Scientists have developed an artificial intelligence (AI) model that can decode human thoughts and convert it into text using technology similar to that of ChatGPT. More fascinating aspect of this breakthrough is that this has been done without the use of any implants.
In a discovery, a University of Texas, Austin, team led by computer science doctoral student Jerry Tang and assistant professor of neuroscience and computer science Alex Huth, used functional magnetic resonance imaging (fMRI) machine to record 16 hours of brain activity from three human subjects as they listened to narrative stories. The researchers were able to identify neural stimuli that corresponded to individual words. The team then employed a custom-trained GPT AI model, something similar to ChatGPT, to decode this brain activity and translate it into text. However, the participants’ precise thoughts could not be captured and only a gist of what the participants were thinking was translated by AI.
In decoding perceived speech, the AI model was 72-82 percent accurate, while in decoding imagined speech, the accuracy was around 41-74 percent. In interpretations of silent movies, the accuracy ranged from 21-45 percent. The results were published in the Nature Neuroscience journal.
The scientists at University of Texas say that with this breakthrough technology, it will be possible to help the people who are physically incapable to express themselves. However, it is still developing and further research is needed before it can actually be used at a clinical setting.