A brain-computer interface, or BCI, is a system that reads electrical activity from the brain and translates it into commands a computer or device can act on. The technology has moved well past the proof-of-concept stage: people with severe paralysis have used implanted BCIs to control robotic arms, type on screens, and even produce decoded speech. At the same time, noninvasive versions worn on the scalp are being tested for everything from wheelchair control to consumer gaming. The gap between what invasive and noninvasive systems can do remains large, but both branches are advancing fast enough that the practical, ethical, and commercial questions surrounding BCIs are becoming urgent.
How a BCI Reads the Brain
Every thought, movement intention, or sensation you experience corresponds to patterns of electrical activity fired by neurons. A BCI’s job is to pick up those patterns, run them through decoding software, and convert them into an output, whether that is moving a cursor, triggering a robotic limb, or generating text. The quality of everything downstream depends on how cleanly the system can capture those signals in the first place.
Invasive BCIs place electrodes directly on or inside the brain, which gives them access to sharp, detailed neural signals. The tradeoff is brain surgery. Noninvasive BCIs sit outside the skull, typically using EEG caps that detect electrical activity through the skin and bone. They avoid surgery entirely, but the signals they pick up are blurry by comparison, limited in spatial resolution and prone to interference from muscle movements and ambient noise.1PubMed Central. A holistic perspective on noninvasive brain-computer interfaces This resolution gap is the central tension in the field: the best-performing systems require opening the skull, while the most accessible ones struggle to decode fine-grained intentions.
What Happens When Electrodes Stay in the Brain
For implanted BCIs, the brain does not simply accept the hardware and move on. Within hours of electrode insertion, the immune system mounts a response. Specialized brain cells called microglia swarm the implant site, triggering inflammation. Over days and weeks, another type of cell, astrocytes, proliferate around the electrodes, forming a layer of scar tissue known as a glial scar. That scar increases the electrical resistance between the electrode and nearby neurons, gradually degrading the quality of recordings and stimulation.2PubMed Central. Enhancing biocompatibility of the brain-machine interface: A review
This is one of the biggest unsolved problems for long-term implants. A device that works beautifully during a clinical trial can lose signal fidelity over months or years as scar tissue builds up. Researchers are attacking the problem from the materials side: flexible electrodes that bend with brain tissue rather than poking rigidly against it cause less mechanical irritation and produce thinner scars.3PubMed Central. Research Progress on the Flexibility of an Implantable Neural Microelectrode One study found that very thin, straight electrode wires produced notably less scarring than thicker or zigzag-shaped alternatives, suggesting that both size and geometry matter.4PubMed Central. Glial scarring around intra-cortical MEA implants with flexible and free microwires inserted using biodegradable PLGA needles The goal is an electrode so soft and slim that the brain barely notices it is there.
Restoring Movement After Paralysis
The most established clinical use of BCIs is helping people with paralysis regain some form of motor control. The approach that has received the most attention involves recording signals from the motor cortex, the strip of brain responsible for planning and executing movement, and using software to decode what motion the person intends. That decoded intention then drives an external device: a robotic arm, a computer cursor, or in some cases the person’s own muscles via electrical stimulation.
One line of research goes a step further by bridging the gap between the brain and the spinal cord. In a rat model of cervical spinal cord injury, researchers showed that brain signals could be decoded and used to trigger spinal stimulation, restoring forelimb movement that the injury had knocked out.5PubMed. Brain-Computer-Spinal Interface Restores Upper Limb Function After Spinal Cord Injury The concept of a brain-to-spine bridge is compelling because it works with the body’s existing wiring rather than replacing it entirely. Animal models remain central to developing and validating these systems before they reach human trials.6PubMed. Monkey models for brain-machine interfaces: the need for maintaining diversity
Bidirectional BCIs add another dimension. Rather than just reading motor commands out of the brain, they also write sensory information back in. In one study, a person with paralysis controlled a robotic arm while receiving touch feedback through tiny electrical pulses delivered to the somatosensory cortex, the brain area that processes tactile sensations. The combination of vision and artificial touch improved their ability to grasp and manipulate objects.7PubMed Central. A brain-computer interface that evokes tactile sensations improves robotic arm control Without that feedback loop, controlling a prosthetic arm is a bit like trying to pick up a glass while your hand is completely numb: you can see it, but you cannot feel how hard you are gripping.
