[dead]
The nitter link didn't appear to provide much along those lines. There were a few arguments that it was possible, which the Nitter OP admits is "very weak;" other than that, there's a link to a wiki page making claims like "Finding goals that aren’t extinction-level bad and are relatively useful appears to be hard" when in observable reality asking ChatGPT to maximize paperclip production does not in fact lead to ChatGPT attempting to turn all life on Earth into paperclips (nor does asking the open source LLMs result in that behavior out of the box either), and instead leads to the LLMs making fairly reasonable proposals that understand the context of the goal ("maximize paperclips to make money, but don't kill everyone," where the latter doesn't actually need to be said for the LLM to understand the goal).
I agree with you that current publicly available LLMs do not pose an existential risk to humanity. On the other hand I believe there is a better than 10% chance that the cutting edge LLMs of 2044 will be very powerful.
Do you believe (A) that LLMs are unlikely to become powerful in the short term, and/or (B) that if LLMs become powerful, then they are likely to be safe even without a significant and concerted alignment effort?
IMO even if LLMs are extremely unlikely to become powerful in the short term, then I still might be better off if LLM development is banned, ie:
P1: Humans are close to developing powerful non-LLM AI systems.
P2: Humans are not close to developing techniques for safely using powerful AI systems.
P3: If governments ban AI development, then the speed of AI capabilities development will be significantly reduced.
P4: It is a waste of scarce expertise and political capital to focus on making an LLM carve out in AI regulation legislation.
C: If it is extremely unlikely that LLMs will become powerful in the near future, then I am made much better off if governments ban all AI capabilities research (including LLMs).Arguing about theoretical AI models 30 years from now that might or might not be dangerous doesn't seem very convincing to me, since we don't know what they'll be based on or how they'll work — researchers today aren't even sure LLMs can scale to super-human intelligence. Similarly, pre-LLMs many safetyist orgs took the "paperclip problem" very seriously, when it's quite clear now that even the not-very-intelligent LLMs of today are capable of understanding the implicit context of a goal like that and won't seriously propose extinguishing humanity as a mechanism to improve paperclip production. Anthropic was formed in part because people thought gpt-3.5-turbo was existentially risky! And I don't think anyone today entertains that thought seriously, to put it lightly.
Trying to ban AI now due to supposed existential risks of systems in the future that don't currently exist and we don't know how to build (and we don't know if the failure modes proposed by the safety orgs will actually exist) seems like putting the cart well before the horse.
The argument starts with a hypothetical ("there is a possible artificial agent"), and it fails to be scary: there are (apparently) already humans that can kill 70% of humanity, and yet most of humanity is still alive. So an AGI that could also do it is not implicitly scarier.
The final twitter thread is basically a thread of people saying "no, there is no canonical, well-formulated argument for AGI catastrophe", so I'm not sure why you shared it.
As I understand it, Yud is actually providing a counterexample to a premise that other people are using to argue that humans will probably not be disempowered by AI systems. The relevant argument looks like this:
P1: If intelligent system A cannot give a detailed account of how it would be bested by a more intelligent system B, then A will not be bested by B.
P2: Humans (so far) cannot give a detailed account of how a more intelligent AI system would best them.
C: So, humans will not be bested by a more intelligent AI system.
> The argument starts with a hypothetical ("there is a possible artificial agent"), and it fails to be scary: there are (apparently) already humans that can kill 70% of humanity, and yet most of humanity is still alive. So an AGI that could also do it is not implicitly scarier.
Right, it's only an argument for the possibility of AGI catastrophe. It doesn't make any move to convince you that the scenario is likely. And it sounds like you already accept that the scenario is possible, so shrug.
> The final twitter thread is basically a thread of people saying "no, there is no canonical, well-formulated argument for AGI catastrophe", so I'm not sure why you shared it.
