Researchers at The University of Texas at Austin have developed a groundbreaking semantic decoder that translates brain activity into a continuous text stream, as detailed in a study published in Nature Neuroscience. This non-invasive AI system, trained using a transformer model akin to those in systems like ChatGPT, may offer new communication avenues for individuals unable to speak due to conditions such as strokes. During the training phase, participants listen to podcasts while undergoing fMRI scans, allowing the decoder to learn correlations between their brain activity and corresponding text.
The decoder functions by capturing the essence of thoughts rather than creating exact transcripts; it generates text based on brain activity during listening or imagination tasks. For instance, one participant’s encoded thought shifted from “I don’t have my driver’s license yet” to “She has not even started to learn to drive yet.” While the system shows promise, it currently relies on significant fMRI time, limiting practical application. Researchers are hopeful that adaptations to portable brain-imaging technologies like near-infrared spectroscopy may broaden its use. They emphasize ethical considerations, ensuring it operates only with consent and cooperation. Future developments may require regulatory measures to protect user privacy.
