there’s a natural statistical reference class, seems like you can easily get a giant Bayes factor that’s “statistically well justified”, and to all the counterarguments you can say “well the likelihood is 1 in 10^5 that the pyramids would have a latitude that matches to the speed of light in m/s”. That’s a good reductio for taking even fairly well justified sounding subjective bayes factors at face value.
And I think that it’s built into your criticism that because the problem is social and hidden evidence filtering going on, there will also tend to be an explanation on the meta-level too for why my coincidence finding is different from your coincidence finding.
Tom Davidson’s report: https://docs.google.com/document/d/1rw1pTbLi2brrEP0DcsZMAVhlKp6TKGKNUSFRkkdP_hs/edit?usp=drivesdk
My old 2020 post: https://www.lesswrong.com/posts/66FKFkWAugS8diydF/modelling-continuous-progress
In my analysis of Tom Davidson’s “Takeoff Speeds Report,” I found that the dynamics of AI capability improvement, as discussed in the context of a software-only singularity, align closely with the original simplified equation ( I′(t) = cI + f(I)I^2 ) from my 4 year old post on Modeling continuous progress. Essentially, that post describes how we switch from exponential to hyperbolic growth as the fraction of AI research done by AIs improves along a logistic curve. These are all features of the far more complex mathematical model in tom’s report.
In this equation, ( I ) represents the intelligence or capability of the AI system. It correlates to the cognitive output or the efficiency of the AI as described in the report, where the focus is on the software improvements contributing to the overall effectiveness of AI systems. The term ( cI ) in the equation can be likened to the constant external effort or input in improving AI systems, which is consistent with the ongoing research and development efforts mentioned in the report. This part of the equation represents the incremental improvements in AI capabilities due to human-led development efforts.
The second term in the equation, ( f(I)I^2 ), is particularly significant in understanding the relationship with the software-only singularity concept. Here, ( f(I) ) is a function that determines the extent to which the AI system can use its intelligence to improve itself, essentially a measure of recursive self-improvement (RSI). In the report, the discussion about a software-only singularity uses a similar concept, where the AI systems reach a point where their self-improvement significantly accelerates their capability growth. This is analogous to ( f(I) ) increasing, leading to a more substantial contribution of ( I^2 ) (the AI’s self-improvement efforts) to the overall rate of intelligence growth, ( I′(t) ). As the AI systems become more capable, they contribute more to their own development, a dynamic that both the equation and the report capture. The report has a ‘FLOP gap’ from when AIs start to contribute to research at all to when they fully take over which is essentially the upper and lower bounds to fit the f(I) curve to. Otherwise, the overall rate of change is sharper in tom’s report as I ignored increasing investment and increasing compute in my model focussing only on software self improvement feedback loops.
One other thing I liked about Tom’s report is it’s focus on relatively outside viewy bio anchors and epoch AIs direct approach estimates for what is needed for TAI.
Maybe this is an unreasonable demand, but one concern I have about all of these alleged attempts to measure the ability of an AI to automate scientific research, is that this feels like a situation where it’s unusually slippery and unusually easy to devise a metric that doesn’t actually capture what’s needed to dramatically accelerate research and development. Ideally, I’d like a metric where we know, as a matter of necessity, that a very high score means that the system would be able to considerably speed up research.
For example, the direct approach estimation does have this property, where if you can replicate to a certain level of accuracy what a human expert would say over a certain horizon length, you do in some sense have to be able to match or replicate the underlying thinking that produced it, which means being able to do long horizon tasks, but of course, that’s a very vague upper bound. It’s not perfect, the Horizon Length metric might only cover the 90th percentile of tasks at each time scale. The remaining 10th percentile might contain harder, more important tasks necessary for AI progress
I think trying to anticipate and list in a task all the capabilities you think you need to automate scientific progress when we don’t really know what those are will lead to a predictable underestimate of what’s required.
I thought it was worth commenting here because to me the 3 way debate with Eliezer Yudkowsky, Nora Belrose, and Andrew Critch managed to collectively touch on just about everything that I think the common debate gets wrong about AI “doom” with the result that they’re all overconfident in their respective positions.