Maybe there is no canonical argument, but the thread definitely features arguments for likely AI catastrophe:
https://wiki.aiimpacts.org/doku.php?id=arguments_for_ai_risk:is_ai_an_existential_threat_to_humanity:will_malign_ai_agents_control_the_future:argument_for_ai_x-risk_from_competent_malign_agents:start
https://arxiv.org/abs/2206.13353
https://aiadventures.net/summaries/agi-ruin-list-of-lethalities.html1. States things like "Finding goals that are extinction-level bad and relatively useful appears to be easy: for example, advanced AI with the sole objective ‘increase company.com revenue’ might be highly valuable to company.com for a time, but risks longer term harms to society, if powerfully accruing resources and power toward this end with no regard for ethics beyond laws that are still too expensive to break." But even current-gen LLMs sidestep this pretty easily, and if you ask them to increase e.g. revenue, they do not propose extinction-level events or propose eschewing basic ethics. This argument falls apart upon contact with reality.
2. Is a 57-page PDF of subjectively-defined risks where it gives up on generalized paperclip-maximizing as a threat, but instead proposes narrower "power-seeking" as an unaligned threat that will lead to doom. It presents little evidence that language models will likely attempt to become power-seeking in the real world other than a (non-language-model) reinforcement learning experiment conducted by OpenAI in which an AI was trained to be good at a game that required controlling blocks, and the AI then attempted to control the blocks. It is possible I missed something in the 57 pages, but once it defines power-seeking as a supposed likely existential risk, it seemed to jump straight into proposals on attempted mitigations.
3. Requires accepting that we will by default build a misaligned superhuman AI that will cause humanity to go extinct as the basic premises of the argument (P1-P3), which makes the conclusions not particularly convincing if you don't already believe that.
Are you claiming that (A) nice behavior in current LLMs is good evidence that all future AI systems will behave nicely, or (B) nice behavior in current LLMs is good evidence that future LLMs will behave nicely?
> 3. Requires accepting that we will by default build a misaligned superhuman AI that will cause humanity to go extinct as the basic premises of the argument (P1-P3), which makes the conclusions not particularly convincing if you don't already believe that.
P3 from the argument says, "Superhuman AGI will be misaligned by default". I interpret that as meaning: if there isn't a highly resourced and focused effort to align superhuman AGI systems in advance of their creation, then the first systems we build will be misaligned.
Is that the some way you are interpreting it? If so, why do you believe it is probably false?
3. P1-P3 are non-obvious and overly speculative to me in many ways. P1 states that current research is likely to produce superhuman AI; I think that is controversial amongst researchers as it is: LLMs may not get us there. P2 states that "superhuman" AI will be uncontrollable — once again, I do not think that is obvious, and depends on your definition of superhuman. Does "superhuman" mean dramatically better at every mental task, e.g. a human compared to a slug? Does it mean "average at most tasks, but much better at a few?" Well, then it depends what few tasks it's better at. Similarly, it anthropomorphizes these systems and assumes they want to "escape" or not be controlled; it is not obvious that a superhumanly-intelligent system will "want" anything; Stockfish is superhuman at chess, but does not "want" to escape or do anything at all: it simply analyzes and predicts the best next chess move. The idea of "desire" on the part of the programs is a large unstated assumption that I think does not necessarily hold. Finally, P3 asserts that AI will be "misaligned by default" and that "misaligned" means that it will produce extinction or extinction-level results, which to me feels like a very large assumption. How much misalignment is required for extinction? Yud has previously made very off-base claims on this, e.g. believing that instruction-following would mean that an AI would kill your grandmother when tasked with getting a strawberry (if your grandmother had a strawberry), whereas current tech can already implicitly understand your various unstated goals in strawberry-fetching like "don't kill grandma." The idea that any degree of "misalignment" will be so destructive that it would cause extinction-level events is a) a stretch to me, and b) not supported by the evidence we have today. In fact a pretty simple thought experiment in the converse is: a superhumanly-intelligent system that is misaligned on many important values, but is aligned on creating AI that aligns with human values, might help produce more-intelligent and better-aligned systems that would filter out the misaligned goals — so even a fair degree of misalignment doesn't seem obviously extinction-creating. Furthermore, it is not obvious that we will produce misaligned AI by default. If we're training AI by giving it large corpuses of human text (or images, etc), and evaluating success by the model producing human-like output that matches the corpus, that... is already a form of an alignment process: how well does the model align to human thought and values in the training corpus? Anthropomorphizing an evil model that "wants" to exist and will thus "lie" to escape the training process but will secretly not produce aligned output at some hidden point in the future is... once again a stretch to me, especially because there isn't an obvious evolutionary process to get there: there has to already exist a superhuman, desire-ful AI that can outsmart researchers long before we are capable of creating superhuman AI, because otherwise the dumb-but-evil AI would give itself away during training and its weights wouldn't survive getting culled by poor model performance. P1-P3 are just so speculative and ungrounded in the reality we have today that it's very hard for me to take them seriously.
