The first time the robotic arm obeyed his mind, the man could only watch in astonishment. A small cube, previously motionless on a table, was suddenly raised into the air, grasped by mechanical fingers.

For the first time in years, he had moved something by himself. The patient had been paralyzed by a stroke years earlier before he joined as a volunteer for a novel brain-computer interface (BCI) prototype.

The BCI interface, which is powered by artificial intelligence, enabled the paralyzed man to control a robotic arm with his mind — not just for a day or two, but for seven months straight.

“This blending of learning between humans and AI is the next phase for these brain-computer interfaces,” said Dr. Karunesh Ganguly, a neurologist at the University of California, San Francisco (UCSF) and senior author of the study. “It’s what we need to achieve sophisticated, lifelike function.”

From Imagination to Action

BCIs are simply amazing. ZME Science previously reported how such technology allowed paralyzed patients to use their thoughts to type fast, connect to an iPhone, and control mechanical arms, exoskeletons, and even drones.

But as amazing as these demonstrations have been, one major challenge has stood in the way of their mass adoption: the brain is never static.

Usually, that’s a good thing. Brain plasticity is what allows us to stay in a constant state of learning, constantly adapting to our surroundings and the challenges that they might bring.

This can be a problem when you’re trying to control a third-party mechanical limb. Even in a healthy person, the neural activity associated with a specific movement changes slightly each day. When someone learns to play the piano, swing a tennis racket, or even just grasp a cup, their brain refines and shifts the way it fires signals.

The same thing happens when a paralyzed person imagines moving a limb. Over time, these shifts can throw off a BCI’s ability to interpret brain signals accurately.

Ganguly and his team suspected that this was why earlier BCIs failed so quickly. To test their theory, they worked with a study participant who had been paralyzed by a stroke. The man, who cannot speak or move, had tiny sensors implanted on the surface of his brain. These sensors picked up electrical activity as he imagined moving different parts of his body, like his hands or feet.

The researchers discovered that while the overall shape of the brain’s movement signals remained consistent, their precise locations shifted slightly from day to day. So they programmed an AI to track and adjust these shifts.

The result is a system that can maintain its accuracy over months.

Stepping up BCI technology

The next step was training the AI to translate the participant’s thoughts into precise movements. Over two weeks, the man practiced imagining simple actions, like moving his fingers or thumbs, while the sensors recorded his brain activity. This data was used to further train the AI, which then controlled a virtual robotic arm on a computer screen.

At first, the movements were clumsy. But with practice — and real-time feedback from the virtual arm — the participant improved. Eventually, he was able to transfer his skills to a real robotic arm. In a matter of days, he could pick up blocks, turn them, and move them to new locations. He even opened a cabinet, retrieved a cup, and held it up to a water dispenser.

“I’m very confident that we’ve learned how to build the system now, and that we can make this work,” Ganguly said.

Months after the initial training, the participant could still control the robotic arm after just a 15-minute “tune-up” to recalibrate the AI. