Stuart Russell: The Foundations of Artificial Intelligence
The Gradient: Perspectives on AI
00:00
Is Meta Reasoning the Solution to Bounded Optimism?
The same algorithm can play backgammon, can do single agent tasks like finding roots through a complicated obstacle course and all sorts of stuff. And so this is a good candidate that you have met a reasoning capability that works no matter what the nature of the underlying decision problem is. The way to think about it is that it's one possible agent architecture. Within that class of Asian architectures, you could try to optimize the amount of meta reasoning that happens or how it's implemented. But there's no optimal solution because there's no such thing as "optimal Meta Reasoning"
Transcript
Play full episode
