An In-Silico Audit of Topographic Organization in a Multimodal Brain-Response Model
An audit of TRIBE v2, a multimodal model predicting fMRI responses across the cortical surface from video, audio and text, tests whether it reproduces classic topographic maps of sensory cortex. The model shows no contralateral hemifield response or polar-angle map, no tonotopic gradient in primary auditory cortex, and an eccentricity-like gradient that appears to reflect global image statistics rather than visual-field position. Motion selectivity is the exception: 22% of the 1% most motion-selective vertices fall in area MT+, which covers only 0.42% of cortex, indicating feature-identity experiments are viable while position-based mapping is not.
It delivers concrete, quantified evidence about which in-silico neuroscience experiments a leading brain-response model can and cannot support, a useful capability boundary for readers tracking AI models of the brain.
Comments & Questions (0)
No comments yet
Be the first to comment!