Perception is modelled as the brain generating predictions and processing only the error between prediction and input, rather than building percepts from the senses upward.
On this account the sensory stream mostly confirms what was already predicted, and what travels up the hierarchy is the discrepancy. It explains sensory attenuation, several illusions, and the cost of surprise economically. The unease among critics is that a framework which explains everything from perception to action to psychiatry may be doing so by being unfalsifiable.
In practice
- A word misheard in noise is heard as the word that was expected, and the expectation arrived first.
- A familiar route feels shorter because almost nothing along it needs processing.
- Possible explanations
- The explanation is that the brain maintains a generative model and passes upward only the discrepancy between prediction and input, which is cheaper than transmitting everything and explains why the expected is barely noticed.
- In its favour
- It accounts for a wide range of findings in one framework, including sensory attenuation, several illusions and aspects of attention, and it has generated testable predictions in perception research.
- Against it
- Its scope is the main criticism, since a framework that accommodates almost any result is hard to refute, and critics question whether its central formalism is doing explanatory work or being fitted after the fact.
- Standing
- An ambitious unifying framework whose scope is its main criticism: it accommodates a very wide range of findings, and critics question what observation could count against it.
- Subject
- Science & Epistemology
- Essay
- How knowledge grows, and how inquiry fools itself