What does a single neuron in your brain 'see'? A new study from Stanford and UT Austin shows that AI can now answer that question in plain English.
The Research
In a paper published on arXiv in May 2026, scientists led by Vedang Lad used digital twins of the macaque visual cortex to automatically generate language descriptions of what individual neurons respond to. They focused on two areas: V1 (early visual cortex) and V4 (a higher area involved in shape and color). By showing neurons thousands of images, they built AI models that could predict each neuron's firing rate. Then, they used a language model to 'translate' the neuron's preferences into concise sentences.
For V1, descriptions included terms like 'oriented edge' or 'specific spatial frequency' — matching existing mathematical models. For V4, the language became more complex, like 'conjunction of red color and vertical texture.' To test accuracy, the team generated new images based on these descriptions. In V4, those synthetic images activated 96.1% of neurons above the 95th percentile of responses to natural images (suppression images drove 97.6% below the 5th percentile). Random images only worked about 10% of the time.
They also found that while the text descriptions lost some information compared to visual features, rendering the text back into images recovered most of the lost alignment with actual neural activity.
Why It Matters
This work shows that language can capture neural selectivity in a way that is both interpretable and testable. For the first time, we have an automated method to generate hypotheses about what any visual neuron encodes, scaling up from single cells to populations. This could accelerate neuroscience research and eventually lead to better brain-computer interfaces or personalized cognitive feedback.
What You Can Do
While you can't yet access your neurons' descriptions, you can explore your own visual cognition with IQGenio's brain training games. Puzzles involving patterns, colors, and shapes challenge the same neural circuits studied here.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