Starting with Eliezer and Nora’s argument. Her statement:
“Alien shoggoths are about as likely to arise in neural networks as Boltzmann brains are to emerge from a thermal equilibrium.”
To which Eliezer responds,
“How blind to ‘try imagining literally any internal mechanism that isn’t the exact thing you hope for’ do you have to be—to think that, if you erase a brain, and then train that brain solely to predict the next word spoken by nice people, it ends up nice internally?”
I agree that it’s a mistake to identify niceness with predicting nice behaviour, and I agree that Nora is overconfident in no generalisation failures as a result of making a similar mistake. If your model says it’s literally as unlikely as a boltzmann brain appearing from nowhere then something has gone wrong. But, I don’t think that her point is as straightforward as just conflating a nice internal mechanism with nice feedback. I’m going to try and explain what I think her argument is.
I think that Eliezer has an implicit model that there’s zillions of potential generalizations to predict niceness which a model could learn, that all are pretty much equally likely to get learned a priori, and actually being nice is just one of them so it’s basically impossible for RLHF to hit on it, so RLHF would require tremendous cosmic coincidences to work.
Maybe this is true in some sense for arbitrarily superintelligent AI. But, as Paul Christiano said, I think that this tells us not much about what to expect for “somewhat superhuman” AI. Which is what we care about for predicting whether we’ll see misalignment disasters in practice.
Rather, “actually learning to be nice” is how humans usually learn to predict nice behaviour. Of all the possible ways that generalisation from nice training could happen, this is privileged as a hypothesis somewhat, it stands out from the background haze of random mechanisms that could be learned.
If the reasons this strategy worked for humans are transferable to the LLM case (and that is highly arguable and unclear), then yes, it might be true that giving agents rewards for being nice causes them to internally develop a sort of pseudo-niceness representation that controls their behaviour and planning even up to superhuman levels, even out of distribution. It’s not for ‘literally no reason’ or ‘by coincidence’ or ‘because of a map-territory conflation’, but because its possible such a mechanism in the form of a model inductive bias really exists and we have some vague evidence in favor of it.
Okay, so what’s the internal mechanism that I’m imagining which gets us there? Here’s a sketch, based on an “easy world” outlined in my alignment difficulty post.
Suppose that (up to some level of competence that’s notably superhuman for most engineering tasks), LLMs just search over potential writers of text, with RLHF selecting from the space of agents that have goals only over text completion. They can model the world, but since they start out modelling text, that’s what their goals range over, even up to considerably superhuman competence at a wide range of tasks. They don’t want things in the real world, and only model it to get more accurate text predictions. Therefore, you can just ask RLHF’d GPT-10, “what’s the permanent alignment solution?”, and it’ll tell you.
People still sometimes say, “doesn’t this require us to get unreasonably, impossibly lucky with generalisation?”. No, it requires luck but you can’t say it’s unbelievable impossible luck just based on not knowing how generalisation works. I also think recent evidence (LLMs getting better at modelling the world without developing goals over it) suggests this world is a bit more likely than it seemed years ago as Paul Christiano argues here:
“I think that a system may not even be able to “want” things in the behaviorist sense, and this is correlated with being unable to solve long-horizon tasks. So if you think that systems can’t want things or solve long horizon tasks at all, then maybe you shouldn’t update at all when they don’t appear to want things.”
But that’s not really where we are at—AI systems are able to do an increasingly good job of solving increasingly long-horizon tasks. So it just seems like it should obviously be an update, and the answer to the original question
Could you give an example of a task you don’t think AI systems will be able to do before they are “want”-y? At what point would you update, if ever? What kind of engineering project requires an agent to be want-y to accomplish it? Is it something that individual humans can do? (It feels to me like you will give an example like “go to the moon” and that you will still be writing this kind of post even once AI systems have 10x’d the pace of R&D.)
But, again, I’m not making the claim that this favourable generalisation that gets RLHF to work is likely, just that it’s not a random complex hypothesis with no evidence for it that’s therefore near-impossible.