Decoding Speech from Brain Activity
For people who have lost the ability to speak, BCIs offer a route to communication that does not depend on any residual muscle control. Researchers have shown that electrodes in the motor cortex can pick up the neural patterns associated with intended speech, even in people who can no longer move their lips, tongue, or vocal cords. The decoded signals are then mapped to words or phonemes and displayed as text or synthesized as audio.
The hardest test case is someone with locked-in syndrome, a condition where a person is fully conscious but almost completely unable to move or speak. A recent study demonstrated that an intracortical BCI could decode intended speech from a person with advanced ALS who had long-standing anarthria, meaning they had lost the ability to produce speech sounds entirely, and who was dependent on a ventilator. The system decoded phonemes, individual words, and higher-level language units above chance levels.8PubMed Central. Decoding intended speech with an intracortical brain-computer interface in a person with long-standing anarthria and locked-in syndrome Sentence-level accuracy was lower than what has been achieved in participants who still have some ability to move their speech muscles, but the result was a significant milestone: it showed that the motor cortex retains enough speech-related neural patterns to be useful even years after a person stops being able to vocalize.9Cell Reports. Area 6v encodes orofacial movements and phonemes in a locked-in participant
The researchers noted that future improvements could come from training participants to imagine speech in a closed-loop setting and from decoding linguistic units rather than individual sounds. The finding that usable speech decoding from motor cortex may be feasible in people with complete loss of speech articulation opens a path toward communication devices for patients who currently have almost no options.
Avoiding Open-Brain Surgery
A major barrier to wider adoption of implanted BCIs is the surgery itself. Craniotomy, where a section of skull is temporarily removed, carries risks of infection, bleeding, and tissue damage. An alternative approach sidesteps the skull entirely by threading an electrode through a blood vessel. A device called the Stentrode uses an endovascular approach: it is delivered via the jugular vein, navigated up into a blood vessel sitting on the surface of the motor cortex, and expanded like a stent. The first human trial of this device, involving four patients with severe paralysis, showed that it was possible to record usable neural signals from inside a blood vessel without ever opening the brain.10PubMed Central. Assessment of Safety of a Fully Implanted Endovascular Brain-Computer Interface for Severe Paralysis in 4 Patients
The signal quality from a blood vessel is not as high as what you get from electrodes sitting directly on or in brain tissue. But the tradeoff may be worth it for many patients: the procedure is more like an angiogram than brain surgery, and the recovery is far less involved. For people whose medical condition makes craniotomy too risky, or who simply are not willing to undergo open-brain surgery, endovascular BCIs could be a practical middle ground.
