Brain-Computer Interfaces: From the Utah Array to Neuralink

In a research lab, a man who has not moved his arms in years watches a cursor glide across a screen and click on a folder. He is not touching a mouse. He is not speaking. He is simply thinking about moving, and a grid of electrodes resting on the surface of his motor cortex is catching the faint electrical chatter of those intentions and translating them into commands. For the participant, the experience can feel almost ordinary, like operating a familiar tool. For the engineers and neuroscientists watching, it is the payoff of a decades-long effort to build a bridge between living neurons and silicon.

This is the promise of the brain-computer interface, or BCI: a direct channel between the nervous system and a machine, bypassing the muscles and nerves that disease or injury have silenced. The field has produced genuinely astonishing results, and it has also generated some of the most overheated headlines in modern technology. Understanding BCIs means learning how they actually read and write brain signals, where the technology came from, and how to tell the durable science from the marketing.

How the brain becomes a signal

Every thought, movement, and sensation in your body rides on electricity. Neurons communicate by firing brief voltage spikes called action potentials, and when many of them fire together, they produce electrical fields large enough to be detected. A brain-computer interface is, at its core, a device that listens to this activity and finds patterns inside it.

The methods fall along a spectrum of how invasive they are. Non-invasive BCIs sit outside the skull. The most common is electroencephalography, or EEG, which uses electrodes on the scalp to read the summed electrical activity of millions of neurons. EEG is safe, cheap, and has been used in clinics since the 1920s, but the skull and scalp blur the signal, so the spatial resolution is poor. It is like trying to follow a single conversation by standing outside a stadium and listening to the roar.

Invasive BCIs place electrodes directly on or into the brain tissue, which buys far sharper signals at the cost of surgery. A middle option, electrocorticography, lays a sheet of electrodes on the brain's surface beneath the skull. The most precise approach pushes tiny needle-like electrodes into the cortex itself, close enough to eavesdrop on individual neurons. The closer you get to the source, the clearer the message, and the higher the medical risk.

The Utah Array and the birth of intracortical recording

The workhorse of invasive human BCI research is a small device called the Utah Array, developed at the University of Utah in the late 1980s and 1990s. It is a square silicon base roughly four millimeters across, studded with a bed of about 100 sharp microelectrodes, each one designed to sit among the neurons of the cortex and record their spikes. The whole thing looks a little like a tiny hairbrush, and it remains one of the most important tools in the field.

The Utah Array became famous through the BrainGate research program, a long-running academic collaboration. In a landmark study published in the early 2000s, a participant with paralysis used a Utah Array implanted in his motor cortex to control a computer cursor and operate simple devices purely by intending to move. In the years since, BrainGate participants have used the same basic technology to control robotic arms, including reaching out to drink from a bottle, and to type by selecting letters with a thought-driven cursor. These results are well documented in the peer-reviewed literature and have been replicated across multiple participants.

The Utah Array proved a profound point: the brain keeps generating movement-related signals even when the body cannot act on them. Years after paralysis, the motor cortex still lights up with the intention to move, and a machine can learn to decode it.

Decoding: turning spikes into intentions

Recording the brain is only half the problem. The harder half is decoding, the process of turning a messy stream of electrical activity into a usable command. This is where modern machine learning has transformed the field.

A decoder is a statistical model that learns the relationship between neural activity and intended action. During a calibration session, the participant imagines or attempts a set of movements while the system records which neurons fire and how. The decoder learns, for example, that a particular pattern of activity corresponds to "move right" and another to "move up." Once trained, it can take live brain signals and predict intent in real time, often within milliseconds.

Speech decoding has become a striking frontier. Recent academic work has used electrode arrays over the speech-related regions of the cortex to help people who have lost the ability to speak, including some with paralysis from stroke or ALS. By decoding the brain's attempts to produce words, these systems have driven a digital avatar or text output at conversational speeds far faster than older letter-by-letter methods. The accuracy is not perfect and the vocabulary and conditions are constrained, but the demonstrations are real and were published in respected journals. They show that even abstract intentions like words can, in principle, be read from cortical activity.

Writing back: stimulation and the two-way street

So far we have talked about reading the brain. The deeper ambition is to write to it, to feed information in rather than only pulling it out. This is done by electrically stimulating neurons, nudging them to fire with small pulses of current.

