Overview
- Stanford researchers implanted microelectrode arrays in the motor cortex of four participants with ALS or post-stroke paralysis to capture neural signals associated with inner speech.
- The AI-driven system translated these signals into text in real time, recording word accuracy as high as 74 percent in small-vocabulary tests and error rates up to 54 percent with larger word sets.
- To prevent unintended decoding of spontaneous thoughts, the team added algorithmic filters and required users to think the passphrase “Chitty-Chitty-Bang-Bang,” which was detected nearly 99 percent of the time.
- Experts warn that the system remains too slow and inconsistent across different tasks and vocabulary sizes and that many spontaneous thoughts may not form clear linguistic sentences.
- Researchers plan to expand coverage to additional brain regions, boost processing speed and accuracy, and address ethical and regulatory considerations before clinical deployment.