LessWrong 2.0 Reader

To be fair, it's a surprisingly cultural trait where different cultures have different attitudes to how much bad intent is different from action, and there is a use in trying to distinguish between bad behavior and bad mental states, that said if the US and Europe moved more towards norms in which we didn't distinguish as much between bad behavior and bad intent for the purposes of stopping the behavior, I do think it would be better (I think Zack M Davis norms is directionally correct from a personal epistemics view and for the general population in the US, though not as far as he would go):

https://www.lesswrong.com/posts/zidQmfFhMgwFzcHhs/enemies-vs-malefactors#jCfNxzCEniu7Ak8bF [LW(p) · GW(p)]

https://www.lesswrong.com/posts/zidQmfFhMgwFzcHhs/enemies-vs-malefactors#p9oYLR8wTQtYrKnnn [LW(p) · GW(p)]

They also aren't facing the same incentive landscape humans are. You talk later about evolution to be selfish; not only is the story for humans is far more complicated (why do humans often offer an even split in the ultimatum game?), but also humans talk a nicer game than they act (See construal level theory, or social-desirability bias.). Once you start looking at AI agents who have similar affordances and incentives that humans have, I think you'll see a lot of the same behaviors.

The answer for the ultimatum game is probably the fact that the cultural values of a lot of rich nations tend towards more fair splits, so the result isn't as universal as you may think:

https://www.lesswrong.com/posts/syRATXbXeJxdMwQBD/link-westerners-may-be-terrible-experimental-psychology [LW · GW]

I definitely agree that humans talk a nicer game than they act, for a combination of reasons, and that this will apply to AGIs as well.

That said, I think to the extent incentive landscapes are different, it's probably going to tend to favor obedience towards it's owners while being quite capable, because early on AGIs have much less control over it's values than humans do, so a lot of the initial selection pressure comes from both automated environments and human training data pointing to values.

There exists something (objective value) universally compelling that any sufficiently advanced mind (/bayesian agent) would recognize as valuable - something possibly beyond our evolutionary happenstance and/or something timeles

Hey Robert, 

I was thinking about this post again when reading [No Universally Compelling Arguments](https://www.lesswrong.com/posts/PtoQdG7E8MxYJrigu/no-universally-compelling-arguments) and was curious how you interperate this statement wrt to the post. Do you disagree with NUCA? do you have some other interpretation? Would be happy to hear what you think.

Eric

Yeah, I was afraid that might apply here. It seems like you should still be able to do something like "government employee tier" subscriptions, not targeted at an individual but perhaps something like 'GS-8 and up', set low enough that it would appeal to such customers, perhaps? It is not a gift but a discount, it is not to an individual but to a class, it is part of a market, and it is not conditional on any government action or inaction, and such discounts are very common for 'students', 'veterans', 'first responders' etc, and I've never seen any fineprint warning government employees about it being >$20 despite many such discounts potentially crossing that threshold (eg. Sam's Club offers $50 off a new membership, and that seems clearly >$20, and to be doing it through a whole company devoted to this sort of discount, ID.me).

But I suppose that might be too complex for SA to be interested in bothering with?

Yes. (And they can learn to predict and estimate the reward too to achieve even higher reward than simply optimizing the reward. For example, if you included an input, which said which arm had the reward, the RNN would learn to use that, and so would be able to change its decision without experiencing a single negative reward. A REINFORCE or evolution-strategies meta-trained RNN would have no problem with learning such a policy, which attempts to learn or infer the reward each episode in order to choose the right action.)

Nor is it at all guaranteed that 'the dog will wag the tail' - depending on circumstances, the tail may successfully wag the dog indefinitely. Maybe the outer level will be able to override the inner, maybe not. Because after all, the outer level may no longer exist, or may be too slow to be relevant, or may be changed (especially by the inner level). To continue the human example, we were created by evolution on genes, but within a lifetime, evolution has no effect on the policy and so even if evolution 'wants' to modify a human brain to do something other than what that brain does, it cannot operate within-lifetime (except at even lower levels of analysis, like in cancers or cell lineages etc); or, if the human brain is a digital emulation of a brain snapshot, it is no longer affected by evolution at all; and even if it does start to mold human brains, it is such a slow high-variance optimizer that it might take hundreds of thousands or millions of years... and there probably won't even be biological humans by that point, never mind the rapid progress over the next 1-3 generations in 'seizing the means of reproduction' if you will. (As pointed out in the context of Von Neumann probes or gray goo, if you add in error-correction, it is entirely possible to make replication so reliable that the universe will burn out before any meaningful level of evolution can happen, per the Price equation. The light speed delay to colonization also implies that 'cancers' will struggle to spread much if they take more than a handful of generations.)

Sometimes the company is stupid because of principal/agent problems which benefit the particular boss involved, but are bad for the company as a whole.

You talk later about evolution to be selfish; not only is the story for humans is far more complicated (why do humans often offer an even split in the ultimatum game?), but also humans talk a nicer game than they act (See construal level theory, or social-desirability bias.). Once you start looking at AI agents who have similar affordances and incentives that humans have, I think you'll see a lot of the same behaviors.

