Particle.news

AI Trained on One Person’s Brain Activity Identifies Colors Seen by Another

The experiment aligned retinotopic maps across people to reveal conserved color signatures.

Overview

  • Researchers at the University of Tübingen report in JNeurosci that group-trained models decoded which of three hues and two brightness levels a left-out participant viewed.
  • Fifteen adults underwent fMRI while seeing red, green or yellow stimuli, and a linear classifier using a leave-one-out design predicted both color and luminance from brain activity.
  • The team built a common response space from retinotopic mapping, allowing comparisons across individuals without relying on each person’s anatomy.
  • Visual areas showed consistent, region-specific biases across the visual field, with central locations tending toward yellow preferences and the periphery toward red.
  • The authors stress that shared neural patterns do not prove identical subjective experiences and note limits including small sample size, narrow stimulus set and fMRI’s coarse resolution.