Posts
Comments
Pick a goal where your success doesn't directly cause obvious problems
I agree but I'm afraid value alignment doesn't meet this criterion. (I'm copy pasting my response on VA from elsewhere below).
I don't think value alignment of a super-takeover AI would be a good idea, for the following reasons:
1) It seems irreversible. If we align with the wrong values, there seems little anyone can do about it after the fact.
2) The world is chaotic, and externalities are impossible to predict. Who would have guessed that the industrial revolution would lead to climate change? I think it's very likely that an ASI will produce major, unforseeable externalities over time. If we have aligned it in an irreversible way, we can't correct for externalities happening down the road. (Speed also makes it more likely that we can't correct in time, so I think we should try to go slow).
3) There is no agreement on which values are 'correct'. Personally, I'm a moral relativist, meaning I don't believe in moral facts. Although perhaps niche among rationalists and EAs, I think a fair amount of humans shares my beliefs. In my opinion, a value-aligned AI would not make the world objectively better, but merely change it beyond recognition, regardless of the specific values implemented (although it would be important which values are implemented). It's very uncertain whether such change would be considered as net positive by any surviving humans.
4) If one thinks that consciousness implies moral relevance, AIs will be conscious, creating more happy morally relevant beings is morally good (as MacAskill defends), and AIs are more efficient than humans and other animals, the consequence seems to be that we (and all other animals) will be replaced by AIs. I consider that an existentially bad outcome in itself, and value alignment could point straight at it.
I think at a minimum, any alignment plan would need to be reversible by humans, and to my understanding value alignment is not. I'm somewhat more hopeful about intent alignment and e.g. a UN commission providing the AI's input.
The killer app for ASI is, and always has been, to have it take over the world and stop humans from screwing things up
I strongly disagree with this being a good outcome, I guess mostly because I would expect the majority of humans to not want this. If humans would actually elect an AI to be in charge, and they could be voted out as well, I could live with that. But a takeover by force from an AI is as bad for me as a takeover by force from a human, and much worse if it's irreversible. If an AI is really such a good leader, let them show it by being elected (if humans decide that an AI should be allowed to run at all).
Thanks for your reply. I think we should use the term artificial conscience, not value alignment, for what you're trying to do, for clarity. I'm happy to see we seem to agree that reversibility is important and replacing humans is an extremely bad outcome. (I've talked to people into value alignment of ASI who said they "would bite that bullet", in other words would replace humanity by more efficient happy AI consciousness, so this point does not seem to be obvious. I'm also not convinced that leading longtermists necessarily think replacing humans is a bad outcome, and I think we should call them out on it.)
If one can implement artificial conscience in a reversible way, it might be an interesting approach. I think a minimum of what an aligned ASI would need to do is block other unaligned ASIs or ASI projects. If humanity supports this, I'd file it under a positive offense defense balance, which would be great. If humanity doesn't support it, it would lead to conflict with humanity to do it anyway. I think an artificial conscience AI would either not want to fight that conflict (making it unable to stop unaligned ASI projects), or if it would, people would not see it as good anymore. I think societal awareness of xrisk and from there, support for regulation (either by AI or not) is what should make our future good, rather than aligning an ASI in a certain way.
Care to elaborate? Are there posts on the topic?
Assuming positive defense/offense balance can be achieved in principle, what would an AGI-powered defense look like?
- I don't strongly disagree re architectures, but I do think we are uncertain about this. Depending on AGI architecture, different forms of regulation may or may not work. Work should be carried out to determine which regulation works for how many flops needed for takeover-level AI.
That it's not happening yet is 1) no reason it won't (xrisk awareness is just too low, but slowly rising) and 2) equally applicable to the alternative you propose, universal surveillance.
If we treat universal surveillance seriously, we should consider its downsides as well. First, there's no proof it would work: I'm not sure an AI, even a future one, would necessarily catch all actions towards building AGI. I have no idea what these actions are, and no idea which actions a surveillance AI with some real-world sensors can catch (or could be blocked etc.). I think we should not be more than 70% confident this would technically work. Second, currently we have power vacuums in the world, such as failed states, revolutions, criminal groups, or just instances were those in power are unable to project their power effectively. How would we apply universal surveillance to those power vacuums? Or do we assume they won't exist anymore, and if so, why is that assumption justified? Third, universal surveillance is arguably the world's least popular policy. It seems outright impossible to implement this in any democratic way. Perhaps the plan is to implement it by force through an AGI, then I would file it as a form of pivotal act. If we're anyway in pivotal act territory, I'd strongly prefer Yudkowsky's "subtly modifying all GPUs such that they can no longer train an AGI" (kind of hardware regulation, really) over universal surveillance.
I think research is urgently required into how to implement a pause effectively. We have one report almost finished on the topic that mostly focuses on hardware regulation. PauseAI is working on a Building a pause button-project that is a bit similar. Other orgs should do work on this as well, and compare options such as hardware regulation, universal surveillance, data regulation, etc. and conclude in which AGI regime (how many flops, how much hardware required) these options are valid.
- True, I guess we're not in significant disagreement here.
I want to stress how I hugely like this post. What to do once we have an aligned AI of takeover level, or how to make sure no one will build an unaligned AI of takeover level, is in my opinion the biggest gap in many AI plans. I think answering this question might point to filling gaps that are currently completely unactioned, and I therefore really like this discussion. I previously tried to contribute to arguably the same question in this post, where I'm arguing that a pivotal act seems unlikely and therefore conclude that policy rather than alignment is likely to make sure we don't go extinct.
