Casey Harrell: Pioneering Speech Through a Brain-Computer Interface
Casey Harrell’s care partner begins each day by connecting the cables of his brain-computer interface (BCI), then leaves him to navigate the world on his own. The device implanted in his brain—comprising four arrays of 64 electrodes each—rests on the strip of cortex once responsible for sending speech commands to his mouth. This technology now compensates for the motor neurons that ALS has rendered nonfunctional. With it, Harrell sends emails, works, talks to his daughter, and continues his climate advocacy—all activities he pursued before ALS stole the muscles needed for speech. Remarkably, he uses this device for an average of five and a half hours daily, every day, for nearly two years, without a scientist present in the room.
While cochlear implants restore hearing by bypassing damaged hair cells to stimulate the auditory nerve directly, speech BCIs perform an even more remarkable feat. They bypass the damaged motor pathways entirely, decoding the intention to speak straight from neural activity in the cortex before any muscle movement or breath shapes sound. Harrell has been living with this second invention for almost three years. His device has accumulated more usage hours than any speech BCI in history, earning him an unprecedented title in the field: the first “power user.”
The term “power user” carries weight because for decades, BCI research primarily focused on brief, supervised demonstrations. Patients would perform speech tasks in labs under researcher supervision, with promising results followed by headlines—but rarely did the technology transition into daily life. Most late-stage ALS patients rely on slow communication tools like gaze trackers that spell out one letter at a time.
Harrell’s case, reported recently by MIT Technology Review and published in Nature Medicine, breaks this mold. He uses the implant independently at home, without researcher supervision, and for real-world purposes, including his professional work.
What 3,800 Hours of Independent Use Reveals
The Nature Medicine study highlights a staggering 3,800 hours of independent use during the first 22.6 months after Harrell’s implantation. This equates to roughly five and a half hours daily, every day, with no scientific supervision. To contextualize, most existing BCI studies report results measured in mere hours total, not thousands. This shift from supervised experiments to a genuinely usable, everyday tool represents a critical milestone.
Accuracy further underscores this breakthrough. On the first day the device was activated in August 2023, Harrell achieved 99.6% accuracy on a limited 50-word vocabulary. The research team at UC Davis then expanded the vocabulary to 125,000 words—covering essentially every English word an average speaker might need—while maintaining an impressive 97.5% accuracy. Today, the system runs at approximately 99% accuracy. These figures are not simply marketing claims but reflect a fundamental difference between a device tolerated for brief trials and one embraced as an essential part of daily life.
Decoding Speech Before It Is Spoken
In July 2023, four arrays of 64 electrodes each—totaling 256—were implanted into Harrell’s speech motor cortex as part of the pioneering BrainGate clinical trial. The speech motor cortex sends commands to muscles controlling the jaw, tongue, larynx, and lips. ALS causes the death of motor neurons carrying these commands, but the intention to speak remains intact. The implanted arrays capture the cortical activity patterns generated when Harrell attempts to say a word.
The recorded neural signals are then translated by a machine learning model trained on Harrell’s unique brain activity. This model decodes phonemes—the 39 fundamental sound units of American English—and assembles them into words and sentences. A synthesized voice then articulates the decoded speech aloud. This phoneme-based approach allows the system to scale from a 50-word vocabulary to 125,000 words without retraining, since phonemes combine to form all words.
Earlier landmark studies published in the New England Journal of Medicine, including a 2023 paper from the UC Davis group and a parallel study from Stanford, established that high-density electrode arrays could decode attempted speech faster and more accurately than previous technologies. The new Nature Medicine paper adds a crucial dimension: durability. Unlike previous devices whose signal quality deteriorated over weeks, Harrell has maintained stable, high-accuracy use for nearly two years.
The Quiet Ambition Behind Becoming a “Power User”
Mariska Vansteenel, a BCI researcher at Utrecht Medical Center with extensive experience in implanted patients, emphasizes that the real test for BCIs is not peak accuracy in a controlled environment but sustained, independent use in daily life. As reported by MIT Technology Review, Vansteenel argues that for BCIs to be genuinely relevant, they must demonstrate value and functionality without constant researcher involvement—in kitchens, bedrooms, and home offices, with caregivers rather than scientists handling setup.
Jane Huggins, who develops non-invasive BCIs at the University of Michigan, bluntly calls long-term, independent use with efficient and accurate communication the “holy grail” of BCI research.
The challenge is biological, not mathematical. While decoding algorithms have steadily improved over the past decade, implanted electrodes provoke immune responses. Scar tissue forms around the arrays, degrading signals over time. Vansteenel references a previous patient who used a fully implanted device for seven years before signal quality declined—a long run, but ultimately finite.
Harrell’s 22.6 months of stable, high-accuracy use is therefore a remarkable biological achievement. His brain has accepted the implant well enough for it to become an integral part of his life.