Since we don’t know how generalisation works, we can’t even say “we should have a uniform prior over internal mechanisms which I can describe that could get high reward”. Rather, if you don’t know, you really just don’t know, and the mechanism involving actually learning to be nice to predict niceness, or actually staying in the domain you were initially trained on when planning, might be favoured by inductive biases in training.
But even if you disagree with me on that, the supposed mistake is not (just) as simple as literally conflating the intent of the overseers with the goals that the AI learns, rather there’s a thought that replicating the goals that produced the feedback and simply adopting them as your own is a natural, simple way to learn to predict what the overseer wants even up to fairly superhuman capabilities so it’s what will get learned by default even if it isn’t the globally optimal reward-maximiser. Is this true? Well, I don’t know, but it’s at least a more complicated mistake if false. This point has been made many times in different contexts, there’s a summary discussion here that outlines 6 different presentations of this basic idea.
If I had to sum it up, I think that while Nora maybe confuses the map with the territory, Eliezer conflates ignorance with positive knowledge (from ‘we don’t know how generalisation works’ to ‘we should have a strong default uniform prior over every kind of mechanism we could name’).
Then there’s Andrew Critch, who I think agrees with and understands the point I’ve just made (that Nora’s argument is not a simple mistake of the map for the territory), but then makes a far more overreaching and unjustifiable claim than Eliezer or Nora in response.
In the Nora/Eliezer case, they were both very confident in their respective models of AI generalisation, which is at least the kind of thing about which you could be extremely confident, should you have strong evidence (which I don’t think we do). Social science and futurism is not one of those things. Critch says,
″ I think literally every human institution will probably fail or become fully dehumanized by sometime around (median) 2040.”
The “multipolar chaos” prediction, which is that processes like a fast proliferating production web will demolish or corrupt all institutional opposition and send us to dystopia with near-certainty, I just don’t buy.
I’ve read his production web stories and also heard similar arguments from many people, and it’s hard to voice my objections as specific “here’s why your story can’t happen” (I think many of them are at least somewhat plausible, in fact), but I still think there’s a major error of reasoning going on here. I think it’s related to the conjunction fallacy, the sleepwalk bias and possibly not wanting to come across as unreasonably optimistic about our institutions.
Here’s one of the production web stories in brief but you can read it in full along with my old discussion here,
In the future, AI-driven management assistant software revolutionizes industries by automating decision-making processes, including “soft skills” like conflict resolution. This leads to massive job automation, even at high management levels. Companies that don’t adopt this technology fall behind. An interconnected “production web” of companies emerges, operating with minimal human intervention and focusing on maximizing production. They develop a self-sustaining economy, using digital currencies and operating beyond human regulatory reach. Over time, these companies, driven by their AI-optimized objectives, inadvertently prioritize their production goals over human welfare. This misalignment leads to the depletion of essential resources like arable land and drinking water, ultimately threatening human survival, as humanity becomes unable to influence or stop these autonomous corporate entities.
My object-level response is to say something mundane along the lines of, I think each of the following is more or less independent and not extremely unlikely to occur (each is above 1% likely):
Wouldn’t governments and regulators also have access to AI systems to aid with oversight and especially with predicting the future? Remember, in this world we have pseudo-aligned AI systems that will more or less do what their overseers want in the short term.
Couldn’t a political candidate ask their (aligned) strategist-AI ‘are we all going to be killed by this process in 20 years’ and then make a persuasive campaign to change the public’s mind with this early in the process, using obvious evidence to their advantage
If the world is alarmed by the expanding production web and governments have a lot of hard power initially, why will enforcement necessarily be ineffective? If there’s a shadow economy of digital payments, just arrest anyone found dealing with a rogue AI system. This would scare a lot of people.
We’ve already seen pessimistic views about what AI regulations can achieve self-confessedly be falsified at the 98% level—there’s sleepwalk bias to consider. Stefan schubert: Yeah, if people think the policy response is “99th-percentile-in-2018”, then that suggests their models have been seriously wrong. So maybe the regulations will be both effective, foresightful and well implemented with AI systems forseeing the long-run consequences of decisions and backing them up.