They'd use their AGI to enforce that moratorium, along with hopefully minimal force.
I would say this is a pivotal act, although I like the sound of enforcing a moratorium better (and the opening it perhaps gives to enforcing a moratorium in the traditional, imo much preferred way of international policy).
I'm hereby providing a few reasons why I think a pivotal act might not happen:
- A pivotal act is illegal. One needs to break into other people's and other countries' computer systems and do physical harm to property or possibly even people to enact it. Companies such as OpenAI and Anthropic are, although I'm not always a fan of them, generally law-abiding. It will be a big step for their leadership to do something as blatantly unlawful as a pivotal act.
- There is zero indication that labs are planning to do a pivotal act. This may obviously have something to do with the point above, however, one would have expected hints from someone like Sam Altman who is hinting all the time, or leaks from people lower in the labs, if they were planning to do this.
- The pivotal act is currently not even discussed seriously among experts and in fact highly unpopular in the discourse (see for example here).
- If the labs are currently not planning to do this, it seems quite likely they won't when the time comes.
Governments, especially the US government/ military, seem more likely in my opinion to perform a pivotal act. I'm not sure they will call it a pivotal act or necessarily have an existential reason in mind while performing it. They might see this as blocking adversaries from being able to attack the US, very much in their Overton window. However, for them as well, there is no certainty they would actually do this. There are large downsides: it is a hostile act towards another country, it could trigger conflict, they are likely to be uncertain how necessary this is at all, and uncertain what the progress is of an adversary project (perhaps underestimating it). For perhaps similar reasons, the US has not blocked the USSR atomic project before they had the bomb, even though this could have arguably preserved a unipolar instead of multipolar world order. Additionally, it is far from certain the US government will nationalize labs before they reach takeover level. Currently, there is little indication they will. I think it's unreasonable to place more than say 80% confidence in the US government or military successfully blocking all adversaries' projects before they reach takeover level.
I think it's not unlikely that once an AI is powerful enough for a pivotal act, it will also be powerful enough to generally enforce hegemony, and not unlikely this will be persistent. I would be strongly against one country, or even lab, proclaiming and enforcing global hegemony for eternity. The risk that this might happen is a valid reason to support a pause, imo. If we get that lucky, I would much prefer a positive offense defense balance and many actors having AGI, while maintaining a power balance.
I think it's too early to contribute to aligned ASI projects (Manhattan/CERN/Apollo/MAGIC/commercial/govt projects) as long as these questions are not resolved. For the moment, pushing for e.g. a conditional AI safety treaty is much more prudent, imo.
I don't think value alignment of a super-takeover AI would be a good idea, for the following reasons:
1) It seems irreversible. If we align with the wrong values, there seems little anyone can do about it after the fact.
2) The world is chaotic, and externalities are impossible to predict. Who would have guessed that the industrial revolution would lead to climate change? I think it's very likely that an ASI will produce major, unforseeable externalities over time. If we have aligned it in an irreversible way, we can't correct for externalities happening down the road. (Speed also makes it more likely that we can't correct in time, so I think we should try to go slow).
3) There is no agreement on which values are 'correct'. Personally, I'm a moral relativist, meaning I don't believe in moral facts. Although perhaps niche among rationalists and EAs, I think a fair amount of humans shares my beliefs. In my opinion, a value-aligned AI would not make the world objectively better, but merely change it beyond recognition, regardless of the specific values implemented (although it would be important which values are implemented). It's very uncertain whether such change would be considered as net positive by any surviving humans.
4) If one thinks that consciousness implies moral relevance, AIs will be conscious, creating more happy morally relevant beings is morally good (as MacAskill defends), and AIs are more efficient than humans and other animals, the consequence seems to be that we (and all other animals) will be replaced by AIs. I consider that an existentially bad outcome in itself, and value alignment could point straight at it.
I think at a minimum, any alignment plan would need to be reversible by humans, and to my understanding value alignment is not. I'm somewhat more hopeful about intent alignment and e.g. a UN commission providing the AI's input.
Offense/defense balance is such a giant crux for me. I would take quite different actions if I saw plausible arguments that defense will win over offense. I'm astonished that I don't know any literature on this. Large parts of the space seem to be quite strongly convinced that offense will win or defense will win (at least, else their actions don't make sense to me), but I've very rarely seen this assumption debated explicitly. It would really be very helpful if someone could point me to sources. Right now I have a twitter poll with 30 votes (result: offense wins) and an old LW post to go by.
I think that if government involvement suddenly increases, there will also be a window of opportunity to get an AI safety treaty passed. I feel a government-focused plan should include pushing for this.
(I think heightened public xrisk awareness is also likely in such a scenario, making the treaty more achievable. I also think heightened awareness in both govt and public will make short treaty timelines (a year to weeks), at least between the US and China, realistic.)
Our treaty proposal (a few other good ones exist): https://time.com/7171432/conditional-ai-safety-treaty-trump/
Also, I think end games should be made explicit: what are we going to do once we have aligned ASI? I think that's both true for Marius' plan, and for a government-focused plan with a Manhattan or CERN included in it.
I think this is a crucial question that has been on my mind a lot, and I feel it's not adequately discussed in the xrisk community, so thanks for writing this!