The most established example is not a research curiosity but a mainstream medical device: the cochlear implant, which has restored a sense of hearing to hundreds of thousands of people worldwide. A cochlear implant bypasses damaged parts of the ear and stimulates the auditory nerve directly, and while it is not a cortical device, it proves that the nervous system can learn to interpret artificial electrical input as meaningful sensation.

In experimental BCIs, researchers have used stimulation of the sensory cortex to give participants a crude artificial sense of touch, so that a robotic hand can send back a faint signal when it grips an object. Deep brain stimulation, meanwhile, is an approved therapy in which implanted electrodes deliver pulses deep in the brain to ease the tremors of Parkinson's disease and to treat certain other disorders, helping many thousands of patients. These two-way and stimulation-based systems are still far less mature than read-only decoding, and writing rich, specific information into the brain remains one of the field's great unsolved challenges.

Enter Neuralink and the new wave of companies

For most of its history, BCI research lived in universities and hospitals. That changed as well-funded companies entered the field, the most visible being Neuralink, founded by Elon Musk in 2016. Neuralink's stated long-term goals are sweeping, but its near-term work is squarely medical: helping people with paralysis control computers by thought.

Neuralink's technical contribution is mostly about scale and packaging. Rather than a rigid Utah-style array, its implant uses many thin, flexible threads carrying a large number of electrodes, inserted by a purpose-built surgical robot, with the electronics sealed in a small wireless device beneath the skull. The aims are more recording channels, no wires poking through the skin, and a more practical path to everyday use. In 2024, Neuralink reported implanting its device in its first human participants under a clinical trial, and shared demonstrations of a participant moving a cursor and playing simple games. Other companies, including Synchron with a stent-based electrode threaded through a blood vessel and Precision Neuroscience with a thin surface array, are pursuing less invasive routes to similar goals.

It is worth being precise about what is new here. The core ability, decoding cursor or device control from motor cortex, was established years earlier by academic groups like BrainGate. What the new companies bring is engineering: higher electrode counts, wireless operation, scalable manufacturing, and the capital to push toward an approved product. That is genuinely important, but it is evolution of a proven idea, not a sudden leap into science fiction.

Separating promise from hype

The gap between BCI reality and BCI marketing is wide, so it helps to hold a few honest distinctions in mind.

First, restoring function is real; enhancing healthy brains is not. The strong, replicated results all involve restoring lost abilities to people with serious injury or disease. The popular vision of typing at the speed of thought, downloading skills, or merging with artificial intelligence has no experimental basis today. Decoding intended movement is a tractable engineering problem; reading the full content of a person's private thoughts is not something any current device can do.

Second, the body fights back. Implanted electrodes provoke the brain's immune response, and over months scar tissue can build up around them, degrading the signal. Keeping an array recording cleanly for years remains a major hurdle, and long-term reliability is an active area of research rather than a solved problem.

Third, bandwidth is tiny. Even the most advanced arrays sample a vanishingly small fraction of the brain's roughly 86 billion neurons. We are reading a few hundred or a few thousand cells in one small region, which is enough for cursor control but nowhere near enough for the grand visions sometimes advertised.

Fourth, surgery is serious. Any device that opens the skull carries real risks of infection and damage, which is why much of the field is also pursuing safer, less invasive approaches even at the cost of signal quality.

Key Takeaways

Brain-computer interfaces are one of the most genuinely impressive achievements in modern neuroscience, and also one of the most over-promised. The core science is solid: the brain encodes intentions as electrical activity, devices like the Utah Array can record that activity even after paralysis, and machine-learning decoders can translate it into cursor movements, robotic control, and even attempted speech, while stimulation-based systems like cochlear implants and deep brain stimulation prove the channel can run both ways. Companies like Neuralink have advanced the engineering with higher-density, wireless implants and have begun human trials, but they are building on decades of academic groundwork rather than inventing the field overnight. The honest picture is a powerful medical tool, still limited by tiny bandwidth, biological wear on implanted electrodes, and the risks of surgery, with the dream of thought-reading mind-uploads firmly in the realm of speculation. Understanding that distinction, between the remarkable things BCIs already do and the fantastical things they are imagined to do, is the most useful thing a curious reader can carry away.