What if the lead project is unitary and a singleton or the few lead projects quickly band together because they’re foresightful, so none of this race to the bottom stuff happens in the first place?
If it gets to the point where water or the oxygen in the atmosphere is being used up (why would that happen again, why wouldn’t it just be easier for the machines to fly off into space and not have to deal with the presumed disvalue of doing something their original overseers didn’t like?) did nobody build in ‘off switches’?
Even if they aren’t fulfilling our values perfectly, wouldn’t the production web just reach some equilibrium where it’s skimming off a small amount of resources to placate its overseers (since its various components are at least somewhat beholden to them) while expanding further and further?
And I already know the response is just going to be “Moloch wouldn’t let that happen..” and that eventually competition will mean that all of these barriers disappear. At this point though I think that such a response is too broad and proves too much. If you use the moloch idea this way it becomes the classic mistaken “one big idea universal theory of history” which can explain nearly any outcome so long as it doesn’t have to predict it.
A further point: I think that someone using this kind of reasoning in 1830 would have very confidently predicted that the world of 2023 would be a horrible dystopia where wages for workers wouldn’t have improved at all because of moloch.
I agree that it’s somewhat easier for me to write a realistic science fiction story set in 2045 that’s dystopian compared to utopian, assuming pseudo-aligned AGI and no wars or other obvious catastrophic misuse. As a broader point, I along with the great majority of people, don’t really want this transition to happen either way, and there are many aspects of the ‘mediocre/utopian’ futures that would be suboptimal, so I get why the future forecasts don’t ever look normal or low-risk.
But I think all this speculation tells us very little with confidence what the default future looks like. I don’t think a dystopian economic race to the bottom is extremely unlikely, and with Matthew Barnett I am worried about what values and interests will influence AI development and think the case for being concerned about whether our institutions will hold is strong.
But saying that moloch is a deterministic law of nature such that we can be near-certain of the outcome is not justifiable. This is not even the character of predictions about which you can have such certainty.
Also, in this case I think that a reference class/outside view objection that this resembles failed doomsday predictions of the past is warranted.
I don’t agree that these objections have much weight when we’re concerned about misaligned AI takeover as that has a clear, singular obvious mechanism to be worried about.
However, for ‘molochian race to the bottom multipolar chaos’, it does have the characteristic of ignoring or dismissing endogenous responses, society seeing what’s happening and deciding not to go down that path, or just unknown unknowns that we saw with past failed doomsday predictions. This I see as absolutely in the same reference class as people who in past decades were certain of overpopulation catastrophes or the people now who are certain or think a civilizational collapse from the effects of climate change are likely. It’s taking current trends and drawing mental straight lines on them to extreme heights decades in the future.
I also expect that if implemented the plans in things like Project 2025 would impair the ability of the government to hire civil servants who are qualified and probably just degrade the US Government’s ability to handle complicated new things of any sort across the board.
If you want a specific practical example of the difference between the two: we now have AIs capable of being deceptive when not specifically instructed to do so (‘strategic deception’) but not developing deceptive power-seeking goals completely opposite what the overseer wants of them (‘deceptive misalignment’). This from Apollo research on Strategic Deception is the former not the latter,
Doc Xardoc reports back on the Chinese alignment overview paper that it mostly treats alignment as an incidental engineering problem, at about a 2.5 on a 1-10 scale with Yudkowsky being 10
I’m pretty sure Yudkowsky is at around an 8.5 actually (I think he thinks it’s not impossible in principle for ML like systems but maybe it is). 10 would be impossible in principle.
I think that aside from the declaration and the promise for more summits the creation of the AI Safety Institute and its remit are really good, explicitly mentioning auto-replication and deception evals and planning to work with the likes of Apollo Research and ARC evals to test for:
Also, NIST is proposing something similar.