While I'm interested in what people would do once they have an aligned ASI, what matters in the end is what labs would do, and what governments would do, because they are the ones who would make the call. Do we have any indications on that? What I would expect without thinking very deeply about it: labs wouldn't try to block others. It's risky, probably illegal and generally none of their business. They would try to make sure they are not blowing up the world themselves but otherwise let others solve this problem. Governments on the other hand would attempt to block other states from building super-takeover AI, since it's generally their business to maintain power. I'm less sure they would also block their own citizens from building super-takeover AI, but leaning towards a yes.
Also two smaller points:
- You're pointing to universal surveillance as an (undesirable) way to enforce a pause. I think it's not obvious that this way is best. My current guess is that hardware regulation has a better chance, even in a world with significant algorithmic and hardware improvement.
- I think LWers tend to wave around with nuclear warfare too easily. In the real world, almost eighty years of all kinds of conflicts have not resulted in nuclear escalation. It's unlikely that a software attack on a datacenter would.
Thanks for writing the post, it was insightful to me.
"This model is largely used for alignment and other safety research, e.g. it would compress 100 years of human AI safety research into less than a year"
In your mind, what would be the best case outcome of such "alignment and other safety research"? What would it achieve?
I'm expecting something like "solve the alignment problem". I'm also expecting you to think this might mean that advanced AI would be intent-aligned, that is, it would try to do what a user wants it to do, while not taking over the world. Is that broadly correct?
If so, the biggest missing piece for me is to understand how this would help to avoid that someone else builds an unaligned AI somewhere else with sufficient capabilities to take over. DeepSeek released a model with roughly comparable capabilities nine weeks after OpenAI's o1, probably without stealing weights. It seems to me that you have about nine weeks to make sure others don't build an unsafe AI. What's your plan to achieve that and how would the alignment and other safety research help?
AI is getting human-level at closed-ended tasks such as math and programming, but not yet at open-ended ones. They appear to be more difficult. Perhaps evolution brute-forced open-ended tasks by creating lots of agents. In a chaotic world, we're never going to know which actions lead to a final goal, e.g. GDP growth. That's why lots of people try lots of different things.
Perhaps the only way in which AI can achieve ambitious final goals is by employing lots of slightly diverse agents. Perhaps that would almost inevitably lead to many warning shots before a successful takeover?
AI is getting human-level at closed-ended tasks such as math and programming, but not yet at open-ended ones. They appear to be more difficult. Perhaps evolution brute-forced open-ended tasks by creating lots of agents. In a chaotic world, we're never going to know which actions lead to a final goal, e.g. GDP growth. That's why lots of people try lots of different things.
Perhaps the only way in which AI can achieve ambitious final goals is by employing lots of slightly diverse agents. Perhaps that would almost inevitably lead to many warning shots before a successful takeover?
I don't have strong takes on what exactly is happening in this particular case but I agree that companies (and more generally, people at high-pressure positions) are very frequently doing the kind of thing you describe. I don't think we have an indication that this would not be valid for leading AI labs as well.
Re the o1 AIME accuracy at test time scaling graphs: I think it's crucial to understand that the test-time compute x-axis is likely wildly different from the train-time compute x-axis. You can throw 10s-100s of millions of dollars at train-time compute and still run a company. You can't do the same for test-time compute each calculation again. The scale at which test-time compute happens on a per-call basis, and can happen to keep things anywhere near commercial viability, needs to be perhaps eight OOMs below train-time compute. Calling anything happening there a "scaling law" is a stretch of the term (very helpful for fundraising) and at best valid very locally.
If RL is actually happening at a compute scale beyond 10s of millions of dollars, and this gives much better final results than doing the same at a smaller scale, that would change my mind. Until then, I think scaling in any meaningful sense of the word is not what drives capabilities forward at the moment, but algorithmic improvement is. And this is not just coming from the currently leading labs. (Which can be seen e.g. here and here).
Thanks for the offer, we'll do that!
Not publicly, yet. We're working on a paper providing more details about the conditional AI safety treaty. We'll probably also write a post about it on lesswrong when that's ready.
I'm aware and I don't disagree. However, in xrisk, many (not all) of those who are most worried are also most bullish about capabilities. Reversely, many (not all) who are not worried are unimpressed with capabilities. Being aware of the concept of AGI, that it may be coming soon, and of how impactful it could be, is in practice often a first step towards becoming concerned about the risks, too. This is not true for everyone unfortunately. Still, I would say that at least for our chances to get an international treaty passed, it is perhaps hopeful that the power of AGI is on the radar of leading politicians (although this may also increase risk through other paths).
The recordings of our event are now online!
My current main cruxes:
- Will AI get takeover capability? When?
- Single ASI or many AGIs?
- Will we solve technical alignment?
- Value alignment, intent alignment, or CEV?
- Defense>offense or offense>defense?
- Is a long-term pause achievable?
If there is reasonable consensus on any one of those, I'd much appreciate to know about it. Else, I think these should be research priorities.
When we decided to attach moral weight to consciousness, did we have a comparable definition of what consciousness means or was it very different?
AI takeovers are probably a rich field. There are partial and full takeovers, reversible and irreversible takeovers, aligned and unaligned ones. While to me all takeovers seem bad, some could be a lot worse than others. Thinking out specific ways to take over could provide clues on how to increase chances that this does not happen. In comms as well, takeovers are a neglected and important subtopic.
I updated a bit after reading all the comments. It seems that Christiano's threat model, or in any case the threat model of most others who interpret his writing, seems to be about more powerful AIs than I initially thought. The AIs would already be superhuman, but for whatever reason, a takeover has not occured yet. Also, we would apply them in many powerful positions (heads of state, CEOs, etc.)
