“Mind reading” sounds like science fiction. In 2023, a team at the University of Texas at Austin published something that came surprisingly close, and the details are more interesting than the headline.
What did the researchers actually build?
They built a semantic decoder: an AI system that turns brain activity, recorded with a non-invasive fMRI scanner, into a continuous stream of text. No implant, no surgery.
To train it, volunteers spent hours lying in the scanner listening to podcasts. The system learned how each person’s brain responded to meaning, and paired that with a language model to generate candidate sentences.
How accurate was it?
Not word for word, but the meaning often survived. When a participant heard “I don’t have my driver’s license yet,” the decoder produced “She has not even started to learn to drive yet.”
That is the key idea: fMRI is too slow to capture individual words, so the decoder works at the level of ideas. It even described events from short silent films the volunteers watched, which suggests it was picking up something deeper than language itself.
Can it read anyone’s mind?
No, and this is the most important part of the study.
- The decoder only worked on the person it was trained on. On anyone else, the output was unintelligible.
- Training required many hours of willing cooperation inside a scanner.
- Volunteers could defeat it on purpose. Counting by sevens, naming animals or imagining a different story was enough to break the output.
Why does it matter?
The main goal is medical: helping people who have lost the ability to speak to communicate again, without brain surgery. The researchers also raised the obvious question themselves. As brain-reading tools improve, mental privacy will need clear rules about who may decode a mind, and when.
The decoder did not read minds against anyone’s will. It read cooperating minds, imperfectly. That distinction is where the real debate begins.