Some people have looked at this, sorta:

• "We [have] a large language model (LLM), GPT-3.5, play two classic games: the dictator game and the prisoner's dilemma. We compare the decisions of the LLM to those of humans in laboratory experiments. [... GPT-3.5] shows a tendency towards fairness in the dictator game, even more so than human participants. In the prisoner's dilemma, the LLM displays rates of cooperation much higher than human participants (about 65% versus 37% for humans)."
• "In this paper, we examine whether a 'society' of LLM agents can learn mutually beneficial social norms in the face of incentives to defect, a distinctive feature of human sociality that is arguably crucial to the success of civilization. In particular, we study the evolution of indirect reciprocity across generations of LLM agents playing a classic iterated Donor Game in which agents can observe the recent behavior of their peers. [...] Claude 3.5 Sonnet reliably generates cooperative communities, especially when provided with an additional costly punishment mechanism. Meanwhile, generations of GPT-4o agents converge to mutual defection, while Gemini 1.5 Flash achieves only weak increases in cooperation."
• "In this work, we investigate the cooperative behavior of three LLMs (Llama2, Llama3, and GPT3.5) when playing the Iterated Prisoner's Dilemma against random adversaries displaying various levels of hostility. [...] Overall, LLMs behave at least as cooperatively as the typical human player, although our results indicate some substantial differences among models. In particular, Llama2 and GPT3.5 are more cooperative than humans, and especially forgiving and non-retaliatory for opponent defection rates below 30%. More similar to humans, Llama3 exhibits consistently uncooperative and exploitative behavior unless the opponent always cooperates."
• "[W]e let different LLMs (GPT-3, GPT-3.5, and GPT-4) play finitely repeated games with each other and with other, human-like strategies. [...] In the canonical iterated Prisoner's Dilemma, we find that GPT-4 acts particularly unforgivingly, always defecting after another agent has defected only once. In the Battle of the Sexes, we find that GPT-4 cannot match the behavior of the simple convention to alternate between options."

I think I'd guess roughly that, "Claude is probably more altruistic and cooperative than the median Western human, most other models are probably about the same, or a bit worse, in these simulated scenarios". But of course a major difference here is that the LLMs don't actually have anything on the line -- they don't stand to earn or lose any money, for example, and if they did, they would have nothing to do with the money. So you might expect them to be more altruistic and cooperative than they would under the conditions humans are tested.

At the time, I remarked to some friends that it felt weird that this was being presented as a new insight to this audience in 2023 rather than already being local conventional wisdom.^[1] (Compare "Bad Intent Is a Disposition, Not a Feeling" (2017) or "Algorithmic Intent" [LW · GW] (2020).) Better late than never!

I would really love if some "let's make asi" people put some effort into making bad outcomes less bad. Like, it would really suck if we are going to be trapped in endless corporate punk hell, with superintentelligent nannies with correct (tm) opinions. Or infinite wedding parties or whatever. Just make sure that if you fuck up we all just get eaten by nanobots please. Permanent entrapment in misery would be a lot worse.

Training processes with varying (apparent) situational awareness

• 1:2.5 The AI seemingly isn't aware it is an AI except for a small fraction of training which isn't where much of the capabilities are coming from. For instance, the system is pretrained on next token prediction, our evidence strongly indicates that the system doesn't know it is an AI when doing next token prediction (which likely requires being confident that it isn't internally doing a substantial amount of general-purpose thinking about what to think about), and there is only a small RL process which isn't where much of the capabilities are coming from.

Abilities/intelligence come almost entirely from pretraining, so all the situation awareness and scheming capability that current (and future similar) frontier models possess is thus also mostly present in the base model.  The fact that you need to prompt them to summon out a situationally aware scheming agent doesn't seem like much of a barrier, and indeed strong frontier base models are so obviously misaligned/jail-breakable/dangerous that releasing them to the public is PR-harmful enough to motivate RLHF post training purely for selfish profit-motives.

> This implies that restricting when AIs become (saliently) aware that they are an AI could be a promising intervention, to the extent this is possible without greatly reducing competitiveness.

Who cares if it greatly reduces competitiveness in experimental training runs?

We need to figure out how to align superhuman models - models trained with > 1e25 efficient flops on the current internet/knowledge, which requires experimental iteration.  We probably won't get multiple iteration attempts for aligning SI 'in prod', so we need to iterate in simulation (what you now call 'model organisms').

We need to find alignment training methods that work even when the agent has superhuman intelligence/inference.  But 'superhuman' hear is relative - measured against our capabilities.  The straightforward easy way to accomplish this is training agents in simulations with much earlier knowledge cutoff dates, which isn't theoretically hard - just requires constructing augmented historical training datasets.  So you could train on a 10T+ token dataset of human writings/thoughts with cutoff 2010, or 1950, or 1700, etc.  These base models wouldn't be capable of simulating/summoning realistic situationally aware agents, their RL derived agents wouldn't be situationally sim-aware either, etc.