normative
Intrinsic Reward for Information Drives Real Learning
A more promising approach than reward-maximizing reinforcement learning is an agent whose reward is intrinsic—gaining information or finding out more about how the world or system works rather than getting cheese, utility, or winning—so it will take actions that reveal how the world works even when they don't pay off in the short run, mirroring the scientist's intrinsic reward of figuring things out.
normativepending
Speaker
Alison GopnikEvidence Quote
“what you're calculating is, 'Did I get more information? Did I find out more about the world?'... you're gonna do the things that will tell you something about how the world works, even if they don't make you any better off in the short run”
Created: 6/13/2026, 12:24:35 AM
Your notes
Loading notes...