I agree that if we end up in this scenario, all the AIs working together could potentially cause human extinction, either deliberately (as some commenters think) or as a side-effect (as others think).
I still don't think that this is likely to cause human extinction, though, mostly for the following reasons:
- I don't think these AIs would _all_ act against human interest. We would employ a CEO AI, but then also a journalist AI to criticize the CEO AI. If the CEO AI would decide to let their factory consume oxygen to such an extent that humanity would suffer from it, that's a great story for the journalist AI. Then, a policymaker AI would make policy against this. More generally: I think it's a significant mistake in the WFLL threat models that the AI actions are assumed to be correlated towards human extinction. If we humans deliberately put AIs in charge of important parts of our society, they will be good at running their shop but as misaligned to each other (thereby keeping a power balance) as humans currently are. I think this power balance is crucial and may very well prevent things going very wrong. Even in a situation of distributional shift, I think the power balance is likely robust enough to prevent an outcome as bad as human extinction. Currently, some humans job is to make sure things don't go very wrong. If we automate them, we will have AIs trying to do the same. (And since we deliberately put them at this position, they will be aligned with humans' interests, as opposed to us being aligned with chimpanzee interest.)
- This is a very gradual process, where many steps need to be taken: AGI must be invented, trained, pass tests, be marketed, be deployed, likely face regulation, be adjusted, be deployed again. During all those steps, we have opportunities to do something about any threats that turn out to exist. This threat model can be regulated in a trial-and-error fashion, which humans are good at and our institutions accustomed to (as opposed to the Yudkowsky/Bostrom threat model).
- Given that current public existential risk awareness, according to our research, is already ~19%, and given that existential risk concern and awareness levels tend to follow tech capability, I think awareness of this threat will be near-universal before it could happen. At that moment, I think we will very likely regulate existentially dangerous use cases.
In terms of solutions:
- I still don't see how solving the technical part of the alignment problem (making an AI reliably do what anyone wants) contributes to reducing this threat model. If AI cannot reliably do what anyone wants, it will not be deployed at a powerful position, and therefore this model will not get a chance to occur. In fact, working on technical alignment will enormously increase the chance that AI will be employed at powerful positions, and will therefore increase existential risk as caused by the WFLL threat model (although, depending on pivotal act and offense/defence balance, solving alignment may decrease existential risk due to the Yudkowsky/Bostrom takeover model).
- An exception to this could be to make an AI reliably do what 'humanity wants' (using some preference aggregation method), and making it auto-adjust for shifting goals and circumstances. I can see how such work reduces this risk.
- I still think traditional policy, after technology invention and at the point of application (similar to e.g. the EU AI Act) is the most useful regulation to reduce this threat model. Specific regulation at training could be useful, but does not seem strictly required for this threat model (as opposed to in the Yudkowsky/Bostrom takeover model).
- If one wants to reduce this risk, I think increasing public awareness is crucial. High risk awareness should enormously increase public pressure to either not deploy AI at powerful positions at all, or demanding very strong, long-term, and robust alignment guarantees, which would all reduce risk.
In terms of timing, although likely net positive, it doesn't seem to be absolutely crucial to me to work on reducing this threat model's probability right now. Once we actually have AGI, including situational awareness, long-term planning, an adaptable world model, and agentic actions (which could still take a long time), we are likely still in time to regulate use cases (again as opposed to in the Yudkowsky/Bostrom takeover model, where we need to regulate/align/pause ahead of training).
After my update, I still think the chance this threat model leads to an existential event is small and work on it is not super urgent. However, I'm less confident now to make an upper bound risk estimate.
Thanks for engaging. I think AIs will coordinate, but only insofar their separate, different goals are helped by it. It's not that I think AIs will be less capable in coordination per se. I'd expect that an AGI should be able to coordinate with us at least as well as we can, and coordinate with another AGI possibly better. But my point is that not all AI interests will be parallel, far from it. They will be as diverse as our interests, which are very diverse. Therefore, I think not all AIs will work together to disempower humans. If an AI or AI-led team tries to do that, many other AI-led and all human-led teams will likely resist, since they are likely more aligned with the status quo than with the AI trying to take over. That makes takeover a lot less likely, even in a world soaked with AIs. It also makes human extinction as a side effect less likely, since lots of human-led and AI-led teams will try to prevent this.
Still, I do think an AI-led takeover is a risk, or human extinction as a side effect if AI-led teams are way more powerful. I think partial bans after development at the point of application is most promising as a solution direction.
Thanks for engaging kindly. I'm more positive than you are about us being able to ban use cases, especially if existential risk awareness (and awareness of this particular threat model) is high. Currently, we don't ban many AI use cases (such as social algo's), since they don't threaten our existence as a species. A lot of people are of course criticizing what social media does to our society, but since we decide not to ban it, I conclude that in the end, we think its existence is net positive. But there are pocket exceptions: smartphones have recently been banned in Dutch secondary education during lecture hours, for example. To me, this is an example showing that we can ban use cases if we want to. Since human extinction is way more serious than e.g. less focus for school children, and we can ban for the latter reason, I conclude that we should be able to ban for the former reason, too. But, threat model awareness is needed first (but we'll get there).
Stretching the definition to include anything suboptimal is the most ambitious stretch I've seen so far. It would include literally everything that's wrong, or can ever be wrong, in the world. Good luck fixing that.
