Brain Inspired BI 219 Xaq Pitkow: Principles and Constraints of Cognition
4 snips
Aug 27, 2025 Xaq Pitkow, director of the Lab for the Algorithmic Brain at Carnegie Mellon University, dives into the principles of cognition and their applications. He shares insights on 'inverse rational control,' illustrating how organisms tackle tasks using strategies that often diverge from optimal solutions. The conversation covers the role of probabilistic graph networks in brain computations and introduces a new ecological neuroscience project with collaborative efforts. Xaq emphasizes the intersection of neuroscience and AI, advocating for a diverse range of models to understand cognitive processes.
AI Snips
Chapters
Transcript
Episode notes
Principles Guide Neuroscience Models
- Xaq Pitkow frames his work as finding normative principles under realistic biological constraints.
- He treats physics and evolutionary legacy as distinct sources of constraints shaping brain computation.
Rationality Relative To The Animal's Beliefs
- Inverse rational control assumes animals act optimally under their own (possibly wrong) beliefs.
- This reframing explains apparently suboptimal behavior as rational relative to the animal's internal model.
Scale Enables New Neuroscience Questions
- Increasing task complexity and richer data lets neuroscience explore more naturalistic cognition.
- Large descriptive models (foundation models) can summarize data but don't by themselves reveal algorithmic mechanisms.