David Knill and Alexandre Pouget published “The Bayesian brain: the role of uncertainty in neural coding and computation” in Trends in Neurosciences in December 2004 (Vol. 27, pages 712-719). The review crystallized the Bayesian-brain hypothesis as a research program.
The core claim is that the brain does not just process its best single estimate of the world; it represents and reasons with uncertainty. Sensory signals are noisy and ambiguous, so an effective perceptual system should weight each source of evidence by how reliable it is and combine sources the way Bayesian probability theory prescribes. Knill and Pouget surveyed behavioral experiments, such as how people fuse visual and touch cues, showing that human perception often matches the predictions of optimal Bayesian inference closely.
The review went further by asking how populations of neurons could actually carry probability distributions rather than point values, sketching the idea of probabilistic population codes. This connected an abstract computational claim to a concrete question about neural representation, which is what made the paper a durable reference.
For a general reader, the Bayesian-brain idea matters because it is the bridge between probability theory, the foundation of modern machine learning, and biological cognition: it suggests both brains and good AI systems succeed by tracking how much they should trust their own evidence.