I’ve noticed a subtle assumption/hypothesis which I call “feedback realism”:
The way in which we provide feedback, directly imprints itself into the type signature of the motivations of the agent. If we give feedback in short episodes, the agent cares about things within short episodes. If we give feedback over outcomes, the agent cares about outcomes in particular.
I think there is some correlational truth to this, but that it’s a lot looser / more contingent / less clean than many people seem to believe.
I’ve noticed a subtle assumption/hypothesis which I call “feedback realism”:
I think there is some correlational truth to this, but that it’s a lot looser / more contingent / less clean than many people seem to believe.