What changed. Earlier brain–computer interfaces (BCIs) had already shown that neural activity from attempted speech can be turned into text with high accuracy. Two things had not been shown: that a person could use such a system independently, at home, with no technical team in the room, and that it would keep working for years rather than a few supervised sessions. This result closes both gaps in one participant, a 45-year-old man with paralysis and severe dysarthria (loss of clear speech) from amyotrophic lateral sclerosis, the disease that progressively severs the brain's control over the muscles, including those used to speak. About 25,000 Americans were living with ALS as of the most recent full federal estimate (CDC National ALS Registry, 2017); for many, losing speech is among its most isolating effects.
The science. Four small electrode arrays sit in the speech-and-movement region of his cortex and record individual neurons firing as he attempts to speak the intention to move the lips, tongue and vocal tract persists even after the muscles stop responding. A "brain-to-text" decoder converts those firing patterns into words on a screen, and the system can speak them aloud in a synthesized version of his own pre-ALS voice; a second decoder turns attempted hand movement into computer-cursor control. The work comes from the BrainGate2 consortium, developed at the University of California, Davis, with Brown University and the Mass General Brigham Neuroscience Institute.
The data. He used the BCI for more than 3,800 hours at home across nearly two years near-daily and unsupervised and across 183,060 sentences, about 92% were decoded at least mostly correctly (53% completely correct, another 13% corrected by him on screen, 26% mostly correct). In structured tests, where he repeated words shown on a screen, the system reached 99.2% word accuracy against a 125,000-word vocabulary at about 31 words per minute; using a faster "silent speech" strategy he reached roughly 50 words per minute at 96.5% accuracy. Either way, that is many times faster than the roughly 6 words per minute his previous assistive tools allowed. He also used the speech decoder as a keyboard and the cursor decoder as a mouse sending texts and emails, browsing the internet, joining video calls, and holding down full-time employment while paralyzed. His longest single correctly decoded utterance ran to 215 words.
What's next. The figure that matters most here is not the 99.2% but the 3,800 hours: the study's real claim is that the system held up through ordinary daily life, unattended, for nearly two years the threshold a lab demonstration has to clear before it can become technology someone actually lives with. Studies in more participants, work toward wireless or fully implantable hardware, and the long road through regulatory review are what would turn one man's restored conversations into something available to others. For now, the result establishes that it is possible that someone locked out of speech by ALS can, with an implant he controls himself, hold real conversations with the people in his life for years.