The University of Texas at Austin has created a device that can “read a person’s mind” and translate it into text using artificial intelligence.
Jerry Tang, a computer science doctoral student, and Alex Huth, a neuroscience assistant professor, led the academic team that used a model similar to that which forms the basis of ChatGPT to convert brain activity released while listening to stories or imagining stories into text.
The technology, which the researchers have dubbed a “semantic decoder,” eliminates the need for implants and is entirely noninvasive.
Huth called the development “a real leap forward” compared to previous efforts, which were mostly limited to single phrases or sentences. The model is trained to decode natural language over long periods and across complex topics.
While a user listens to podcasts, an fMRI scanner measures brain activity to learn how to translate the user’s thoughts. The decoder then generates text that, rather than being a literal transcription of the words entering the person’s brain, is meant to reflect the intended meaning of the ideas.
Participants who had not voluntarily trained the system or purposefully wanted to defy it could not have their words decoded by the system, as revealed in a recent study published in Nature Neuroscience.
“We take very seriously the concerns that it could be used for bad purposes and have worked to avoid that,” Tang said. He said the only time individuals should utilize these technologies is if they find them useful and they want to prevent that.
Although the current version of the system is dependent on an fMRI machine, it has implications for stroke victims and others who are cognitively intact but physically unable to communicate.
Major corporations and investors have spent a lot of money in recent months creating artificial intelligence (AI) tools for consumer goods and business solutions. After investing billions in ChatGPT’s developer, OpenAI, Microsoft stated that the system would be integrated into Bing and the Edge web browser, making it easier for users to get relevant results.
