Bayesian Brains without Probabilities.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Bayesian Brain Hypothesis

Background:

• Bayesian models are influential in explaining cognitive functions like perception, learning, and language.
• Despite theoretical Bayesian frameworks, human performance on probability tasks is often poor.
• This discrepancy raises questions about how the brain processes probabilistic information.

Purpose of the Study:

• To resolve the paradox between Bayesian cognitive models and human probabilistic reasoning deficits.
• To propose an alternative model of brain function that accounts for observed reasoning errors.

Main Methods:

• Theoretical analysis of Bayesian inference and sampling processes.
• Examination of cognitive biases (e.g., unpacking effect, base-rate neglect, conjunction fallacy) within a sampling framework.

Main Results:

• The brain may function as a Bayesian sampler, not a direct probability calculator.
• Finite sampling inherently leads to systematic probabilistic reasoning errors.
• This sampling model explains common biases observed in human judgment.

Conclusions:

• The Bayesian brain hypothesis may be better understood through a sampling mechanism.
• This perspective reconciles the success of Bayesian models with empirical findings of human irrationality.
• Finite sampling provides a unified explanation for various cognitive biases in probabilistic reasoning.