A young adult man with closely cropped dark brown hair sits in a wheelchair in a dimly lit room, facing a computer screen on which a simple phrase is typed out: “How are you today?” Underneath the question on the screen, three dots appear—similar to the ones messaging apps use to signal that your conversation partner is composing a response—and flash with anticipation. One by one, new words materialize: “I am very good.”
At first glance, this brief conversation doesn’t seem particularly special, but the man’s situation is anything but ordinary. On top of his head is a small box, about one inch square in size, connected to a circular portal in his skull about the size of a nickel. Protruding from the box is a thin black cable. The man, who goes by the nickname Pancho, is paralyzed and unable to speak, but a BCI records brain waves from the part of his cortex that is responsible for generating speech, the vocal motor cortex, and translates them into words on the computer screen.
In 2003, at the age of twenty, Pancho experienced a devastating brainstem stroke while recovering from a car accident and major surgery. After a week in a coma, he woke up paralyzed and unable to speak. The connections between his brain and his vocal muscles disrupted by the stroke, he couldn’t even make sounds as he cried.
For years, Pancho’s main method of verbal communication relied on a pointer attached to a baseball cap. He’d use the limited mobility that remained in his neck to move his head forward and down to tap keys on a computer screen, one at a time. Spelling out words like this enabled him to communicate, albeit at the agonizingly slow pace of five words per minute. “I always was told by doctors that I had zero chance of getting better,” he wrote in a 2021 interview. But in recent years, a groundbreaking brain implant and a research collaboration with scientists at the University of California, San Francisco has been, for Pancho, “very much like getting a second chance to talk again.”
The lab of neurosurgeon and scientist Edward Chang builds technologies that help paralyzed people like Pancho recover their ability to communicate by translating their brain activity into language. The Chang lab’s speech prosthesis project has incorporated many of the elements of the neuroscience tool kit we’ve discussed in other chapters; for example, like other BCIs, it uses precise e-phys recordings from specific parts of the brain that are important for generating behaviors. Before working with Pancho, Chang and his colleagues spent over a decade studying the basic neurobiology of speech in the human brain using similar recordings from non-paralyzed subjects. In brain-mapping studies that recalled Wilder Penfield’s pioneering work, his team steadily uncovered fundamental principles of how speech is represented by neural activity. Pancho’s implant, a delicate array of electrodes spread over a flat, flexible rectangular device like a business card, is placed on the surface of his vocal motor cortex, which governs speech production. Wire-thin cables protruding from the implant exit Pancho’s skull through the coin-sized opening. These electrode arrays, which contain anywhere from 128 to just over 200 separate electrodes, record the collective activity of many neurons and send the neural waveforms to a computer for analysis.
These are not the fever dreams of a tech CEO; they are the result of decades of ingenuity and perseverance in the lab and in the clinic.
The BCI relies on a phenomenon called persistent somatotopy, which is when a stroke or other injury prevents outgoing neural signals from the brain to the body (such as the mouth and tongue), but the cortical neurons that generate those signals are still functioning properly. Consequently, BCIs can read out the muscle movements these patients try to make. The idea is that, although their paralysis blocks the neural signals from reaching their muscles, those signals are still generated in the brain. In Pancho’s case, Chang and his team had no way of knowing what those signals would be like, and to what degree they would represent language with fidelity after sixteen years of paralysis (Pancho was thirty-six at the start of his collaboration with Chang).
Thankfully, Pancho’s vocal motor cortex still shows a healthy amount of neural activity, but even in ideal circumstances, it’s no small feat to translate those neural impulses to language. Here is where A.I. comes in. A model called an artificial neural network looks for patterns in the brain activity associated with phonemes, or the auditory building blocks of words, that the user reads on a screen—a feat that A.I. accomplishes much more quickly and accurately than a human data analyst. Whereas previous models matched neural activity to a limited vocabulary of words, this updated A.I. model is able to dramatically expand patients’ vocabularies by focusing on phonemes instead. The set of phonemes the model can recognize from brain activity, when combined in different ways to create different words, covers a vocabulary that amounts to nearly 90 percent of the English language.
Over time, an individual’s A.I. model improves at matching their neural activity to the words they read on a screen and focus intently on trying to say. Eventually, even when the person tries to say something spontaneously—that is, from their own mind rather than when prompted—the A.I. can predict their words.
Over the past twenty years or so, paralyzed people have been able to communicate verbally using BCIs or other devices, but the process remained slow. For example, one type of BCI recorded signals from a part of the motor cortex that let the user control the movement of a cursor on a screen. Moving it across an array of letters laid out like a keyboard, that user could type out words one letter at a time. Another device tracked its user’s eye movements, which are retained in otherwise paralyzed people who have a brainstem disorder called “locked-in syndrome,” as they looked at individual letters. Here, too, words needed to be typed out one letter at a time—not ideal for speed. These types of devices could only read out about 10 words per minute—double the speed of Pancho’s pointer-on-a-baseball-cap technique, but a long way from the 120 to 150 words per minute characteristic of natural speech. The recent BCI from Chang’s team enables a word output of almost 80 words per minute.