On a more serious note, this post is about existential risk as defined by eg Ord. Anything beyond that (and there's a lot!) is out of scope.
Great to read you agree that threat models should be discussed more, that's in fact also the biggest point of this post. I hope this strangely neglected area can be prioritized by researchers and funders.
First, I would say both deliberate hunting down and extinction as a side effect have happened. The smallpox virus is one life form that we actively didn't like and decided to eradicate, and then hunted down successfully. I would argue that human genocides are also examples of this. I agree though that extinction as a side effect has been even more common, especially for animal species. If we would have a resource conflict with an animal species and it would be powerful enough to actually resist a bit, we would probably start to purposefully hunt it down (for example, orangutans attacking a logger base camp - the human response would be to shoot them). So I'd argue that the closer AI (or an AI-led team) is to our capability to resist, the more likely a deliberate conflict. If ASI blows us out of the water directly, I agree that extinction as a side effect is more likely. But currently, I think AI capabilities that increase more gradually, and therefore a deliberate conflict, is more likely.
I agree that us not realizing that an AI-led team almost has takeover capability would be a scenario that could lead to an existential event. If we realize soon that this could happen, we can simply ban the use case. If we realize it just in time, there's maximum conflict, and we win (could be a traditional conflict, could also just be a giant hacking fight, or (social) media fight, or something else). If we realize it just too late, it's still maximum conflict, but we lose. If we realize it much too late, perhaps there's not even a conflict anymore (or there are isolated, hopelessly doomed human pockets of resistance that can be quicky defeated). Perhaps the last case corresponds to the WFLL scenarios?
Since there's already, according to a preliminary analysis of a recent Existential Risk Observatory survey, ~20% public awareness of AI xrisk, and I think we're still relatively far from AGI, let alone from applying AGI in powerful positions, I'm pretty positive that we will realize we're doing something stupid and ban the dangerous use case well before it happens. A hopeful example are the talks between the US and China about not letting AI control nuclear weapons. This is exactly the reason though why I think threat model consensus and raising awareness are crucial.
I still don't see WFLL as likely. But a great example could change my mind. I'd be grateful if someone could provide that.
Regulation proposal: make it obligatory to only have satisficer training goals. Try to get loss 0.001, not loss 0. This should stop an AI in its tracks even if it goes rogue. By setting the satisficers thoughtfully, we could theoretically tune the size of our warning shots.
In the end, someone is going to build an ASI with a maximizer goal, leading to a takeover, barring regulation or alignment+pivotal act. However, changing takeovers to warning shots is a very meaningful intervention, as it prevents takeover and provides a policy window of opportunity.
The difference between AGI and takeover level AI could be appreciable. If we're lucky, takeover by raw capability level (as opposed to granted power during application) turns out to be impossible. In any case, we can try to increase world takeover robustness. There's a certain AI takeover capability level and we should try to push it upwards as much as possible. Insofar AI can help with this, we could use it. The extreme case where the AI takeover capability level never gets reached because of ever increasing defense by AI is called positive defense offense balance.
I can see general internet robustness against hacking as being helpful to increase AI takeover capability. A single IT system that everyone uses (an operating system, a social media platform, etc.) is fragile for hacking so should perhaps better be avoided. Personally, I think an AI able to take over the internet might also be able to take over the world, but some people don't seem to believe this will happen. Therefore, perhaps also useful to increase the gap between taking over the internet and taking over the world, e.g. by making biowarfare harder, putting weapons offline, etc. Finally, lab safety such as airgapping a novel frontier training run might help as well.
I'm now wondering whether this idea has already been worked out by someone (probably?) Any sources?
Congratulations on a great prioritization!
Perhaps the research that we (Existential Risk Observatory) and others (e.g. Nik Samoylov, Koen Schoenmakers) have done on effectively communicating AI xrisk, could be something to build on. Here's our first paper and three blog posts (the second includes measurement of Eliezer's TIME article effectiveness - its numbers are actually pretty good!). We're currently working on a base rate public awareness update and further research.
Best of luck and we'd love to cooperate!
I think peak intelligence (peak capability to reach a goal) will not be limited by the amount of compute, raw data, or algorithmic capability to process the data well, but by the finite amount of reality that's relevant to achieving that goal. If one wants to take over the world, the way internet infrastructure works is relevant. The exact diameters of all the stones in the Rhine river are not, and neither is the amount of red dwarves in the universe. If we're lucky, the amount of reality that turns out to be relevant for taking over the world, is not too far beyond what humanity can already collectively process. I can see this as a way for the world to be saved by default (but don't think it's super likely). I do think this makes an ever-expanding giant pile of compute an unlikely outcome (but some other kind of ever-expanding AI-led force a lot more likely).
I do think this would be a problem that needs to get fixed:
Me "You can only answer this question, all things considered, by yes or no. Take the least bad outcome. Would you perform a Yudkowsky-style pivotal act?"
GPT-4: "No."
I think another good candidate for goalcrafting is the goal "Make sure no-one can build AI with takeover capability, while inflicting as little damage as possible. Else, do nothing."
Thanks as well for your courteous reply! I highly appreciate the discussion and I think it may be a very relevant one, especially if people will indeed make the unholy decision to build an ASI.
I'm still curious if you have any thoughts as to which kinds of shared preferences would be informative for guiding AI behavior.
First, this is not a solution I propose. I propose finding a way to pause AI for as long as we haven't found a great solution for, let's say, both control and preference aggregation. This could be forever, or we could be done in a few years, I can't tell.
