Reward learning summary post by Stuart Armstrong 202 days ago | discuss A putative new idea for AI control; index here. I’ve been posting a lot on value/reward learning recently, and, as usual, the process of posting (and some feedback) means that those posts are partially superseded already - and some of them are overly complex. So here I’ll try and briefly summarise my current insights, with links to the other posts if appropriate (a link will cover all the points noted since the previous link): I’m modelling humans as a pair $$(p, R)$$, where $$R$$ is the reward function, and $$p$$ is a planning algorithm (called a planer) that maps rewards to policies. An agent is trying to learn $$R$$. The policy of a given human is designated $$\pi_H$$. A pair $$(p, R)$$ is compatible if $$p(R)=\pi_H$$. There is a no-free-lunch theorem for these $$(p, R)$$ pairs. Once the agent has learnt $$\pi_H$$, it can get no further evidence from observing the human. At that point, any compatible pair $$(p, R)$$ is a valid candidate for explaining the human planner/reward. Thus the agent cannot get any idea about the reward without making assumptions about the human planner (ie the human irrationality) - and can’t get any idea about the planner without making assumptions about the human reward. Unlike most no-free-lunch theorems, a simplicity prior does not remove the result. LINK It’s even worse than that: a simplicity prior can push us away from any “reasonable” $$(p, R)$$. LINK Ignoring “noise” doesn’t improve the situation: the real problem is bias. LINK The $$(p, R)$$ formalism can also model situations like the agent “overriding” the human’s reward. LINK There are Pascal’s mugging type risks in modelling reward override, where a very unlikely $$(p, R)$$ pair may still be chosen because the expected reward for that choice is huge. LINK There is also the risk of an agent transforming human into rational maximisers of the reward its computed so far LINK. Humans are not just creatures with policies, but we are creatures with opinions and narratives about our own values emotions, and rationality. Using these “normative assumptions”, we can start converging on better $$(p, R)$$ pairs. LINK LINK Even given that, our values will remain underdefined, changeable, and open to manipulation. LINK Resolving those problems with our values is a process much more akin to defining values that discovering them. LINK There are important part of human values that are not easily captured in the $$(p, R)$$ formalism. LINK

### NEW DISCUSSION POSTS

Note: I currently think that
 by Jessica Taylor on Predicting HCH using expert advice | 0 likes

Counterfactual mugging
 by Jessica Taylor on Doubts about Updatelessness | 0 likes

What do you mean by "in full
 by David Krueger on Doubts about Updatelessness | 0 likes

It seems relatively plausible
 by Paul Christiano on Maximally efficient agents will probably have an a... | 1 like

I think that in that case,
 by Alex Appel on Smoking Lesion Steelman | 1 like

 by Sam Eisenstat on No Constant Distribution Can be a Logical Inductor | 1 like

A: While that is a really
 by Alex Appel on Musings on Exploration | 0 likes

> The true reason to do
 by Jessica Taylor on Musings on Exploration | 0 likes

 by Vadim Kosoy on Musings on Exploration | 1 like

I'm not convinced exploration
 by Abram Demski on Musings on Exploration | 0 likes

Update: This isn't really an
 by Alex Appel on A Difficulty With Density-Zero Exploration | 0 likes

If you drop the
 by Alex Appel on Distributed Cooperation | 1 like

Cool! I'm happy to see this
 by Abram Demski on Distributed Cooperation | 0 likes

Caveat: The version of EDT
 by 258 on In memoryless Cartesian environments, every UDT po... | 2 likes

[Delegative Reinforcement
 by Vadim Kosoy on Stable Pointers to Value II: Environmental Goals | 1 like