However, speech is more than just phonemes and words. Though it’s easy to take this for granted in an age when so much communication takes place through text, speech is biologically linked to facial expression and tone of voice. These nonverbal characteristics are important for conveying emotion, and paralyzed people unable to move their facial muscles lose their ability to communicate in this critical dimension. Recognizing this, Chang’s team designed their BCI and A.I. models to read and interpret the activity in the vocal motor regions of the brain to restore the personalized sound of someone’s voice and the facial expressions that match their speech. Though not yet able to reanimate a paralyzed person’s real face, these technologies allow them to communicate with speech and facial expressions via a humanlike avatar on a screen.
The patient at the center of this research is a woman named Ann who experienced a devastating stroke while playing volleyball with friends on a sunny day when she was only thirty years old.
Rather than just being displayed on a screen, Ann’s words, predicted by the recordings from her BCI, are “spoken” by a computer-generated avatar of a woman whose voice sounds remarkably like Ann’s used to. Creating a synthetic voice from recordings is a process referred to as “voice banking.” Patients with a neurodegenerative disease, such as amyotrophic lateral sclerosis (ALS, or Lou Gehrig’s disease), can preserve a library of vocal recordings before their condition progresses and impacts their speech. Later, the recordings can be used to create more realistic-sounding speech from a BCI like Ann’s. In her case, the avatar’s voice was generated based on recordings taken from before her stroke, including her wedding speech captured on video.
The Chang lab’s A.I. model also enables Ann to communicate nonverbally by animating the facial movements of her avatar. This most recent development was driven by another state-of-the-art feature of her speech BCI. The human vocal motor cortex comprises subregions that drive movements in discrete parts of the vocal tract, such as the tongue, jaw, and lips. Ann’s BCI translates neural activity into the avatar’s speech by focusing not on words or even phonemes, but on movements of these vocal muscles. By breaking speech down into the muscular components of individual phonemes, the team was able to re-create both speech and the individual elements of facial expression that were sent to the avatar animation. This allows Ann’s BCI to translate neural activity into facial expressions conveying emotion—in fact, with or without accompanying speech.
In their 2023 Nature paper about Ann’s BCI, the research team noted that the A.I. model made mistakes about a quarter of the time, though sentences were often still comprehensible when several of the component words were incorrect. However, as technology continues to improve, the team hopes to minimize errors and facilitate real-time communication between paralyzed patients and the people around them. One example of this is the recent development of a bilingual speech prosthesis that can seamlessly translate neural signals into English or Spanish. To accomplish this, researchers used transfer learning—the same technique used by Mackenzie Mathis to enable DeepLabCut to recognize and track body parts from lab videos. Because different languages might consist of different vocabularies but still share phonemes, their A.I. model applied the neural data recorded in one language to learn the other.
When we think about brain implants and A.I. avatars communicating with humans, we can imagine dystopian, sinister, or purely commercial applications: an A.I. with a humanlike avatar meant to dupe us or a brain implant that reads out our neural activity to queue up the next song on our morning drive playlist. Even neuroscientists, toiling away in the lab on our esoteric experiments, might admit these applications can sound cool. But ultimately, this is not what neural circuits research is all about. Pancho and Ann, on the other hand, are two real examples of how this technology—and the neural circuits research that drives it—is actually being used to help patients. These are not the fever dreams of a tech CEO; they are the result of decades of ingenuity and perseverance in the lab and in the clinic, not to mention millions of dollars in public funding.
Funding institutions like the NIH often publicize their “moonshots,” like the Human Genome Project or the BRAIN Initiative, with splashy press conferences and websites. These moonshot projects emphasize nonlinear, long-term scientific advancement, often without a specific cure or even a tangible goal in mind. But what motivates individual scientists to do basic research is often something more personal. Every day, neuroscience labs in the United States and across the world are humming along, focused on a different kind of moonshot—the moonshots that patients with brain disorders and injuries wish for themselves and their loved ones: one more milestone for a baby with cerebral palsy, or one more birthday for a grandmother with Alzheimer’s.
Though science is imperfect and fallible by definition, I think people would feel more trust if they understood a bit more about how it is done, who does it, and why.
When asked about her experience with the Chang lab’s BCI and avatar, Ann said she became emotional listening to her own voice for the first time in eighteen years, remarking that it was also the first time she’d been able to “speak” to her husband in as long. Her “moonshot” is to use the technology to one day return to a career in counseling.
*
Like Ann’s and Pancho’s BCIs, other revolutionary new treatment strategies based on the rapidly evolving neural circuits tool kit won’t necessarily come in the form of a pill. Instead, they might rely on an electrode, an engineered virus, or even a totally noninvasive method like a particular type of sensory stimulus: a flashing light or sound waves at a particular rhythm. The most high-profile example of this in recent years originated in a 2016 Nature paper by the lab of MIT neuroscientist Li-Huei Tsai and collaborators Ed Boyden (of optogenetics fame) and Emery Brown. In it, the researchers detailed a potential new treatment strategy for the devastating effects of Alzheimer’s disease.