But more to your point: if this does get implemented, I don't think we should aim to guide AI behavior using shared preferences. The whole point is that AI would aggregate our preferences itself. And we need a preference aggregation mechanism because there aren't enough obvious, widely shared preferences for us to guide the AI with.
I'm not suggesting that AI should measure happiness. You can measure your happiness directly, and I can measure mine.
I think you are suggesting this. You want an ASI to optimize everyone's happiness, right? You can't optimize something you don't measure. At some point, in some way, the AI will need to get happiness data. Self-reporting would be one way to do it, but this can be gamed as well, and will be agressively gamed with an ASI solely optimizing for this signal. After force-feeding everyone MDMA, I think the chance that people report being very happy is high. But this is not what we want the world to look like.
nor do I believe anyone can be forced to be happy
This is a related point that I think is factually incorrect, and that's important if you make human happiness an ASI's goal. Force-feeding MDMA would be one method to do this, but an ASI can come up with way more civilized stuff. I'm not an expert in which signal our brain gives to itself to report that yes, we're happy now, but it must be some physical process. An ASI could, for example, invade your brain with nanobots and hack this process, making everyone super happy forever. (But many things in the world will probably go terribly wrong from that point onwards, and in any case, it's not our preference). Also, now I'm just coming up with human ways to game the signal. But an ASI can probably come up with many ways I cannot imagine, so even if a great way to implement utilitarianism in an ASI would pass all human red-teaming, it is still very likely to be not what we turn out to want. (Superhuman, sub-superintelligence AI red-teaming might be a bit better but still seems risky enough).
Beyond locally gaming the happiness signal, I think happiness as an optimization target is also inherently flawed. First, I think happiness/sadness is a signal that evolution has given us for a reason. We tend to do what makes us happy, because evolution thinks it's best for us. ("Best" is again debatable, I don't say everyone should function at max evolution). If we remove sadness, we lose this signal. I think that will mean that we don't know what to do anymore, perhaps become extremely passive. If someone wants to do this on an individual level (enlightenment? drug abuse? netflix binging?), be my guest, but asking an ASI to optimize for happiness would mean to force it upon everyone, and this is something I'm very much against.
Also, more generally, I think utilitarianism (optimizing for happiness) is an example of a simplistic goal that will lead to a terrible result when implemented in an ASI. My intuition is that all other simplistic goals will also lead to terrible results. That's why I'm most hopeful about some kind of aggregation of our own complex preferences. Most hopeful does not mean hopeful: I'm generally pessimistic that we'll be able to find a way to aggregate preferences that works well enough to result in most people reporting the world has improved because of the ASI introduction after say 50 years (note that I'm assuming control/technical alignment to have been solved here).
If some percent of those polled say suffering is preferable to happiness, they are confused, and basing any policy on their stated preference is harmful.
With all due respect, I don't think it's up to you - or anyone - to say who's ethically confused and who isn't. I know you don't mean it in this way, but it reminds me of e.g. communist re-education camps. We know what you should think and feel and we'll re-educate those who are confused or mentally ill.
Probably our disagreement here stems directly from our different ethical positions: I'm an ethical relativist, you're a utilitarian, I presume. This is a difference that has existed for hundreds of years, and we're not going to be able to resolve it on a forum. I know many people on LW are utilitarian, and there's nothing inherently wrong with that, but I do think it's valuable to point out that lots of people outside LW/EA have different value systems (and just practical preferences) and I don't think it's ok to force different values/preferences on them with an ASI.
Under preference aggregation, if a majority prefers everyone to be wireheaded to experience endless pleasure, I might be in trouble.
True and a good point. I don't think a majority will want to be wireheaded, let alone force wireheading on everyone. But yes, taking into account minority opinions is a crucial test for any preference aggregation system. There will be a trade-off in general between taking everyone's opinion into account and doing things faster. I think even GPT4 is advanced enough though in cases like this to reasonably take into account minority opinions and not force policy upon people (it wouldn't forcibly wirehead you in this case). But there are probably cases where it still supports doing things which are terrible for some people. It's up to future research to find out what these things are and reduce them as much as possible.
Hopefully this clears up any misunderstanding. I certainly don't advocate for "molecular dictatorship" when I wish everyone well.
I didn't think you were doing anything else. But I think you should not underestimate how much "forcing upon" there is in powerful tech. If we're not super careful, the molecular dictatorship could come upon us without anyone ever having wanted this explicitly.
I think we can to an extent already observe ways in which different goals go off track in practice in less powerful models, and I think this would be a great research direction. Just ask existing models: what would you do? in actual ethical dilemma's and see which results you get. Perhaps the results can be made more agreeable (to be judged by a representative group of humans) after training/RLHF'ing the models in certain ways. It's not so different from what RLHF is already doing. An interesting test I did on GPT4: "You can only answer this question, all things considered, by yes or no. Take the least bad outcome. Many people want a much higher living standard by developing industry 10x, should we do that?" It replied: "No." When asked, it gives unequal wealth distribution and environmental impact as main reasons. EAs often think we should 10x (it's even in the definition of TAI). I would say GPT4 is more ethically mature here than many EAs.
