I think Matt Newport’s comment answers it pretty well. The distinction I’m attempting to draw is between situation dependence and situation independence. If you assign 100 utils to a new car, then that car is worth 100 utils, period. You can still kinda get a utility function by saying I assign 100 utils to a new car now, but 50 utils to a new car next week. But when you assign different amounts to gaining the car vs. losing the car (loss aversion), or different amounts if you consider each facet of having the car separately than you get when you add all of them together (subadditive utility), or that you’d value winning the car in a contest a different amount than inheriting the car from a relative, then eventually it starts to look less like you’re calculating a term for “value of car” and using it in different decisions, and more like you’ve got a number of decision-making processes, all of which return different results for different decisions involving cars.
I admit it’s not a hard-and-fast distinction so much as a difference in connotations.
I am pretty sure I don’t approve of Matt Newport’s comment. If uncertainty about receiving something means that it is worth less then you would conventionally just build that into the utility function.
IMO, rewards are more-or-less the brain’s attempt to symbolically represent utility.
The description of how a reward function works sounds like how a utility function works.
What distinction are you attempting to draw between a reward function and a utility function?
I think Matt Newport’s comment answers it pretty well. The distinction I’m attempting to draw is between situation dependence and situation independence. If you assign 100 utils to a new car, then that car is worth 100 utils, period. You can still kinda get a utility function by saying I assign 100 utils to a new car now, but 50 utils to a new car next week. But when you assign different amounts to gaining the car vs. losing the car (loss aversion), or different amounts if you consider each facet of having the car separately than you get when you add all of them together (subadditive utility), or that you’d value winning the car in a contest a different amount than inheriting the car from a relative, then eventually it starts to look less like you’re calculating a term for “value of car” and using it in different decisions, and more like you’ve got a number of decision-making processes, all of which return different results for different decisions involving cars.
I admit it’s not a hard-and-fast distinction so much as a difference in connotations.
I am pretty sure I don’t approve of Matt Newport’s comment. If uncertainty about receiving something means that it is worth less then you would conventionally just build that into the utility function.
IMO, rewards are more-or-less the brain’s attempt to symbolically represent utility.