15 – Natural Abstractions with John Wentworth

100122 F 4388M 075

The nap of the cloth can affect the run of the balls, especially on slower shots and shots played with sidespin applied to the cue ball. 5. Finally, parallel shift the line above to the centre of the cue ball. United Kingdom and in many former British colonies, is a cue sport that combines the aspects of carom billiards and pool. On a carom table, how to play billiards the K-55 profile is used (with a somewhat sharper angle than pool cushions). If you’re interested in other strategies like drills to improve your pool skills, you can check out our advice here. John Wentworth: I mean the right answer here is things that do not always unambiguously belong to one category and not another. Daniel Filan: Okay. So when I was talking about the easy path to alignment, where you just say what outcomes you do and don’t want, why does the easy path to alignment fail here?

John Wentworth: Right. So that gets us to one of the next problems, which is: so first of all, we don’t currently have a way that we know to robustly say don’t mess up the rest of the world, but I expect that part is relatively tractable. John Wentworth: In practice, there’s just tons of institutions that hold things long term and never trade. John Wentworth: I can’t give you a numerical example off the top of my head because it’s messy as all hell. John Wentworth: looking at Daniel’s notes Isn’t the type signature of human values world to reals. So the first thing is there’s this finding that neural networks are really good at classifying images, but you can also train them to optimality when you just make images by selecting random values for every pixel and giving them random classes. Can you describe a task where here’s this task that we would want an AI system to do, and we need to do really good value learning, or alignment, or something even beyond that.

So we want to measure impact in terms of those. So basically something like, we want to understand abstraction well enough to just take anything and be like, okay, what are the good abstractions here, and know that we know that. But it takes them a bit longer than it takes them to learn actual data sets of real images, which seems like it could be related to the idea that somehow it’s harder for them to find the unnatural abstractions. John Wentworth: I was also, around the same time, looking at markets through a similar lens, like financial markets. John Wentworth: Yes. Basically it tracks the sugar concentration over time. But if the sugar concentration is increasing over time, the number of phosphates will stay at a high level. John Wentworth: So you end up with the equilibrium number of phosphate molecules on there tracking what’s going on with the sugar outside. These seem like really good intuitive models for what’s going on, but we don’t know the right way to translate the low-level dynamics of the system into something that actually captures the semantics of agency. Intuitively, it seems like it’s doing agenty stuff, but we don’t have a good way to go from the low-level dynamical systems representation to it’s trying to do this, it’s modeling the world like that.

He’s interested in like, “Okay, what’s a good rough algorithm we could use? John Wentworth: So they’re using these very agency-loaded intuitions to understand what’s going on with the E. coli, but the actual models they have are just these simple dynamical systems, occasionally with some feedback loops in there. I’m currently mentoring an AI safety camp team, which is working on the modularity thing and it’s going extremely well. Daniel Filan: So I guess the first thing I’d like to ask is I see you as being interested in resolving confusions around agency, or things that we don’t understand about agency. If I could model this financial market as being an expected utility maximizer, that would make it a lot simpler to reason about in a lot of ways. How do I back out something that is semantically its world model? John Wentworth: So then, obvious question: what’s the utility function and the Bayesian model for a financial market? John Wentworth: So in principle, there could be some stocks held by entities that are outside of the market. John Wentworth: So, for instance, in a financial market, what could happen is … John Wentworth: So it can keep track of whether concentrations are increasing or decreasing as it’s slipping.

0