Closed-Loop Stimulation for Parkinson’s Disease
BCIs are not limited to reading signals out of the brain. A related application uses the same sensing technology to deliver stimulation back into the brain in real time, adjusting automatically based on what the sensors detect. This closed-loop approach is being developed for Parkinson’s disease, where conventional deep brain stimulation delivers a constant electrical pulse regardless of the patient’s moment-to-moment state. Adaptive deep brain stimulation senses physiological biomarkers, such as characteristic brain rhythms associated with Parkinson’s symptoms, and modulates the stimulation accordingly.11PubMed Central. Closing the loop in DBS: A data-driven approach
The appeal is precision: instead of blasting the same stimulation whether a person is resting, walking, or sleeping, the system ramps stimulation up when symptoms flare and dials it down when they subside. Early clinical data from chronic adaptive stimulation in Parkinson’s patients showed improvements in overall wellbeing and a trend toward better general movement.12npj Parkinson’s Disease. Chronic adaptive deep brain stimulation for Parkinson’s disease: clinical outcomes and programming strategies The technology is still being refined, particularly in identifying the right biomarkers to guide stimulation and in building systems robust enough for everyday use.13PubMed Central. Closed-Loop Adaptive Deep Brain Stimulation in Parkinson’s Disease: Procedures to Achieve It and Future Perspectives
Memory Prosthetics
Perhaps the most striking frontier for BCIs is the prospect of augmenting or restoring memory. The hippocampus, a brain structure essential for forming new long-term memories, is often damaged by Alzheimer’s disease, stroke, or traumatic injury. Researchers have developed a prosthetic system that mimics how the hippocampus processes information. In rat experiments, when hippocampal function was chemically blocked and the animals lost the ability to form new memories, delivering electrical stimulation patterns derived from a computational model of normal hippocampal activity restored their performance on memory tasks.14PubMed Central. A hippocampal cognitive prosthesis: multi-input, multi-output nonlinear modeling and VLSI implementation
This work has progressed to humans. In a study involving epilepsy patients who already had electrodes implanted for seizure monitoring, researchers tested whether the same model-based stimulation could improve memory encoding. Participants who received stimulation patterned on successful memory trials performed better on recall tasks, and the benefit was most pronounced on harder trials with longer delays between learning and recall.15Journal of Neural Engineering. Developing a hippocampal neural prosthetic to facilitate human memory encoding and recall Random stimulation, by contrast, did not help and sometimes made things worse, which suggests that the specific pattern of stimulation matters enormously. This is not a memory chip you could implant and walk away from; it requires precise modeling of an individual’s neural activity to know what signals to send and when.
The Brain and Machine Learn Together
One underappreciated aspect of BCIs is that they are not a one-way street, even at the level of basic learning. When you first use a BCI, the software is trying to figure out your brain signals, and your brain is simultaneously adjusting how it produces those signals to get better results from the software. Researchers call this co-adaptation, and it turns out to be critical for performance. Systems that update their decoding algorithms in tandem with the user’s neural changes outperform static ones.16PubMed Central. Co-adaptive Training Improves Efficacy of a Multi-Day EEG-Based Motor Imagery BCI Training
This mutual adaptation means that a BCI is less like a TV remote, where you press a button and the device obeys, and more like learning to ride a bicycle, where you and the machine develop a working relationship over time. The interface has to handle the inherent noisiness and variability of brain signals, which change from day to day and even minute to minute.17PubMed. Probabilistic co-adaptive brain-computer interfacing Building adaptive intelligence into the interface, rather than relying on a fixed decoder, transforms the device from a passive tool into something closer to a collaborative partner.18PubMed. Exploiting co-adaptation for the design of symbiotic neuroprosthetic assistants For users, this also means that BCI skill improves with practice. The first day on a new system is typically rough, but over sessions, both the user’s brain and the machine’s algorithms converge on a shared communication strategy.
Consumer EEG Headsets and Their Limits
You can buy wireless EEG headsets for a few hundred dollars that claim to read your brainwaves. These consumer devices do pick up real brain signals, and they can detect certain well-known neural events, such as the P300 response, a spike in electrical activity that occurs about 300 milliseconds after you notice something unexpected. However, the overall temporal pattern of brain activity these devices record is distorted compared to research-grade EEG systems.19PubMed Central. Evaluation of consumer-grade wireless EEG systems for brain-computer interface applications
What this means in practice is that consumer headsets can handle simple, coarse tasks: detecting whether you are focused or relaxed, selecting from a small number of on-screen options, or providing neurofeedback during meditation. They are not going to decode your thoughts, control a prosthetic hand, or do anything close to what clinical-grade BCIs achieve. The marketing around some of these products can blur this distinction, implying capabilities that the hardware simply cannot support at its signal quality level.
