An AI can now read a person’s thoughts from a brain scan and turn them into text, and a new method makes it far quicker to set up. Researchers at the University of Texas at Austin found a way to adapt this kind of AI brain decoder to a brand new person with much less training than before. The work was published in the journal Current Biology in February 2025.
The advance matters because a faster, easier setup brings the technology closer to a real use: helping people with aphasia, a brain disorder that makes it hard to produce and understand language.
How the brain decoder learns a new person fast
The first version of this decoder was powerful but demanding. To train it, a person had to lie still in a brain scanner for about 16 hours while listening to audio stories, and the finished model worked only for that one person.
The new study removes much of that burden. The team wrote a converter algorithm that maps one person’s brain activity onto another’s, a method called functional alignment. With it, an existing decoder can be adapted to a new person using only about an hour of scanning while watching short silent videos such as Pixar shorts.
Reading meaning, even from silent video
The decoder does not spell out the exact words a person heard. Instead it produces text that is close in meaning rather than word for word. When a test passage described someone talking about a job they disliked, the decoder returned a different sentence with the same idea of an unpleasant, boring job.
The surprising part is that a converter trained only on silent films still built a working language decoder. That points to a shared semantic representation that does not depend on the input: the brain seems to treat a story it hears and a story it watches in similar ways. This fits a wider line of research showing that MRI can pick out simple content from brain activity under strict lab conditions.
Why this could help people with aphasia
Aphasia affects about a million people in the United States, and it can block both speaking and understanding speech. That is a problem for older decoders, which were trained by having a person listen to and follow spoken stories.
Because the new approach can train on silent video, it offers a route to a brain-computer interface that does not require language comprehension. To check whether the idea could reach patients, the researchers simulated aphasia-like brain lesion patterns and found the decoder could still translate the story a person was perceiving. The team is now working with an aphasia specialist to test it in people who have the condition.
Limitations and quality of evidence
This is an early, proof-of-concept result. The test participants were neurologically healthy, so the benefit for people with aphasia is suggested by simulations rather than proven in patients. The decoded text stays approximate, and the study involved a small number of people scanned in a large, costly MRI machine.
There is also a built-in privacy limit. The system works only with cooperative participants: if someone whose brain was used for training later resists by thinking about something else, the results are unusable. Readers interested in this technology can follow the peer-reviewed sources below and treat single studies as a starting point, not a settled conclusion.
Sources and related information
Current Biology – Semantic language decoding across participants and stimulus modalities – 2025
The peer-reviewed paper reports that a decoder can transfer across people and modalities, using a converter trained on far less data than the original model required.
UT Austin News – Improved brain decoder holds promise for communication in people with aphasia – 2025
The university’s release describes how an hour of silent video can adapt the decoder to a new person, the aphasia motivation, and the cooperation requirement that limits misuse.
Live Science – AI brain decoder can read a person’s thoughts with just a quick brain scan – 2025
Skyler Ware’s report explains that the decoded text is close in meaning, not exact, and quotes an outside neuroscientist on what the modality-independent result means.
