An artificial intelligence that "reads minds" has been developed
A "mind-reading" artificial intelligence has been developed that turns brain activity into text.Engineers have presented a noninvasive system for decoding language based on brain activity. The research was published in the journal Nature Neuroscience.
Researchers at the University of Texas at Austin have created a noninvasive semantic decoder, a system capable of transforming human brain activity while listening to a story or imagination work into a continuous stream of text. The development in the future will help people who are conscious but unable to speak, for example, after suffering a stroke, to communicate.
To teach the system to recognize speech, it is trained on patterns of a particular person's brain activity. A participant is placed in a functional MRI scanner (fMRI), where they listen to podcasts for several hours while the system records data on brain activity.
A semantic decoder trained on such a data set allows for fairly accurate recognition of a person's thoughts as he listens to other podcasts or stories he tells in his imagination. The researchers use models similar to those used in chatbots like ChatGPT or Bard AI to analyze the data.
The result is not a verbatim transcription of what is said. Instead, it captures the essence of the utterance and conveys the main idea. It generates text that closely (rarely accurately) matches the intended meaning of the source words. For example, for a participant who heard the phrase "I don't have a driver's license yet," the decoder generated the text: "She hasn't even started learning to drive yet."
Although such a system is still imperfect, it's a big advance for noninvasive speech-recognition systems, the researchers note. Previous developments required devices implanted in the brain and worked only with a limited vocabulary. Practical applications will require further development of the technology: both in the accuracy of speech recognition and in the lack of need for the participant to be in the fMRI scanner at all times.