I find this especially interesting because we now know that in the absence of any empirical evidence of any instance of deceptive alignment at least one major government is directing resources to developing deception evals anyway. If your model of politics doesn’t allow for this or considers it unlikely then you need to reevaluate it as Matthew Barnett said.
Additionally, the NIST consortium and AI Safety Institute both strike me as useful national-level implementations of the ‘AI risk evaluation consortium’ idea proposed by TFI.
King Charles notes (0:43 clip) that AI is getting very powerful and that dealing with it requires international coordination and cooperation. Good as far as it goes.
I find it amusing that for the first time in hundreds of years a king is once again concerned about superhuman non-physical threats (at least if you’re a mathematical platonist about algorithms and predict instrumental convergence as a fundamental property of powerful minds) to his kingdom and the lives of his subjects. :)
I don’t like the term pivotal act because it implies without justification that the risk elimination has to be a single action. Depending on the details of takeoff speed that may or may not be a requirement but if the final speed is months or longer then almost certainly there will be many actions taken by humans + AI of varying capabilities that together incrementally reduce total risk to low levels. I talk about this in terms of ‘positively transformative AI’ as the term doesn’t bias you towards thinking this has to be a single action, even if nonviolent.
Seeing the risk reduction as a single unitary action, like seeing it as a violent overthrow of all the world’s governments, also makes the term seem more authoritarian, crazy, fantastical and off-putting to anyone involved in real world politics so I’d recommend that in our thinking we make both the change you suggest and stop thinking of it as necessarily one action.
This as a general phenomenon (underrating strong responses to crises) was something I highlighted (calling it the Morituri Nolumus Mori) with a possible extension to AI all the way back in 2020. And Stefan Schubert has talked about ‘sleepwalk bias’ even earlier than that as a similar phenomenon.
https://twitter.com/davidmanheim/status/1719046950991938001
https://twitter.com/AaronBergman18/status/1719031282309497238
I think the short explanation as to why we’re in some people’s 98th percentile world so far (and even my ~60th percentile) for AI governance success is that if was obvious to you how transformative AI would be over the next couple of decades in 2021 and yet nothing happened, it seems like governments are just generally incapable.
The fundamental attribution error makes you think governments are just not on the ball and don’t care or lack the capacity to deal with extinction risks, rather than decision makers not understanding obvious-to-you evidence that AI poses an extinction risk. Now that they do understand, they will react accordingly. It doesn’t meant that they will react well necessarily, but they will act on their belief in some manner.
Lomborg is massively overconfident in his predictions but not exactly less wrong than the implicit mainstream view that the economic impacts will definitely be ruinous enough to justify expensive policies.
It’s very hard to know, the major problem is just that the existing climate econ models make so many simplifying assumptions that they’re near-useless except for giving pretty handwavy lower bounds on damage, especially when the worst risks to worry about are in correlated disasters and tail risks, and Lomborg makes the mistake of taking them completely literally. I discussed this at length a couple of years ago and John Halstead later wrote a book-length report on what the climate impacts literature can and can’t tell us.
Roon also lays down the beats.
For those who missed the reference
The ARC evals showing that when given help and a general directive to replicate a GPT-4 based agent was able to figure out that it ought to lie to a TaskRabbit worker is an example of it figuring out a self-preservation/power-seeking subgoal which is on the road to general self-preservation. But it doesn’t demonstrate an AI spontaneously developing self-preservation or power-seeking, as an instrumental subgoal to something that superficially has nothing to do with gaining power or replicating.
Of course we have some real-world examples of specification-gaming like you linked in your answer: those have always existed and we see more ‘intelligent’ examples like AIs convinced of false facts trying to convince people they’re true.
There’s supposedly some evidence here that we see power-seeking instrumental subgoals developing spontaneously but how spontaneous this actually was is debatable so I’d call that evidence ambiguous since it wasn’t in the wild.
>APS is less understood and poorly forecasted compared to AGI.
