Your Brain Doesn't Command Your Body. It Predicts It. [Max Bennett]

Brain generates perception from priors, not raw sensory input

Following Hermann von Helmholtz, conscious perception is an active inference about what is likely true in the world, where sensory data serves only as evidence to update prior beliefs rather than directly constructing reality.

Maintaining models despite sensory gaps

The brain sustains its mental model of objects even when sensory input disappears, updating only when contradictory evidence forces a revision, which enables efficient navigation through noisy or dark environments.

Unity of perception prevents simultaneous alternatives

The brain cannot render two simulations at once, explaining bistable illusions like the duck/rabbit where it constructs one coherent 3D hypothesis at a time because no hybrid model exists.

Neocortex enables learning by imagining

The neocortex enables model-based reinforcement learning by building rich world models that can be explored offline through mental simulation without requiring immediate sensory input.

Partnership with subcortical structures

Mental simulation requires collaboration between the neocortex and older structures like the thalamus and basal ganglia, which handle the pausing, intention modeling, and search space pruning necessary for planning.

Integration with Hawkins' thousand brains theory

The neocortex contains redundant overlapping models that must be synergized into a single rendered simulation, aligning with the constraint that we perceive only one coherent interpretation at a time.

Data scarcity in comparative psychology

Our understanding of brain evolution relies on sparse data, such as the complete absence of navigation studies in lamprey fish, forcing scientists to infer ancestral capabilities from distant relatives like teleost fish.

The AI-neuroscience implementation gap

While Carl Friston's active inference provides rich explanatory frameworks for biological intelligence, its principles remain largely unimplemented in practical AI systems like transformers, suggesting either a divergence between artificial and biological intelligence or untapped breakthroughs.

Outsider synthesis across fields

Bennett's technology background provided a cognitive bias toward 'ordered modifications' that enabled productive cross-pollination between comparative psychology, evolutionary neuroscience, and AI engineering.