Re Scott Garrabrant on arguments against independence: He said (personal communication) something like "utility function is a pretty innocent concept if you don't assume that it interacts linearly with probability to form expected utility and instead you allow a greater variety of methods to combine the two (in which case you have a violation of independence)". Also see Stuart Armstrong's post "Expected utility without the independence axiom"
Re decision-tree separability: this seems to be in tension with updatelessness and similar decision-theoretic shenanigans where you actually need to care what happens (would happen) in different branches.
Re money & free energy: not clear to me whether the universe we live in has (convergently?) a single scalar universal currency for everything that may matter to an agent
Re Scott Garrabrant on arguments against independence: He said (personal communication) something like "utility function is a pretty innocent concept if you don't assume that it interacts linearly with probability to form expected utility and instead you allow a greater variety of methods to combine the two (in which case you have a violation of independence)". Also see Stuart Armstrong's post "Expected utility without the independence axiom"
(https://www.lesswrong.com/posts/tGhz4aKyNzXjvnWhX/expected-utility-without-the-independence-axiom).
Re decision-tree separability: this seems to be in tension with updatelessness and similar decision-theoretic shenanigans where you actually need to care what happens (would happen) in different branches.
Re money & free energy: not clear to me whether the universe we live in has (convergently?) a single scalar universal currency for everything that may matter to an agent