The problem is that I don’t see how to integrate this approach for solving this problem with deep learning. It seems like this approach might work well for a model-based RL setup where you can make the AI explicitly select for this utility function.
For my part, I was already expecting AGI to be some kind of model-based RL. So I’m happy to make that assumption.
However, when I tried to flesh out model splintering (a.k.a. concept extrapolation) assuming a model-based-RL AGI—see Section 14.4 here—I still couldn’t quite get the whole story to hang together.
(Before publishing that, I sent a draft to Stuart Armstrong, and he told me that he had a great answer but couldn’t make it public yet :-P )
However, when I tried to flesh out model splintering (a.k.a. concept extrapolation) assuming a model-based-RL AGI—see Section 14.4 here—I still couldn’t quite get the whole story to hang together.
Thanks for linking that!
(Before publishing that, I sent a draft to Stuart Armstrong, and he told me that he had a great answer but couldn’t make it public yet :-P )
For my part, I was already expecting AGI to be some kind of model-based RL. So I’m happy to make that assumption.
However, when I tried to flesh out model splintering (a.k.a. concept extrapolation) assuming a model-based-RL AGI—see Section 14.4 here—I still couldn’t quite get the whole story to hang together.
(Before publishing that, I sent a draft to Stuart Armstrong, and he told me that he had a great answer but couldn’t make it public yet :-P )
Thanks for linking that!
Oooh that is really exciting news.