One of visual neuroscience's oldest puzzles—why spinning black-and-white patterns produce vivid color sensations—may finally have a mechanistic explanation, and the answer has implications for understanding how the brain anticipates rather than simply records sensory input. For health-conscious adults interested in brain aging and neurological resilience, predictive coding frameworks are increasingly central to theories of how healthy cognition is maintained and how conditions like Alzheimer's or schizophrenia may represent failures of the brain's forecasting machinery.
Researchers trained artificial neural networks on natural videos using predictive learning—a framework where the network learns to anticipate the next frame rather than simply classify static images. When these models were then tested with achromatic (black-and-white) rotating stimuli like Benham's top, they generated artificial subjective color percepts despite encountering no actual chromatic information. Critically, the specific colors hallucinated by the network were systematically shaped by the colors of moving objects present in the training videos, not by retinal photoreceptor fatigue as older theories proposed. Experiments across progressively simplified stimuli, from natural scenes to 2D animations, confirmed that motion-color statistical associations—not low-level retinal dynamics—primarily drive the phenomenon.
This finding positions cortical predictive mechanisms as co-generators of perceptual experience, not merely passive recipients of retinal signals. The classical explanation for Benham's top relied on differential fatigue of cone photoreceptors responding to flickering temporal patterns. This new model suggests the brain actively imposes learned chromatic expectations onto ambiguous motion signals, a fundamentally different locus of effect. The implication for broader neuroscience is significant: if the cortex routinely synthesizes color from motion-based priors, predictive coding failures—already implicated in migraine aura, visual snow syndrome, and hallucinations in psychotic disorders—may share a common computational thread. The limitation here is that the model is computational rather than directly neurophysiological; human imaging or electrophysiology would be needed to confirm analogous cortical dynamics. Still, this represents a potentially paradigm-shifting reframe of a 200-year-old observation.