I should clarify that I was talking about the definition used by forecasts like the Direct Approach methodology and/or the definition given in the metaculus forecast or in estimates like the Direct Approach. The latter is roughly speaking, capability sufficient to pass a hard adversarial Turing tests and human-like capabilities on enough intellectual tasks as measured by certain tests.This is something that can plausibly be upper bounded by the direct approach methodology which aims to predict when an AI could get a negligible error in predicting what a human expert would say over a specific time horizon. So this forecast is essentially a forecast of ‘human-expert-writer-simulator AI’, and that is the definition that’s used in public elicitations like the metaculus forecasts.
However, I agree with you that while in some of the sources I cite that’s how the term is defined it’s not what the word denotes (just generality, which e.g. GPT-4 plausibly is for some weak sense of the word), and you also don’t get from being able to simulate the writing of any human expert to takeover risk without making many additional assumptions.
I guess it is down to Tyler’s personal opinion, but would he accept asking IR and defense policy experts on the chance of a war with China as an acceptable strategy or would he insist on mathematical models of their behaviors and responses? To me it’s clearly the wrong tool, just as in the climate impacts literature we can’t get economic models of e.g. how governments might respond to waves of climate refugees but can consult experts on it.
I recently held a workshop with PIBBSS fellows on the MTAIR model and thought some points from the overall discussion were valuable:
The discussants went over various scenarios related to AI takeover, including a superficially aligned system being delegated lots of power and gaining resources by entirely legitimate means, a WFLL2-like automation failure, and swift foom takeover. Some possibilities involved a more covert, silent coup where most of the work was done through manipulation and economic pressure. The concept of “$1T damage” as an intermediate stage to takeover appeared to be an unnatural fit with some of these diverse scenarios. There was some mention of whether mitigation or defensive spending should be considered as part of that $1T figure.
Alignment Difficulty and later steps merge many scenarios
The discussants interpreted “alignment is hard” (step 3) as implying that alignment is sufficiently hard that (given that APS is built), at least one APS is misaligned somewhere, and also that there’s some reasonable probability that any given deployed APS is unaligned. This is the best way of making the whole statement deductively valid.
However, proposition 3 being true doesn’t preclude the existence of other aligned APS AI (hard alignment and at least one unaligned APS might mean that there are leading conscientious aligned APS projects but unaligned reckless competitors). This makes discussion of the subsequent questions harder, as we have to condition on there possibly being aligned APS present as well which might reduce the risk of takeover.
This means that when assessing proposition 4, we have to condition on some worlds where aligned APS has already been deployed and used for defense, some where there have been warning shots and strong responses without APS, some where misaligned APS emerges out of nowhere and FOOMs too quickly for any response, and a slow takeoff where nonetheless every system is misaligned and there is a WFLL2 like takeover attempt, and add up the chance of large scale damage in all of these scenarios, conditioning on their probability, which makes coming to an overall answer to 4 and 5 challenging.
Definitions are value-laden and don’t overlap: TAI, AGI, APS
We differentiated between Transformative AI (TAI), defined by Karnofsky, Barnett and Cotra entirely by its impact on the world, which can either be highly destructive or drive very rapid economic growth; General AI (AGI), defined through a variety of benchmarks including passing hard adversarial Turing tests and human-like capabilities on enough intellectual tasks; and APS, which focuses on long-term planning and human-like abilities only on takeover-relevant tasks. We also mentioned Paul Christiano’s notion of the relevant metric being AI ‘as economically impactful as a simulation of any human expert’ which technically blends the definitions of AGI and TAI (since it doesn’t necessarily require very fast growth but implies it strongly). Researchers disagree quite a lot on even which of these are harder: Daniel Kokotaljo has argued that APS likely comes before TAI and maybe even before (the Matthew Barnett definition of) AGI, while e.g. Barnett thinks that TAI comes after AGI with APS AI somewhere in the middle (and possibly coincident with TAI).
In particular, some definitions of ‘AGI’, i.e. human-level performance on a wide range of tasks, could be much less than what is required for APS depending on what the specified task range is. If the human-level performance is only on selections of tasks that aren’t useful for outcompeting humans strategically (which could still be very many tasks, for example, human-level performance on everything that requires under a minute of thinking), the ‘AGI system’ could almost entirely lack the capabilities associated with APS. However, most of the estimates that could be used in a timelines estimate will revolve around AGI predictions (since they will be estimates of performance or accuracy benchmarks), which we risk anchoring on if we try to adjust them to predict the different milestones of APS.
