In the authors’ preferred formalization of FDT, agents actually iterate over policies
(mappings from observations to actions) rather than actions. This makes a difference in
certain multi-agent dilemmas, but will not make a difference in this paper.
And it does seem that using FDT, but as a function that returns a policy rather than an action, solves this problem. So this is not an intrinsic problem with FDT that UDT doesn’t have, it’s a problem that arises in simpler versions of both theories and can be solved in both with the same modification.
I see. I suppose you’d do this by creating a policy node that is subjunctively upstream of every individual FDT decision, and intervening on that. The possible values would be every combination of FDT decisions, and you’d calculate updateless expected value over them.
This seems to work, though I’ll think on it some more. I’m a little disappointed that this isn’t the formulation of FDT in the paper, since that feels like a pretty critical distinction. But in any case, I should have read more carefully, so that’s on me. Thank you for bringing that up! Your comment is now linked in the introduction :)
In the FDT paper there is this footnote:
And it does seem that using FDT, but as a function that returns a policy rather than an action, solves this problem. So this is not an intrinsic problem with FDT that UDT doesn’t have, it’s a problem that arises in simpler versions of both theories and can be solved in both with the same modification.
I see. I suppose you’d do this by creating a policy node that is subjunctively upstream of every individual FDT decision, and intervening on that. The possible values would be every combination of FDT decisions, and you’d calculate updateless expected value over them.
This seems to work, though I’ll think on it some more. I’m a little disappointed that this isn’t the formulation of FDT in the paper, since that feels like a pretty critical distinction. But in any case, I should have read more carefully, so that’s on me. Thank you for bringing that up! Your comment is now linked in the introduction :)