Why Does the Future Branch? Identifiable Closure Tests for Stochastic Physical World Models
The episode discusses a paper titled "Why Does the Future Branch?" which investigates why a world model predicts uncertainty. The authors show that observations alone cannot distinguish between missing state information and inherent process randomness. They propose an intervention protocol called ClosurePairs, providing a way to determine if uncertainty comes from sensor limitations or inherent noise, guiding where computational resources should be allocated.