predictive coding

Predictive coding is a theory that says the brain is not a passive camera that simply records whatever the senses send it. Instead, it is a tireless guesser: at every moment it is already predicting what it is about to see, hear, and feel, and then it compares that guess against the signals actually arriving from the eyes, ears, and skin. What travels up to the higher parts of the brain is mostly not the raw sensation, but the mismatch — the part the brain got wrong. That leftover surprise is called the prediction error. Picture reading a sentence and your eyes skip lightly over familiar words but snag hard on a typo: the typo is the error, the thing your brain did not expect, and that is exactly what grabs your attention.

How does it work? Imagine the brain as a stack of layers. Higher layers send their best guesses downward — "based on everything I know, the next sound should be this" — and lower layers, closer to the senses, send back only the difference between that guess and reality. If the guess was right, almost nothing needs to be reported, so the brain stays quiet and saves energy. If the guess was wrong, an error signal shoots upward, and the higher layers quickly adjust their next prediction to fit. Round after round, the brain tries to shrink its errors toward zero — which is just another way of saying it is trying to understand the world well enough that nothing surprises it.

Why does this matter? It reframes perception as a kind of controlled hallucination that the senses keep correcting, and it offers one tidy explanation for many puzzles at once: why we can read messy handwriting, why we sometimes hear a word that was never spoken, why we tune out a steady hum but jump at a new noise, and why attention feels like turning up the volume on the errors that matter most. The same idea links to powerful frameworks in artificial intelligence, where machines also learn by predicting and then correcting their mistakes.