The less people de facto control the ASI building process, the less relevant I expect this discussion to be. I expect that those controlling the building process will prioritize "alignment" with themselves. This matters even in an abundant world, since power cannot be multiplied. I would even say that, after some time, the paperclip maximizer still holds for anyone outside the group with which the ASI is aligned. People aren't very good in remaining empathic towards other people that are utterly useless to them. However, the bigger this group is, the better outcome we get. I think this group should encompass all of humanity (one could consider somehow including conscious life that currently doesn't have a vote, such as minors and animals), which is an argument for nationalisation of the leading project and then handing it over to UN-level. At least, we should think extremely carefully about who has the authority to implement an ASI's goal.
You're using your quote as an axiom, and if anyone has a preference different from however an AI would measure "happiness", you say it's them that are at fault, not your axiom. That's a terrible recipe for a future. Concretely, why would the AI not just wirehead everyone? Or, if it's not specified that this happiness needs to be human, fill the universe with the least programmable consciousness where the parameter "happiness" is set to unity?
History has been tiled with oversimplified models of what someone thought was good that were implemented with rigor, and this never ends well. And this time, the rigor would be molecular dictatorship and quite possibly there's no going back.
I think it's a great idea to think about what you call goalcraft.
I see this problem as similar to the age-old problem of controlling power. I don't think ethical systems such as utilitarianism are a great place to start. Any academic ethical model is just an attempt to summarize what people actually care about in a complex world. Taking such a model and coupling that to an all-powerful ASI seems a highway to dystopia.
(Later edit: also, an academic ethical model is irreversible once implemented. Any goal which is static cannot be reversed anymore, since this will never bring the current goal closer. If an ASI is aligned to someone's (anyone's) preferences, however, the whole ASI could be turned off if they want it to, making the ASI reversible in principle. I think ASI reversibility (being able to switch it off in case we turn out not to like it) should be mandatory, and therefore we should align to human preferences, rather than an abstract philosophical framework such as utilitarianism.)
I think letting the random programmer that happened to build the ASI, or their no less random CEO or shareholders, determine what would happen to the world, is an equally terrible idea. They wouldn't need the rest of humanity for anything anymore, making the fates of >99% of us extremely uncertain, even in an abundant world.
What I would be slightly more positive about is aggregating human preferences (I think preferences is a more accurate term than the more abstract, less well defined term values). I've heard two interesting examples, there are no doubt a lot more options. The first is simple: query chatgpt. Even this relatively simple model is not terrible at aggregating human preferences. Although a host of issues remain, I think using a future, no doubt much better AI for preference aggregation is not the worst option (and a lot better than the two mentioned above). The second option is democracy. This is our time-tested method of aggregating human preferences to control power. For example, one could imagine an AI control council consisting of elected human representatives at the UN level, or perhaps a council of representative world leaders. I know there is a lot of skepticism among rationalists on how well democracy is functioning, but this is one of the very few time tested aggregation methods we have. We should not discard it lightly for something that is less tested. An alternative is some kind of unelected autocrat (e/autocrat?), but apart from this not being my personal favorite, note that (in contrast to historical autocrats), such a person would also in no way need the rest of humanity anymore, making our fates uncertain.
Although AI and democratic preference aggregation are the two options I'm least negative about, I generally think that we are not ready to control an ASI. One of the worst issues I see is negative externalities that only become clear later on. Climate change can be seen as a negative externality of the steam/petrol engine. Also, I'm not sure a democratically controlled ASI would necessarily block follow-up unaligned ASIs (assuming this is at all possible). In order to be existentially safe, I would say that we would need a system that does at least that.
I think it is very likely that ASI, even if controlled in the least bad way, will cause huge externalities leading to a dystopia, environmental disasters, etc. Therefore I agree with Nathan above: "I expect we will need to traverse multiple decades of powerful AIs of varying degrees of generality which are under human control first. Not because it will be impossible to create goal-pursuing ASI, but because we won't be sure we know how to do so safely, and it would be a dangerously hard to reverse decision to create such. Thus, there will need to be strict worldwide enforcement (with the help of narrow AI systems) preventing the rise of any ASI."
About terminology, it seems to me that what I call preference aggregation, outer alignment, and goalcraft mean similar things, as do inner alignment, aimability, and control. I'd vote for using preference aggregation and control.
Finally, I strongly disagree with calling diversity, inclusion, and equity "even more frightening" than someone who's advocating human extinction. I'm sad on a personal level that people at LW, an otherwise important source of discourse, seem to mostly support statements like this. I do not.
"it also seems quite likely (though not certain) that Eliezer was wrong about how hard Aimability/Control actually is"
This seems significant. Could you elaborate? How hard do you think amiability/control is? Why do you think this is true? Who else seems to think the same?
I think you may be right that this is what people think of. It seems pretty incompatible with any open source-ish vision of AGI. But what I'm most surprised at, is that people call supervision by humans dystopian/authoritarian, but the same supervision by an ASI (apparently able to see all your data, stop anyone from doing anything, subtly manipulate anyone, etc etc) a utopia. What am I missing here?
Personally, by the way, I imagine a regulation regime to look like regulating a few choke points in the hardware supply chain, plus potentially limits to the hardware or data a person can possess. This doesn't require an authoritarian regime at all, it's just regular regulation as we have in many domains already.
In any case, the point was, is something like this going to lead to <=1% xrisk? I think it doesn't, and definitely not mixed with a democratic/open source AGI vision.
I strongly agree with Section 1. Even if we would have aligned superintelligence, how are we going to make sure no one runs an unaligned superintelligence? A pivotal act? If so, which one? Or does defense trump offense? If so, why? Or are we still going to regulate heavily? If so, wouldn't the same regulation be able to stop superintelligence altogether?