Harrell’s Perspective: Defying Diminished Expectations
The researchers chose to highlight a quote from Harrell that departs from typical clinical narratives. He rejects the common framing that living with ALS requires surrendering hopes and dreams. Instead, he asserts that any one of the capabilities restored by the implant would be a “godsend.” To regain all of them, and more, feels “truly revolutionary.”
This is a powerful correction to assumptions ingrained in conversations about terminal neurodegenerative diseases—that patients should seek less. Harrell’s experience shows he wanted more, and through this technology, he got it.
This shift matters profoundly as BCIs transition from research prototypes to commercial products. The people who stand to benefit most are not simply looking for incremental improvements in communication aids; they seek the restoration of full, meaningful lives they were told to grieve.
The Unsolved Challenge of Restoring the Voice
Currently, the system speaks for Harrell using a synthesized voice. It is intelligible and fast enough for conversation but not his natural voice. The UC Davis team, led by neuroengineers Sergey Stavisky and Nicholas Card, is developing a “brain-to-voice” system that aims to decode not only the words Harrell wants to say but also the prosody—the cadence, inflection, and emotional nuances that transform speech from mere information transfer into rich communication.
Prosody is a far more complex challenge than word decoding. Unlike discrete word selection, prosody is continuous and closely tied to motor control of the larynx and breath—areas still being mapped in detail. Foundational research on cortical speech control shows that the motor cortex encodes articulatory gestures rather than acoustic features directly. This means recreating a natural-sounding voice requires modeling these gestures and resynthesizing the resulting sounds.
If successful, this advancement would allow Harrell not only to express his thoughts but to sound like himself again—restoring a voice lost early in ALS progression. Family members often describe losing a loved one’s voice as a second bereavement, distinct from physical decline. Bringing the voice back would fill a profound gap that current assistive technologies cannot address.
Why This Technology Is Not Yet Widely Available
Despite impressive accuracy and extensive home use, this BCI is not yet a commercial product. Harrell represents a single case, and researchers caution that outcomes may differ across individuals due to variations in brain anatomy, disease progression, and biological responses to implants. Some patients may rationally prefer non-invasive gaze trackers over brain surgery.
The current system also requires a care partner to connect it each day. The hardware is not yet a sleek consumer device like those envisioned by companies such as Neuralink; it remains a research rig with external cabling and processing, designed for clinical trials rather than everyday convenience.
Furthermore, the regulatory path from compassionate-use trials to approved medical devices involves years of multi-patient studies, manufacturing validation, and reimbursement negotiations. Although the global population of potential beneficiaries—including people with locked-in syndrome, late-stage ALS, and severe brainstem strokes—numbers in the hundreds of thousands, this represents a serious medical device market rather than a consumer electronics timeline.
The Competitive Landscape Around Harrell’s Breakthrough
BrainGate is the longest-running invasive BCI trial worldwide, with multiple active sites including UC Davis, where Harrell’s implant was performed. Parallel programs at Stanford, UCSF, and European centers pursue similar goals using various technologies: some deploy surface electrodes resting on the cortex, others use deeper probes, and some favor wireless data transmission over percutaneous connectors.
Neuralink, the most publicly visible private company in this space, has focused on cursor control through implanted devices rather than speech decoding. Synchron takes a less invasive approach by threading electrodes through blood vessels, avoiding open-brain surgery. Precision Neuroscience develops thin-film arrays placed on the cortical surface.
Harrell’s case raises the bar for what constitutes a meaningful result in BCI research. A lab demonstration of cursor control is no longer sufficient. The new standard demands that patients use the device independently, daily, for years, in their own homes to perform real work. Until a system meets this threshold, it remains a prototype.
The Transformative Impact of Restored Speech
ALS progression varies but often affects speech early in bulbar-onset cases. Families describe a particular grief when their loved ones remain mentally alert but lose the ability to participate in conversations, tell jokes with proper timing, or interject naturally. The mind remains vibrant while the room falls silent. Existing assistive technologies enable communication but impose a slow tempo that stifles spontaneous interaction.
A BCI that enables conversation at near-natural speed with 99% accuracy reshapes this dynamic. Patients can interrupt, make jokes at the right moment, and respond before topics shift—restoring the social presence that is often lost. This aspect of Harrell’s experience defies any simple accuracy metric but is deeply meaningful.
The Longer-Term Journey Ahead
One question remains open: after 3,800 hours of use, is the synthesized voice a mere tool, like a keyboard, or has it become a prosthetic extension of Harrell himself, akin to how his brain once controlled his tongue? He has not publicly addressed this nuance, nor have the researchers explored it in their publication. What is clear is that the device enables him to sustain the dreams he was told to relinquish. Connected each morning by someone who loves him, the implant is not just a machine—it is his lifeline to expression and connection.
For more detailed information, visit the original source here.
