One Doubt About Timeless Decision Theories

Time­less De­ci­sions The­o­ries (in­clud­ing var­i­ants like FDT, UDT, ADT, ect.) provide a rather el­e­gant method of solv­ing a broader class of prob­lems than CDT. While CDT re­quires the out­comes of de­ci­sions to be in­de­pen­dent of the in­di­vi­d­ual mak­ing the de­ci­sion (in such a way that causal surgery on a sin­gle node is valid), time­less de­ci­sions the­o­ries can han­dle any prob­lem where the out­come is a func­tion of the choice se­lected (even if this oc­curs in­di­rectly as a re­sult of a pre­dic­tion).

(Epistemic Sta­tus: Thoughts for fur­ther in­ves­ti­ga­tion)

This is an ex­cel­lent rea­son to in­ves­ti­gate these de­ci­sion the­o­ries, yet we need to make sure that we don’t get blinded by in­sight. Be­fore we im­me­di­ately jump to con­clu­sions by tak­ing this im­prove­ment, it is worth­while con­sid­er­ing what we give up. Per­haps there are other classes that we might which to op­ti­mise over which we can no longer op­ti­mise over once we have in­cluded this whole class?

After all, there is a sense in which there is no free lunch. As dis­cussed in the TDT pa­per, for any al­gorithm, we could cre­ate a situ­a­tion where there is an agent that speci­fi­cally pun­ished that al­gorithm. The usual re­sponse is that these situ­a­tions are un­fair, but a) the uni­verse is of­ten un­fair b) there are plau­si­ble situ­a­tions where the al­gorithm cho­sen in­fluences the out­come is slightly less un­fair ways.

Ex­pand­ing on b), there are times when you want to be pre­dictable to simu­la­tors. In­deed, I can even imag­ine agents that wish to elimi­nate agents that they can’t pre­dict. Fur­ther, rather than fac­ing a perfect pre­dic­tor, it seems like it’ll be at least a few or­ders of mag­ni­tude more likely that you’ll face an im­perfect pre­dic­tor. Model­ling these as X% perfect pre­dic­tor, 100-X% ran­dom pre­dic­tor will usu­ally be im­plau­si­ble as pre­dic­tors won’t have a uniform suc­cess rate over all al­gorithms. Th­ese situ­a­tions are slightly more plau­si­ble for sce­nar­ios in­volv­ing AI, but even if you perfectly know an agent’s source code, you are un­likely to know its ex­act ob­ser­va­tional state due to ran­dom noise.

It there­fore seems that the “best” de­ci­sion the­ory al­gorithm might be dom­i­nated by fac­tors other than op­ti­mal perfor­mance on the nar­row class of prob­lems TDT op­er­ates on. It may very well be the case that TDT is ul­ti­mately tak­ing the right ap­proach, but even if this is the case, I thought it was worth­while sketch­ing out these con­cerns so that they can be ad­dressed.