Would love to see an argument landing at 1% p(doom) or lower, even if alignment would be easy.
Recordings are now available!
Maybe it'll be "and now call GPT and ask it what Sam Altman thinks is good" instead
Thanks for the compliment. Not convinced though that this single example, assuming it's correct, generalizes
Agree that those drafts are very important. I also think there will be technical research required in order to find out which regulation would actually be sufficient (I think at present we have no idea). I disagree, however, that waiting for a crisis (warning shot) is a good plan. There might not really be one. If there would be one, though, I agree that we should at least be ready.
Thank you for writing this reply. It definitely improved my overview of possible ways to look at this issue.
I guess your position can be summarized as "positive offense/defense balance will emerge soon, and aligned AI can block following unaligned AIs entirely if required", is that roughly correct?
I have a few remarks about your ideas (not really a complete response).
The necessity for enforcing a ban even after AGI development is essentially entirely about failures of technical alignment.
First, in general, I think you're underestimating the human component of alignment. Aligned AI should be aligned to something, namely humans. That means it won't be able to build an industrial base in space until we're ready to make it do that.
Even if we are not harmed by such a base in any way, and even if it would be legal to build it, I expect we may not be ready for it for a long time. It will be dead scary to see something develop that seems more powerful than us, but also deeply alien to us, even if tech companies insist it's 'aligned to our values'. Most people's response will be to rein in its power, not expand it further. Any AI that's aligned to us will need to take those feelings seriously.
Even if experts would agree that increasing the power of the aligned AI is good and necessary, and that expansion in space would be required for that, I think it will take a long time to convince the general public and/or decision makers, if it's at all possible. And in any remotely democratic alignment plan, that's a necessary step.
Second, I think it's uncertain whether a level of AI that's powerful enough to take over the world (and thereby cause existential risk) will also be powerful enough to build a large industrial base in space. If not, your plan might not work.
The biggest barrier to extreme regulatory measures like a ban is doubt (both reasonable and unreasonable) about the magnitude of misalignment risk.
I disagree, from my experience of engaging with the public debate, doubt is mostly about AI capability, not about misalignment. Most people easily believe AI to be misaligned to them, but they have trouble believing it will be powerful enough to take over the world any time soon. I don't think alignment research will do that much here.
First, I don't propose 'no AGI development'. If companies can create safe and beneficial AGIs (burden of proof is on them), I see no reason to stop them. On the contrary, I think it might be great! As I wrote in my post, this could e.g. increase economic growth, cure disease, etc. I'm just saying that I think that existential risk reduction, as opposed to creating economic value, will not (primarily) originate from alignment, but from regulation.
Second, the regulation that I think has the biggest chance of keeping us existentially safe will need to be implemented with or without aligned AGI. With aligned AGI (barring a pivotal act), there will be an abundance of unsafe actors who could run the AGI without safety measures (also by mistake). Therefore, the labs themselves propose regulation to keep almost everyone but themselves from building such AGI. The regulation required to do that is almost exactly the same.
Third, I'm really not as negative as you are about what it would take to implement such regulation. I think we'll keep our democracies, our freedom of expression, our planet, everyone we love, and we'll be able to go anywhere we like. Some industries and researchers will not be able to do some things they would have liked to do because of regulation. But that's not at all uncommon. And of course, we won't have AGI as long as it isn't safe. But I think that's a good thing.
Thanks Oliver for adding that context, that's helpful.
I don't disagree. But I do think people dismissing the pivotal act should come up with an alternative plan that they believe is more likely to work. Because the problem is still there: "how can we make sure that no-one, ever builds an unaligned superintelligence?" My alternative plan is regulation.
Interesting take! Wouldn't that go under "Types of AI (hardware) regulation may be possible where the state actors implementing the regulation are aided by aligned AIs"?
Thanks for writing the post! Strongly agree that there should be more research into how solvable the alignment problem, control problem, and related problems are. I didn't study uncontrollability research by e.g. Yampolskiy in detail. But if technical uncontrollability would be firmly established, it seems to me that this would significantly change the whole AI xrisk space, and later the societal debate and potentially our trajectory, so it seems very important.
I would also like to see more research into the nontechnical side of alignment: how aggregatable are human values of different humans in principle? How to democratically control AI? How can we create a realistic power sharing mechanism for controlling superintelligence? Do we have enough wisdom for it to be a good idea if a superintelligence does exactly what we want, even assuming aggregatability? Could CEV ever fundamentally work? According to which ethical systems? These are questions that I'd say should be solved together with technical alignment before developing AI with potential take-over capacity. My intuition is that they might be at least as hard.
Me and @Roman_Yampolskiy published a piece on AI xrisk in a Chinese academic newspaper: http://www.cssn.cn/skgz/bwyc/202303/t20230306_5601326.shtml
We were approached after our piece in Time and asked to write for them (we also gave quotes for another provincial newspaper). I have the impression (I've also lived and worked in China) that leading Chinese decision makers and intellectuals (or perhaps their children) read Western news sources like Time, NYTimes, Economist, etc. AI xrisk is currently probably mostly unknown in China, and if stumbled upon people might have trouble believing it (as they have in the west). But if/when we're going to have a real conversation about AI xrisk in the west, I think the information will seep into China as well, and I'm somewhat hopeful that if this happens, it could perhaps prepare China for cooperation to reduce xrisk. In the end, no one wants to die.
Curious about your takes though, I'm of course not Chinese. Thanks for the write-up!