by Alex Mennen 285 days ago | link | parent The behavior of players with finite strategies is not computable. To see this, let $$n$$-HaltingBot be the strategy that cooperates iff the $$n$$th Turing machine halts (on null input) in fewer steps than the length of the string of predictions. If the $$n$$th Turing machine halts, then the depth of this strategy is the number of steps it takes for it to halt. If it doesn’t halt, then the depth of this strategy is $$0$$. Either way, it’s a finite strategy. But you’re paying attention to the limiting behavior as the length of the string of predictions approaches infinity, and in that sense, the $$n$$-HaltingBot cooperates with the other player iff the $$n$$th program halts. However, I’m pretty sure the behavior of players with provable finite bounds on the depths of their strategies are computable. In terms of philosophical reasonableness, I’m kind of skeptical of this. In an actual implementation of this with real agents, presumably there isn’t some prediction oracle that will be assisting the agents, so the agents will have to generate these predictions themselves, and then they’ll actually have to compute the limits of their and the other player’s behavior in order to decide what to actually do. You also want to avoid getting exploited by players who act like ProbabilisticFairBot when the string of predictions is finite but defect when the string is infinite. So I think they’ll need to look for proofs that the other player is an agent that behaves according to this framework, with some bound on the depth of their strategy. Two players doing this then end up doing a Lobian handshake that they’ll both act according to this framework, but if they need to do a Lobian handshake, they may as well use it to cooperate, instead of using it to adopt a different framework that will lead them to cooperate.

NEW DISCUSSION POSTS

This is exactly the sort of
 by Stuart Armstrong on Being legible to other agents by committing to usi... | 0 likes

When considering an embedder
 by Jack Gallagher on Where does ADT Go Wrong? | 0 likes

The differences between this
 by Abram Demski on Policy Selection Solves Most Problems | 0 likes

Looking "at the very
 by Abram Demski on Policy Selection Solves Most Problems | 0 likes

 by Paul Christiano on Policy Selection Solves Most Problems | 1 like

>policy selection converges
 by Stuart Armstrong on Policy Selection Solves Most Problems | 0 likes

Indeed there is some kind of
 by Vadim Kosoy on Catastrophe Mitigation Using DRL | 0 likes

Very nice. I wonder whether
 by Vadim Kosoy on Hyperreal Brouwer | 0 likes

Freezing the reward seems
 by Vadim Kosoy on Resolving human inconsistency in a simple model | 0 likes

Unfortunately, it's not just
 by Vadim Kosoy on Catastrophe Mitigation Using DRL | 0 likes

>We can solve the problem in
 by Wei Dai on The Happy Dance Problem | 1 like

Maybe it's just my browser,
 by Gordon Worley III on Catastrophe Mitigation Using DRL | 2 likes

At present, I think the main
 by Abram Demski on Looking for Recommendations RE UDT vs. bounded com... | 0 likes

In the first round I'm
 by Paul Christiano on Funding opportunity for AI alignment research | 0 likes

Fine with it being shared
 by Paul Christiano on Funding opportunity for AI alignment research | 0 likes