For more than 3,800 hours over 22 and a half months, Casey Harrell has been plugged into a computer that speaks for him, alone at home, with only a part-time carer to make the connection. The numbers, reported by MIT Technology Review, are the new milestone in a small but accelerating field: a speech brain-computer interface that has left the lab and settled into a working life.
Harrell has amyotrophic lateral sclerosis, or ALS, the progressive neurological disease that has already taken his movement and his voice. In 2023, a surgical team implanted a small array of electrodes on the surface of his brain's speech motor cortex, the region that would normally direct the lips, tongue, and larynx to form words. A port called a pedestal sits on his skull. When a cable links that pedestal to a computer, software reads the electrical activity from the cortex and decodes it, phoneme by phoneme (each a basic unit of sound), into sentences displayed and spoken on the machine. The system first let Harrell "speak" short sentences in 2023, and the research team has been adding capability since.
What changed, according to the researchers writing in the journal Nature Medicine, is not the surgery or the electrode array. It is what Harrell has done with it. Within 22.6 months of implantation he had logged more than 3,800 hours of use at home without any researcher in the room. A carer, not an engineer, handles the daily plug-in. The hours accumulated while Harrell did ordinary things: held conversations, looked things up, performed paid work, and tried out new features the team has shipped to him over time. "Casey is the first power user of a speech BCI," said Sergey Stavisky, a neuroengineer at UC Davis and a co-author, in the MIT Technology Review account.
The framing is a deliberate one. Speech brain-computer interfaces have existed in research labs for years, and a handful of paralyzed patients have demonstrated the technology in clinic sessions. What is different here is duration, environment, and operator. Harrell uses the device largely on his own. The team has added software that lets him browse the web and do his job, and he has done both, for thousands of hours, on a single implant.
The limits of the milestone are visible in the same numbers. Three thousand eight hundred hours over 22.6 months, spread across a typical waking day, is roughly an hour or two a day, not full-time use. Harrell is one patient. The decoding vocabulary is bounded, the latency is not zero, and accents and spontaneous speech remain harder than rehearsed sentences. The pedestal, the small port on the skull that the cable mates to, is a daily commitment and a visible one. The "power user" label, as Stavisky's own quote makes clear, is a description of one man's extraordinary hours on a single device, not a population claim.
That matters because the next questions are not engineering ones. They are access questions. A surgical brain implant is not the kind of thing a patient chooses lightly, and ALS itself narrows the window in which a patient is well enough to benefit. The current array uses 4×64 electrodes on the speech cortex, and the surgery, electrode vendor, and trial registration details were not fully disclosed in the excerpt MIT Technology Review provided. The Nature Medicine paper itself will be the place to look for decoding accuracy, comparator cohorts, and adverse-event rates. What that paper cannot answer yet is durability beyond three years, language coverage, or who else will get the chance to log 3,800 hours of their own.
For Harrell, the math is simpler. "Living with a disease like ALS, you are supposed to have diminished dreams," he told MIT Technology Review. "I do not. To have all of them, and many, many more, is truly revolutionary." The system that helped him say that is, for now, the longest field test any speech brain-computer interface has had. The next test is whether anyone else gets to take it home.