Brain as active predictor theory
Explanation
Karl Friston, together with Andy Clark, Jakob Hohwy, Anil Seth and others, has developed a conception of the brain as a fundamentally predictive system. It is not an organ that passively receives sensations and then responds; it is a system that constantly generates predictions about what it will sense, and only adjusts its models when there is a discrepancy between prediction and input (prediction errors).
This theory, combining computational neuroscience, Bayesian inference and information theory, proposes that the brain implements a hierarchy of generative models. Each level predicts the representations of the level below; prediction errors are passed up to correct the models. The whole system adjusts itself by minimising variational free energy, a statistical measure of discrepancy between model and world.
An important distinction: we do not only predict, we also act on the world to make predictions come true. This is active inference. Bodily movements are not responses to stimuli, but actions the brain undertakes to confirm its predictions (moving the eyes towards where it expects to see something, reaching for an object it predicts to be there). Action and perception are two faces of the same predictive process.
This perspective transforms how we understand conscious perception. What we see is not what comes in through the eyes, but the brain's best hypothesis about the causes of sensory inputs. Anil Seth puts it this way: perception is controlled hallucination. Conscious reality is predictive cerebral construction, corrected moment by moment by sensory errors.
For consciousness, this view suggests that subjective experience is the most sophisticated global predictive model of the organism, integrating sensory, interoceptive, proprioceptive and cognitive information. Consciousness of the self is a prediction about one's own body and its states; consciousness of the world is a prediction about the external causes of sensory inputs.
This theory is one of the most unifying and productive in contemporary neuroscience. It connects perception, action, learning, attention, emotion, consciousness and pathology (psychosis as miscalibration of prediction errors, autism as hypersensitivity to errors, etc.). It is a common theoretical framework being explored and empirically validated with promising results.
Strengths
- Integrates perception, emotion and self in a single framework.
- Empirical support in bodily illusions and psychopharmacology.
- Clear and accessible articulation.
- Coherent with predictive processing.
Main critiques
- Does not directly solve the hard problem.
- Risk of identifying the self with prediction without explaining phenomenology.
- Some generalisations are empirically partial.
- Competes with other predictive theories.