UC Davis brain implant lets ALS patient speak with 99% accuracy and work full time, no researchers needed

A man with ALS has been using a brain implant to speak independently for more than 3,800 hours over the past two years, producing nearly 2 million words with an average speed of 56 words per minute. The study, published Monday in Nature Medicine by researchers at the University of California, Davis, represents the longest sustained demonstration that a brain-computer interface can function as a practical daily communication tool outside a laboratory. Casey Harrell, the 47-year-old participant, has used the system to return to full-time work as an environmental advocate.

The implant consists of four microelectrode arrays placed in Harrell’s left precentral gyrus, the brain region that coordinates speech, recording activity from 256 cortical electrodes. Machine learning algorithms built into a software platform called BRAND, developed by UC Davis postdoctoral fellow Nicholas Card, translate that neural activity into English-language phonemes, then map those phonemes to words and sentences. The system reads out the decoded text in a synthesised version of Harrell’s pre-ALS voice.

In controlled testing with a 125,000-word vocabulary, the system scored over 99% word accuracy. In daily use outside the lab, Harrell rated 92% of sentences as accurate or mostly correct. During the study period, he communicated more than 183,000 sentences.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!

“The key thing to me is that it’s enabling everyday communication for a guy who wants to talk but can’t,” neurosurgeon David Brandman, who implanted the device in 2023 and co-led the study, told The Register. “Despite being paralysed, he has gone back to work full time and has meaningful conversations with his daughter who’s never heard the sound of his voice.”

The study’s significance lies not just in accuracy but in independence. Previous BCI systems required researchers to be present in the patient’s home whenever the device was in use, or required the patient to travel to a lab. Harrell’s system is operated by his home care team, with no researcher support needed.

Based on the study’s timeline, he averaged more than five hours of daily use.

The UC Davis team is part of BrainGate, the consortium of universities and the US Department of Veterans Affairs developing brain-computer interfaces for speech restoration, computer control, and movement recovery. The hardware itself is not custom-built, using existing microelectrode arrays produced by Blackrock Neurotech. The breakthrough is in the software, specifically the BRAND platform’s machine learning algorithms that decode attempted speech from neural signals in real time.

Brandman compared the current state of BCI technology to early pacemakers, which in the 1950s required external wiring to large batteries or wall power. Seventy years later, pacemakers are implanted in outpatient procedures. “We’re at the early stages of this kind of technology,” Brandman said.

Harrell is still wired to external computers, but the UC Davis team’s AI advances combined with hardware miniaturisation work at companies like Neuralink, Synchron, and Paradromics point toward a future where the setup is far less cumbersome.

The competitive landscape in BCI is accelerating. Neuralink has implanted devices in at least 21 patients under research protocols but lacks commercial approval. China approved the first commercially available invasive BCI earlier this year.

Other approaches to restoring speech for people with ALS use AI voice conversion rather than brain implants, but those methods require the patient to retain some vocal ability.

What distinguishes the UC Davis work is its demonstration that a BCI can cross from laboratory experiment to sustained, practical daily tool. The 3,800 hours of brain recording also constitute the largest individual neural dataset with single-neuron resolution ever collected, according to co-principal investigator Sergey Stavisky, which will inform future improvements to the decoding algorithms.

The system remains an investigational device, limited by federal law to research use, and has been tested on a single patient. Whether the results generalise to other ALS patients, or to people with different neurological conditions, is not yet known. Scaling the technology from a clinical trial to a prescribed medical device will require regulatory approval, hardware miniaturisation, and cost reduction that could take years.

“I want desperately to not be unique or special, because that will mean I no longer have the disease or that everyone that has the disease like me can get this prescribed to them,” Harrell said through his BCI system.