Scientists have shown that artificial intelligence can make an experimental bionic eye more precise by learning how an individual brain responds to electrical stimulation. Researchers from UC Santa Barbara and collaborators published the proof-of-concept study in Neuron, as described in a press release.
A bionic eye sends electrical signals directly to the visual cortex, the brain region that processes vision. It bypasses the eyes and optic nerves, so it could help people who lost sight to stroke, disease, or injury. In 2024 the team tested its approach in Spain with a 27-year-old man blinded by a traumatic brain injury. He had a temporary implant of 96 electrodes in his visual cortex. When the electrodes delivered current, he perceived phosphenes, meaning spots, flashes, or shapes of light.
Learning from the brain's responses
The researchers trained a deep neural network to predict how the man's cortex would react to different stimulation settings. The model also received measurements of resting brain activity just before each test, letting it account for the brain's current state. Stimulation patterns chosen by the AI hit their neural targets more accurately than standard methods while using less electrical current.
A second result may matter just as much. Brain activity recorded after stimulation predicted what the man reported seeing, including a phosphene's shape, size, brightness, and color, better than the stimulation settings themselves. In other words, measuring what the brain actually did revealed more about perception than knowing which electrodes had been activated.
The study matters because it points toward implants that adapt to each user instead of relying on fixed settings. Most visual prostheses send predetermined signals into brains that differ from person to person and day to day. A device that monitors the brain's responses and adjusts its signals could give more consistent results. The technology remains experimental, and the participant saw only flashes and shapes of light, not detailed images. Even so, the work shows how AI could close the loop between an artificial vision device and the brain.