Year: 2026
Type: preprint
Abstract: Probabilistic reasoning in the brain has been studied primarily using simple two-choice tasks that are experimentally and theoretically tractable but have unclear relevance to more complex problems. Here we generalize principles from these studies using information geometry, which treats probability distributions (“beliefs”) as points on statistical manifolds whose geometry captures relationships between those beliefs. We introduce an “information compass,” a context-dependent, one-dimensional axis along which an arbitrarily complex belief is reinforced or repudiated. Projecting high-dimensional belief updates onto this axis casts even complex decisions into familiar forms of deliberation and commitment used for simpler decisions. Our framework provides a unified account of how biological and artificial systems can use geometric forms of probabilistic reasoning to make decisions across a range of tasks and domains.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: National Science Foundation
OpenAlex ID: W7222022001