One of the naturally occurring patterns of electrical activity in the brain is rhythmic neural firing at a frequency of 40 hertz, referred to as the gamma frequency. Gamma is associated with concentration and active engagement in tasks. Previous studies have found evidence that the brain’s gamma activity is significantly reduced in Alzheimer’s patients, motivating Tsai and colleagues to ask what happens when this brain rhythm is artificially restored. They used a genetically modified mouse strain that has long been used in the lab as a model for early onset Alzheimer’s disease. These mice have some of the same features of brain damage that Alzheimer’s patients do, most notably the formation of tangled proteins and debris in the brain and progressively deteriorating performance on cognitive tasks. In the 2016 study, the researchers asked what would happen if they stimulated neurons to fire at a gamma frequency, but instead of using electrodes, they used a noninvasive approach. They presented sensory stimuli to the mice that alternated at 40 hertz, such as flashing lights in front of the eyes, repetitive sound stimuli toward the ears, or vibrations to the skin.
To their surprise, the researchers found that when they coaxed neurons in the brains of Alzheimer’s mice at 40 hertz with sensory stimuli for just one hour, there was a marked reduction in debris in their brain tissue. Subsequent studies have given a clearer picture of why: the 40-hertz gamma frequency associated with cognitive engagement also appears to be an important brain rhythm for recruiting the activity of microglia. These activated microglia likely aid in the clearance of waste from brain tissue. In other words, the same brain rhythm that indicates heavy engagement of neural circuits—something likely to produce metabolic waste—also serves as a signal for the brain’s cleanup crew to come in and do their job.
Gamma is also a frequency that is ideally suited for activating inhibitory neurons, which scientists believe are dysfunctional in Alzheimer’s and play a major role in balancing the amount of activity in neural circuits. Interestingly, Alzheimer’s mice showed improvements in recognizing familiar objects and important spatial locations after receiving the gamma stimulation, suggesting that the disease-related debris clearance may drive improvements in cognition.
Though most of these studies have been carried out in mice—researchers like to lament that it’s the best time in history to have Alzheimer’s if you’re a mouse—the team has worked hard to replicate their results in human subjects and boost their data with clinical trials. They tested the 40-hertz stimulation technique, called GENUS (Gamma ENtrainment Using Sensory stimuli), on human participants in a study published in late 2022.
Results were promising. In most Alzheimer’s patients, a noticeable atrophy develops over the course of only a few months of disease progression. However, those who used GENUS exhibited less brain atrophy than the control group. What’s more, they also showed better sleep patterns than controls and better connectivity between brain structures associated with learning and memory. Although they didn’t show a clear difference in cognitive performance compared to the non-GENUS group, the study was only performed over the course of a few months due to disruptions from the COVID-19 pandemic, and researchers hope to see more long-term benefits in future studies. The GENUS procedure did not cause any troubling side effects—a clear benefit of using simple sensory stimuli rather than potentially toxic drugs—and is delivered with a device that provides light flashes and sound pulses at 40 hertz, which is not only inexpensive but something that patients can easily and consistently use at home.
As I look back on every thing that has changed about being a neuroscientist since I started my graduate training back in 2009, I’m exhilarated by the rapid pace of progress. Back then, I struggled to record spikes from one neuron at a time in an awake animal. I barely paid attention to the fact that the neuron could be one of many different types, because it wasn’t routine to identify what type the neuron was. When we analyzed data, we typically only looked for a few crude features from a handful of cells. It feels like that was the Stone Age compared to where we are now.
But given every thing I’ve related to you about how fast scientific progress can happen with the right resources, collaboration, and hard work, it’s deeply concerning to me that in 2025, as I write this, the funding environment for basic brain research in the United States is under unprecedented threat. Since the 1990s era of bipartisan support for the National Institutes of Health, which enabled landmark achievements like the Human Genome Project, public enthusiasm for science has waned. The NIH as a whole has taken a reputational hit in the aftermath of the COVID-19 pandemic, and these days expertise takes a back seat to political ideology, research institutions are expected to curry favor with a hostile government, and the ability of the United States to attract the best and brightest scientists from all over the world is being undermined. Even if the average person still supports much of the mainstream biomedical research community, there’s a level of skepticism and cynicism toward science that is perhaps unprecedented in our lifetimes. Though science is imperfect and fallible by definition, I think people would feel more trust if they understood a bit more about how it is done, who does it, and why.
_______________________________
Adapted from Circuit Breakers: How Neuroscientists Get Inside Your Head by Anna Chambers. Copyright © 2026. Available from Abrams Books.
Discover more from DrWeb's Domain
Subscribe to get the latest posts sent to your email.