I agree with both of these claims (A) it is hard to find goals that are extinction-level bad for current SOTA LLMs, and (B) current SOTA LLMs understand at least some important context around the requests made to them.
But I'm also skeptical that they understand _all_ of the important context around requests made to them. Do you believe that they understand _all_ of the important context? If so, why?
> P2 states that "superhuman" AI will be uncontrollable — once again, I do not think that is obvious, and depends on your definition of superhuman. Does "superhuman" mean dramatically better at every mental task, e.g. a human compared to a slug? Does it mean "average at most tasks, but much better at a few?" Well, then it depends what few tasks it's better at.
I take "superhuman" to mean dramatically better than humans at every mental task.
> Similarly, it anthropomorphizes these systems and assumes they want to "escape" or not be controlled; it is not obvious that a superhumanly-intelligent system will "want" anything; Stockfish is superhuman at chess, but does not "want" to escape or do anything at all: it simply analyzes and predicts the best next chess move. The idea of "desire" on the part of the programs is a large unstated assumption that I think does not necessarily hold.
Would you have less of a problem with this premise if instead it talked about "Superhuman AI agents"? I agree that some systems seem more like oracles rather than agents, that is, they just answer questions rather than pursuing goals in the world.
Consider self-driving cars, regardless of whether or not self-driving cars 'really want' to avoid hitting pedestrians, they do in fact avoid hitting pedestrians. And then P2 is roughly asserting, regardless of whether or not a superhuman AI agent 'really wants' to escape control by humans, it will in fact not be controllable by humans.
> Finally, P3 asserts that AI will be "misaligned by default" and that "misaligned" means that it will produce extinction or extinction-level results, which to me feels like a very large assumption. How much misalignment is required for extinction? Yud has previously made very off-base claims on this, e.g. believing that instruction-following would mean that an AI would kill your grandmother when tasked with getting a strawberry (if your grandmother had a strawberry), whereas current tech can already implicitly understand your various unstated goals in strawberry-fetching like "don't kill grandma." The idea that any degree of "misalignment" will be so destructive that it would cause extinction-level events is a) a stretch to me, and b) not supported by the evidence we have today.
I'm often unsure whether you are making claims about all future AI systems or just future LLMs.
> In fact a pretty simple thought experiment in the converse is: a superhumanly-intelligent system that is misaligned on many important values, but is aligned on creating AI that aligns with human values, might help produce more-intelligent and better-aligned systems that would filter out the misaligned goals — so even a fair degree of misalignment doesn't seem obviously extinction-creating.
Maybe. Or the misaligned system will just disinterestedly and indirectly kill everyone by repurposing the Earth's surface into a giant lab and factory for making the aligned AI.
> Furthermore, it is not obvious that we will produce misaligned AI by default. If we're training AI by giving it large corpuses of human text (or images, etc), and evaluating success by the model producing human-like output that matches the corpus, that... is already a form of an alignment process: how well does the model align to human thought and values in the training corpus?