In general it is challenging to use the probabilities from one metric like TAI to inform other predictions like APS, because each definition includes many assumptions about things that don’t have much to do with AI progress (like how qualitatively powerful intelligence is in the real world, what capabilities are needed for takeover, what bottlenecks are there to economic automation or research automation etc.) In other words, APS and TAI are value-laden terms that include many assumptions about the strategic situation with respect to AI takeover, world economy and likely responses.
APS is less understood and more poorly forecasted compared to AGI. Discussants felt the current models for AGI can’t be easily adapted for APS timelines or probabilities. APS carries much of the weight in the assessment due to its specific properties: i.e. many skeptics might argue that even if AGI is built, things which don’t meet the definition of APS might not be built.
Alignment and Deployment Decisions
Several discussants suggested splitting the model’s third proposition into two separate components: one focusing on the likelihood of building misaligned APS systems (3a) and the other on the difficulty of creating aligned ones (3b). This would allow a more nuanced understanding of how alignment difficulties influence deployment decisions. They also emphasized that detection of misalignment would impact deployment, which wasn’t sufficiently clarified in the original model.
Advanced Capabilities
There was a consensus that ‘advanced capabilities’ as a term is too vague. The discussants appreciated the attempt to narrow it down to strategic awareness and advanced planning but suggested breaking it down even further into more measurable skills, like hacking ability, economic manipulation, or propaganda dissemination. There are, however, disagreements regarding which capabilities are most critical (which can be seen as further disagreements about the difficulty of APS relative to AGI).
If strategic awareness comes before advanced planning, we might see AI systems capable of manipulating people, but not in ways that greatly exceed human manipulative abilities. As a result, these manipulations could potentially be detected and mitigated and even serve as warning signs that lower total risk. On the other hand, if advanced capabilities develop before strategic awareness or advanced planning, we could encounter AI systems that may not fully understand the world or their position in it, nor possess the ability to plan effectively. Nevertheless, these systems might still be capable of taking single, highly dangerous actions, such as designing and releasing a bioweapon.
Outside View & Futurism Reliability
We didn’t cover the outside view considerations extensively, but various issues under the “accuracy of futurism” umbrella arose which weren’t specifically mentioned.
The fact that markets don’t seem to have reacted as if Transformative AI is a near-term prospect, and the lack of wide scale scrutiny and robust engagement with risk arguments (especially those around alignment difficulty), were highlighted as reasons to doubt this kind of projection further.
The Fermi Paradox implies a form of X-risk that is self-destructive and not that compatible with AI takeover worries, while market interest rates also push the probability of such risks downward. The discussants recommended placing more weight on outside priors than we did in the default setting for the model, suggesting a 1:1 ratio compared to the model’s internal estimations.
Discussants also agreed with the need to balance pessimistic surviva- is-conjunctive views and optimistic survival-is-disjunctive views, arguing that the Carlsmith model is biased towards optimism and survival being disjunctive but that the correct solution is not to simply switch to a pessimism-biased survival is conjunctive model in response.
Difficult to separate takeover from structural risk
There’s a tendency to focus exclusively on the risks associated with misaligned APS systems seeking power, which can introduce a bias towards survival being predicated solely on avoiding APS takeover. However, this overlooks other existential risk scenarios that are more structural. There are potential situations without agentic power-seeking behavior but characterized by rapid changes could for less causally clear reasons include technological advancements or societal shifts that may not necessarily have a ‘bad actor’ but could still spiral into existential catastrophe. This post describes some of these scenarios in more detail.
