Georgetown Law
Technology Review

In January 2024, a thirty-year-old quadriplegic named Noland Arbaugh became the first person to receive Neuralink’s implanted brain chip. Within weeks he was moving a cursor across a screen, playing chess, and browsing the internet—all by thinking about moving his hand. A few months later, researchers at the University of California, Davis activated a different kind of brain implant in Casey Harrell, a man with amyotrophic lateral sclerosis (ALS) who had lost the ability to speak. Within minutes, the system was converting his brain activity into text at roughly 97% accuracy. And in January 2026, a startup called Merge Labs emerged from stealth with over $250 million in funding, promising to read and write brain signals using ultrasound rather than electrodes.

These are all examples of brain-computer interfaces, or BCIs. Not long ago, the concept of BCIs could be comfortably filed under science fiction: telepathic communication, mind‑controlled machines, seamless merging of humans and computers. But today, a convergence of better hardware, smarter algorithms, and serious investment has moved BCIs from the realm of laboratory curiosity to early clinical reality. In 2023, Nature Electronics named BCIs its “technology of the year,” reflecting both technical progress and new clinical demonstrations.

The basic premise is simple to describe: a BCI measures activity in the brain, translates those signals into a form a computer can understand, and uses that translation to control an external device. The systems underlying these BCIs, however, sit at the intersection of neuroscience, electrical engineering, and machine learning—fields many people understandably have not studied since college, if at all. This explainer therefore aims to open that black box. Part II introduces core BCI concepts in plain language. Part III surveys the main ways BCIs measure brain activity, from fully implanted electrodes to noninvasive caps and emerging ultrasound‑based systems. Part IV walks through how software and machine‑learning models decode neural signals. Part V highlights a few flagship applications. Part VI describes what has changed in the last few years. Part VII addresses limits and misconceptions. Finally, part VIII closes with a forward‑looking synthesis of where the field appears to be headed and why it matters.