Researchers have developed an AI system capable of converting brain activity into detailed text descriptions of a person’s thoughts, according to a report from DNYUZ.
Scientists have developed a new artificial intelligence system capable of translating human brain activity into detailed text descriptions, according to a report from DNYUZ. The technology, known as “mind captioning,” uses MRI scans and advanced AI models to produce real-time commentary of what a person is thinking.
Developed by researchers at the University of California, Berkeley, and Japan’s NTT Communication Science Laboratories, the system combines AI training on more than 2,000 short video clips with brain-scan data. Each clip is paired with captions to generate a “meaning signature,” described as a digital fingerprint of the narrative content.
The report states that a second AI model then learns to match those signatures to MRI readings taken from volunteers who watched the same clips. Once trained, the decoder can analyze a new scan and predict what a person is seeing, often producing sentences that closely align with the actual visual content.
In one test, a participant viewed a video of someone jumping off a waterfall. The AI initially predicted “spring flow” but later refined its description to “a person jumps over a deep water fall on a mountain ridge.” Across multiple trials, the system successfully identified the correct video from 100 options half the time.
Researchers say the technology could aid individuals with paralysis or speech loss, though it also raises privacy concerns. As UC Berkeley’s Alex Huth said, “Nobody has shown you can do that, yet.”