Brain implant system decodes thought-speech with 97% accuracy
Researchers demonstrated that an AI-powered brain-computer interface can decode words people think silently with 97% accuracy. While the technology could help people with speech disabilities, it raises significant privacy concerns about accessing thoughts without explicit intent to communicate.
What happened
Researchers at UC Davis Health's BrainGate2 program implanted microelectrodes in the motor cortex of volunteers to record brain signals. Using AI, the system learned to recognize approximately 6,000 words with over 97% accuracy and synthesize them in the patient's voice.
The new finding is that the system can correctly decode words the participant only thought, without intending to speak them aloud. After additional training, the computer successfully decoded entire sentences that volunteers thought silently.
Reported impact
The capability could help people who have lost the ability to speak due to conditions like amyotrophic lateral sclerosis or stroke, eliminating the physical effort required when speaking is difficult. However, the technology raises serious questions about the privacy of human thoughts, as it can access what people think without their awareness.
Researchers found the system could detect words corresponding to numbers while participants counted colored shapes mentally, suggesting the technology could decode unintended thoughts. Dr. Erin Kunz noted that measures may be needed to decode only intentional speech.
Index analysis
Researchers tested a 'mental password' activation method using the unusual phrase 'Chitty Chitty Bang Bang' to control when the system decodes thoughts. One participant successfully activated the system with 98.75% accuracy, ensuring control over when thoughts are decoded.
Governance controls
The BrainGate2 program is conducted with implanted volunteers in a research setting. Neuroscientist Christian Herff stated that the mental password result transcends purely technical aspects and illuminates the mystery of language.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.