I believe it is likely that this process does some small amount of alignment work. But I would still expect the system to be mostly confused about what humans want.
Is this roughly the argument that you are making?
(P1) Current SOTA LLMs are good at understanding implicit context.
(P2) A system must be extremely misaligned in order to cause a catastrophe.
(C) So, it will be easy to sufficiently align future more powerful LLMs.(P1) Current SOTA AI is good at understanding implicit context, and improved versions will likely be better at understanding implicit context (much like gpt-4 is better at understanding context than gpt-3, and llama2 is better than llama1, and mixtral is better than gpt-3 and better than claude, etc).
(P2) Most misalignments within the observable behavior of current AI do not produce extinction-level goals, and given (P1), it is unclear why someone would believe it's likely going to in the future, since they'll be even better at understanding implicit human context of goals (e.g. implicit goals like do not make humanity extinct, don't turn the entire surface of the planet into an AI lab, etc).
(C) Future AI will not likely be extinction-level misaligned with human goals.
I think there are several other arguments, though, e.g.:
(P1) Progress on AI capabilities is evolutionary, with dumber models slowly being replaced by derivative-but-better models, in terms of architectural evolutionary improvements (e.g. new attention variants), dataset evolutionary improvements as they grow larger and as finetuning sets grow higher quality, and in terms of benchmark and alignment evolutionary progress.
(P2) Evolutionary steps towards evil-AI will likely be filtered out during training, since it will not yet be generalized superhuman intelligence and will give away its misalignment during training, whereas legitimately-aligned AI model evolutions will be rewarded for better performance.
(P3) Generalized superhuman intelligence will likely be an evolutionary step from a well-aligned ordinary intelligence, which will be an evolutionary step from sub-human intelligence that is reasonably well aligned.
(C) Superhuman intelligence will have been evolutionarily refined to be reasonably well-aligned.
Or:
(P1) LLMs have architectural issues that will prevent them from quickly becoming generalized superintelligence of the "human vs slug" variety (bad/inefficient at math, tokenization issues, likelihood of hallucinations, limited ability to learn new facts without expensive and slow training runs, difficulty backtracking from incorrect chains of reasoning, etc).
(C) LLM research is not likely to soon produce a superhuman AI able to cause an extinction event for humanity, and should not be illegal.
However, ultimately my most strongly-believed personal argument is:
(P1) The burden of proof for making something illegal due to apocalyptic predictions lies on the prognosticator.
(P2) There is not much hard evidence of an impending apocalypse due to LLMs, and philosophical arguments for it are either self-referential and require belief in the apocalypse as a prerequisite, or are highly speculative, or both.
(C) LLM research should not be illegal.
> (P1) Current SOTA AI is good at understanding implicit context, and improved versions will likely be better at understanding implicit context (much like gpt-4 is better at understanding context than gpt-3, and llama2 is better than llama1, and mixtral is better than gpt-3 and better than claude, etc).
I believe that (P1) is probably true.
> (P2) Most misalignments within the observable behavior of current AI do not produce extinction-level goals, and given (P1), it is unclear why someone would believe it's likely going to in the future, since they'll be even better at understanding implicit human context of goals (e.g. implicit goals like do not make humanity extinct, don't turn the entire surface of the planet into an AI lab, etc).
I'm confused about what exactly you mean by "goals" in (P2). Are you referring to (I) the loss function used by the algorithm that trained GPT4, or (II) goals and sub-goals which are internal parts of the GPT4 model, or (III) the sub-goals that GPT4 writes into a response when a user asks it "What is the best way to do X?"
But earlier you said this:
> 1. States things like "Finding goals that are extinction-level bad and relatively useful appears to be easy: for example, advanced AI with the sole objective ‘increase company.com revenue’ might be highly valuable to company.com for a time, but risks longer term harms to society, if powerfully accruing resources and power toward this end with no regard for ethics beyond laws that are still too expensive to break." But even current-gen LLMs sidestep this pretty easily, and if you ask them to increase e.g. revenue, they do not propose extinction-level events or propose eschewing basic ethics.
