Nice, I'd read the first but didn't realise there were more. I'll digest later.
I think agents vs optimisation is definitely reality-carving, but not sure I see the point about utility functions and preference orderings. I assume the idea is that an optimisation process just moves the world towards states, but an agent tries to move the world towards certain states i.e. chooses actions based on how much they move the world towards certain states, so it make sense to quantify how much of a weighting each state gets in its decision-making. But it's not obvious to me that there's not a meaningful way to assign weightings to states for an optimisation process too - for example if a ball rolling down a hill gets stuck in the large hole twice as often as it gets stuck in the medium hole and ten times as often as the small hole, maybe it makes sense to quantify this with something like a utility function. Although defining a utility function based on the typical behaviour of the system and then trying to measure its optimisation power against it gets a bit circular.
Anyway, the dynamical systems approach seems good. Have you stopped working on it?
Nice, thanks. It seems like the distinction the authors make between 'building agents from the ground up' and 'understanding their behaviour and predicting roughly what they will do' maps to the distinction I'm making, but I'm not convinced by the claim that the second one is a much stronger version of the first.
The argument in the paper is that the first requires an understanding of just one agent, while the second requires an understanding of all agents. But it seems like they require different kinds of understanding, especially if the agent being built is meant to be some theoretical ideal of rationality. Building a perfect chess algorithm is just a different task to summarising the way an arbitrary algorithm plays chess (which you could attempt without even knowing the rules).
I've been thinking about whether these results could be interpeted pretty differently under different branding.
The current framing, if I understand it correctly, is something like, 'Powerseeking is not desirable. We can prove that keeping your options open tends to be optimal and tends to meet a plausible definition of powerseeking. Therefore we should expect RL agents to seek power, which is bad.'
An alternative framing would be, 'Making irreversible changes is not desirable. We can prove that keeping your options open tends to be optimal. Therefore we should not expect RL agents to make irreversible changes, which is good.'
I don't think that the second framing is better than the first, but I do think that if you had run with it instead then lots of people would be nodding their heads and feeling reassured about corrigibility, instead of feeling like their views about instrumental convergence had been confirmed. That makes me feel like we shouldn't update our views too much based on formal results that leave so much room for interpretation. If I showed a bunch of theorems about MDPs, with no exposition, to two people with different opinions about alignment, I expect they might come to pretty different conclusions about what they meant.
What do you think?
(To be clear I think this is a great post and paper, I just worry that there are pitfalls when it comes to interpretation.)
If you know which variables you want to remove the incentive to control, an alternative to penalising divergence is path-specific objectives, i.e. you compute the score function under an intervention on the model that sets the irrelevant variables to their status quo values. Then the AI has no incentive to control the variables, but no incentive to keep them the same either.