Neocortical computation has been framed as predictive coding, where cortex continually generates hypotheses and signals mismatches as prediction errors. Theoretical development by Karl Friston University College London and earlier formulations by Rajesh P. N. Rao University of Washington provide the EEAT foundation linking algorithmic ideas to identifiable circuit elements.
Laminar motifs: who signals prediction and error
Anatomical and physiological evidence maps the algorithm onto laminar cortical architecture. Superficial pyramidal cells in layers 2 and 3 are proposed to encode bottom-up prediction errors, relaying unexpected sensory information to higher areas. Deep-layer pyramidal cells in layers 5 and 6 convey top-down predictions that target the apical dendrites of superficial cells, implementing subtractive comparison. This laminar segregation is supported by laminar recording studies and circuit-tracing work that show distinct feedforward and feedback projection patterns, consistent with the models advocated by Karl Friston University College London and subsequent circuit-focused interpretations.
Inhibitory interneurons, dendrites, and subcortical loops
Local inhibitory motifs refine error versus prediction signaling. Parvalbumin-expressing interneurons control perisomatic gain and timing, while somatostatin interneurons gate distal dendritic inputs, modulating the impact of feedback on apical dendrites. This interneuron specialization allows cortex to perform approximate arithmetic operations required for subtraction and precision-weighting of errors. Thalamocortical loops further shape which signals are treated as predictions versus errors: specific thalamic nuclei relay feedforward sensory evidence, and higher-order thalamic pathways mediate contextual feedback, enabling alignment across cortical columns.
Relevance, causes, and consequences emerge across scales. Functionally, these motifs explain robust perceptual inference: rapid suppression of expected sensory input frees resources to detect salient deviations. Clinically, disruptions to feedback pathways or inhibitory balance can produce abnormal error signaling implicated in schizophrenia and autism, linking microcircuit dysfunction to perceptual and cognitive symptoms. Culturally and environmentally, predictive mechanisms underlie how humans adapt expectations to local norms and changing ecosystems, influencing learning and decision-making in real-world settings.
Empirical work continues to refine details: laminar-resolved imaging, optogenetic perturbations, and computational modeling are converging to test which neurons carry signed errors, how precision is encoded, and how predictive hierarchies are instantiated across areas. Contributions from theoretical neuroscientists and experimental groups aim to translate abstract models into testable circuit-level predictions, strengthening both mechanistic understanding and translational impact.