Shrinking the Hardware
Size matters for implanted electronics. Larger devices cause more tissue damage, provoke stronger immune responses, and limit where in the nervous system they can be placed. A research platform called neural dust aims to shrink implantable sensors to sub-millimeter dimensions. These tiny devices, sometimes called motes, use ultrasound rather than radio waves for both power and communication, which solves several problems at once: ultrasound penetrates tissue more efficiently than electromagnetic signals, and it can power devices deep in the body without batteries.20PubMed. Recent advances in neural dust: towards a neural interface platform
Early demonstrations showed that neural dust motes could wirelessly transmit high-quality electrical recordings from the peripheral nervous system of rats.21Neuron. Wireless Recording in the Peripheral Nervous System with Ultrasonic Neural Dust A related device called StimDust extended the concept to stimulation, packing both recording and stimulation capabilities into a mote just 1.7 cubic millimeters in volume, small enough to sit on the tip of a pencil.22arXiv. StimDust: A mm-scale implantable wireless precision neural stimulator with ultrasonic power and communication The long-term vision is to scatter many such motes across the brain or body, creating a distributed network of sensors and stimulators that could be implanted with minimal surgery and maintained indefinitely without battery replacements. That vision remains years from clinical reality, but the engineering foundations are being laid.
Cortical Visual Prosthetics
BCIs for vision restoration take the bidirectional concept and apply it to the visual system. The idea is to bypass damaged eyes or optic nerves entirely and stimulate the primary visual cortex directly, producing patterns of light perception called phosphenes. A camera worn on glasses captures the scene, a processor converts it into electrical stimulation patterns, and an electrode array on the cortex delivers those patterns. The result is not normal sight, but a grid of light spots that can convey shapes, edges, and movement well enough to help a blind person navigate or recognize objects.23Springer Link / Neurotherapeutics. Brain Machine Interfaces for Vision Restoration: The Current State of Cortical Visual Prosthetics Several groups around the world are actively developing and testing cortical visual prostheses, driven by advances in electrode design and wireless data transmission. The technology is still experimental, and the resolution of artificial vision remains far below that of a healthy eye, but for people with no other options it represents a possible path back to functional sight.
Commercialization Challenges
Turning a BCI from a research prototype into a product people can actually get has proven extremely difficult. The patient populations that need implanted BCIs, such as people with locked-in syndrome or high-level spinal cord injuries, are relatively small. Manufacturing the devices is expensive. Surgery, hospital stays, and the extensive rehabilitation needed to train both the patient and the system drive costs even higher. Even devices that have received regulatory approval through humanitarian exemption pathways have struggled commercially because the economics do not scale well.24PubMed Central. Brain-computer interface commercialization
Beyond cost, several technical barriers stand in the way of widespread adoption. Long-term implant stability, as discussed with glial scarring, remains an open challenge. Clinical workflows for implantation, calibration, and ongoing support are not yet standardized. And decoding algorithms that work well for one person do not automatically transfer to another, meaning each system requires individualized tuning.25arXiv.org. Clinical Translation of Brain-Computer Interface in China: A Landscape Analysis of Investigator-Initiated Trials, Registered Clinical Trials, and Regulatory Approval The recent entry of well-funded private companies into the space has accelerated hardware development, but the fundamental question of who pays for these devices, and whether health systems will cover them, has no clear answer yet.
Privacy and the Ethics of Reading the Brain
As BCIs become more capable, they raise questions that sit at the intersection of neuroscience and civil liberties. If a device can decode what words you intend to say, could a more advanced version someday decode what you are thinking but choose not to say? The concept of “mental privacy” has generated significant debate. A study involving interviews with Chinese experts in the field concluded that current BCI-based mind reading does not yet pose unique ethical challenges compared to other forms of inferring someone’s mental state, such as polygraphs or behavioral analysis, and therefore does not yet justify creating a distinct legal right to mental privacy.26PubMed Central. Does brain-computer interface-based mind reading threaten mental privacy? ethical reflections from interviews with Chinese experts
That “not yet” does a lot of work. The experts’ reasoning was that today’s BCIs decode specific, intended motor signals, not free-floating thoughts or private emotions. But the technology is on a trajectory toward higher resolution and broader signal interpretation, and governance frameworks tend to lag behind the technology they are meant to regulate. Other concerns are more immediate: who owns the neural data a BCI collects? Can it be subpoenaed? Could an employer require its use? These questions are not hypothetical for the people already enrolled in BCI clinical trials, and the regulatory landscape has barely begun to catch up.