This is a serious problem, but it is under active investigation at the moment, and the binary of regulation or pivotal act is a false dichotomy. Most approaches that I’ve heard of rely on some combination of positively transformative AI tech (basically lots of TAI technologies that reduce risks bit by bit, overall adding up to an equivalent of a pivotal act) and regulation to give time for the technologies to be used to strengthen the regulatory regime in various ways or improve the balance of defense over offense, until eventually we transition to a totally secure future: though of course this assumes at least (somewhat) slow takeoff.
You can see these interventions as acting on the conditional probabilities 4) and 5) in our model by driving down the chance that assuming misaligned APS is deployed, it can cause large-scale disasters.
4) Misaligned APS systems will be capable of causing a large global catastrophe upon deployment,
5) The human response to misaligned APS systems causing such a catastrophe will not be sufficient to prevent it from taking over completely,
6) Having taken over, the misaligned APS system will destroy or severely curtail the potential of humanity.This hasn’t been laid out in lots of realistic detail yet not least because most AI governance people are currently focused on near-term actions like making sure the regulations are actually effective, because that’s the most urgent task. But this doesn’t reflect a belief that regulations alone are enough to keep us safe indefinitely.
Holden Karnofsky has written on this problem extensively,
Oh, we’ve been writing up these concerns for 20 years and no one listens to us.′ My view is quite different. I put out a call and asked a lot of people I know, well-informed people, ‘Is there any actual mathematical model of this process of how the world is supposed to end?’...So, when it comes to AGI and existential risk, it turns out as best I can ascertain, in the 20 years or so we’ve been talking about this seriously, there isn’t a single model done.
I think that MTAIR plausibly is a model of the ‘process of how the world is supposed to end’, in the sense that it runs through causal steps where each individual thing is conditioned on the previous thing (APS is developed, APS is misaligned, given misalignment it causes damage on deployment, given that the damage is unrecoverable), and for some of those inputs your probabilities and uncertainty distribution could itself come from a detailed causal model (e.g. you can look at the Direct Approach for the first two questions.
For the later questions, like e.g. what’s the probability that an unaligned APS can inflict large disasters given that it is deployed, we can enumerate ways that it could happen in detail but to assess their probability you’d need to do a risk assessment with experts not produce a mathematical model.
E.g. you wouldn’t have a “mathematical model” of how likely a US-China war over Taiwan is, you’d do wargaming and ask experts or maybe superforecasters. Similarly, for the example that he gave which was COVID there was a part of this that was a straightforward SEIR model and then a part that was more sociological talking about how the public response works (though of course a lot of the “behavioral science” then turned out to be wrong!).
So a correct ‘mathematical model of the process’ if we’re being fair, would use explicit technical models for technical questions and for sociological/political/wargaming questions you’d use other methods. I don’t think he’d say that there’s no ‘mathematical model’ of nuclear war because while we have mathematical models of how fission and fusion works, we don’t have any for how likely it is that e.g. Iran’s leadership decides to start building nuclear weapons.
I think Tyler Cowen would accept that as sufficiently rigorous in that domain, and I believe that the earlier purely technical questions can be obtained from explicit models. One addition that could strengthen the model is to explicitly spell out different scenarios for each step (e.g. APS causes damage via autonomous weapons, economic disruption, etc). But the core framework seems sufficient as is, and also those concerns have been explained in other places.
What do you think?
Maybe we have different definitions of DSA: I was thinking of it in terms of ‘resistance is futile’ and you can dictate whatever terms you want because you have overwhelming advantage, not that you could eventually after a struggle win a difficult war by forcing your opponent to surrender and accept unfavorable terms.
If say the US of 1965 was dumped into post WW2 Earth it would have the ability to dictate whatever terms it wanted because it would be able to launch hundreds of ICBMS at enemy cities at will. If the real US of 1949 had started a war against the Soviets it would probably have been able to cripple an advance into western Europe but likely wouldn’t have been able to get its bombers through to devastate enough of the soviet homeland with the few bombs they had.
Remember the soviets did just lose a huge percentage of their population and industry in WW2 and kept fighting. The fact that it’s at all debatable who would have won if WW3 started in the late 1940s at all (see e.g. here) makes me think nuclear weapons weren’t at that time a DSA producer.