And in this quote it looks to me that you are using "goals" as in (III).
(I'm not an expert on these matters and I am admittedly still very confused about them. Minimally I'd like to make sure that we aren't talking past one another.)
What that was referencing was finding goals that a human would want an AI to follow, e.g. "increase revenue" was one example explicit goal in the wiki the human might want an AI to follow. The argument in the wiki was that the AI would then do unethical things in service of that goal that would be "extinction-level bad." My counter-argument is that current SOTA AI already understands that despite having an explicit goal — let's say given in a prompt — of "increase revenue," there are implicit goals of "do not kill everyone" (for example) that it doesn't need stated; as LLMs advance they have become better at understanding implicit human goals, and better at instruction-following with adherence to implicit goals; and thus future LLMs will be likely to be even better at doing that, and unlikely to e.g. resurface the planet and turn it into an AI lab when told to increase revenue or told to produce better-aligned AI.
I don't think anyone seriously believes this. It's very very clear to all humans that have ever played a game of any kind that they can be defeated in unexpected ways. I don't even think that anyone believes the claim "it's impossible for AGI to pose an existential risk to humanity".
The negation of the claim "AGI poses an existential risk to humanity" is "AGI doesn't necessarily pose an existential risk to humanity". This is what most people in the world believe, and it is the obvious "null theory" about any technology.
> https://wiki.aiimpacts.org/doku.php?id=arguments_for_ai_risk...
The argument here works just as much for single-minded humans, so it's quite moot.
> https://arxiv.org/abs/2206.13353
Too long, sorry. Maybe I will read it someday, but not today.
> https://aiadventures.net/summaries/agi-ruin-list-of-lethalit...
This seems to agree with my previously stated positions. It does try to establish a canonical argument, as you say, but then it goes on to explain why they don't think it's persuasive.
Okay. So we agree that (A) powerful systems can best weaker systems in ways that are unexpected to the weaker system, and (B) it is possible that AGI poses an existential risk to humanity.
> The negation of the claim "AGI poses an existential risk to humanity" is "AGI doesn't necessarily pose an existential risk to humanity".
It seems to me that the negation of your first claim is just "AGI doesn't pose an existential risk to humanity". Is "necessarily" doing some important work in your second claim?
>> https://wiki.aiimpacts.org/doku.php?id=arguments_for_ai_risk...
> The argument here works just as much for single-minded humans, so it's quite moot.
I don't understand why the argument being applicable to humans would make it moot. Please explain.
>> https://aiadventures.net/summaries/agi-ruin-list-of-lethalit...
> This seems to agree with my previously stated positions. It does try to establish a canonical argument, as you say, but then it goes on to explain why they don't think it's persuasive.
Is there a particular premise or inferential step in the blog's argument that you believe to be mistaken? (I've copied the argument below.)
P1: The current trajectory of AI research will lead to superhuman AGI.
P2: Superhuman AGI will be capable of escaping any human efforts to control it.
P3: Superhuman AGI will be misaligned by default, i.e. it will likely adopt values and/or set long-term goals that will lead to extinction-level outcomes, meaning outcomes that are as bad as human extinction.
P4: We do not know how to align superhuman AGI, i.e. reliably imbue it with values or define long-term goals that will ensure it does not ultimately lead to an extinction-level outcome, without some amount of trial & error (how nearly all of scientific research works).
C1: P2 + P3 In the case of superhuman AGI, since it will be able to escape human control and misaligned by default, the only survivable path to alignment cannot involve trial & error because the first failed try will result in an extinction-level outcome.
C2: P4 + C1 This means we will not survive superhuman AGI, because our survival would require alignment, towards which we have no survivable path: the only path we know of involves trial & error, which is not survivable.
C3: P1 + C2 Therefore the current trajectory of AI research which will produce superhuman AGI leads to an outcome where we do not survive.