Laurent, I agree that the Solomonoff induction should do a good job predicting the next bit (especially if the programming language is symmetric with respect to 1s and 0s).
However, if I'm not mistaken, AIXI relies on SI being able to model all possible futures optimally, rather than just the next bit. Is this wrong? (I haven't yet come up with a full reinforcement learning counterexample that would demonstrate this presumed failure though. Perhaps it could be a game against an intelligent opponent with the 2x2 identity matrix as the payoff matrix) ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
