Filter, Detector, Predictor: The expanding repertoire of retinal computation in vertebrates

The primate retina has traditionally been viewed as a simple, high-resolution encoding stage, performing basic spatial, temporal, and chromatic filtering before transmitting information to cortical areas for higher-level processing. This perspective, formalized in the classical three-channel model, emphasizes three dominant ganglion cell pathways: parasol cells for achromatic signals, midget cells for red-green opponency and spatial detail, and small bistratified cells for blue-yellow signals. However, recent evidence challenges this view, showing that these pathways play more complex and nuanced roles than previously assumed. In non-primate species, the retina is now recognized as a site of substantial computation, performing nonlinear processing, motion analysis, and feature extraction. Emerging data suggest that similar computational complexity exists in the primate retina but has been underappreciated due to sampling biases and the use of overly simplistic stimuli. Anatomical and physiological studies indicate that the primate retina contains at least 20-30 distinct ganglion cell types, many exhibiting feature selectivity analogous to that seen in rodents, rabbits, and amphibians. In this review, we critically examine the classical model of primate retinal processing, its historical foundations, and its limitations. We highlight recent evidence revealing a more complex and diverse primate retinal output and discuss how conserved and primate-specific circuits contribute to early visual computations. Together, these findings reposition the primate retina as an active computational structure, not merely a simple front-end encoder.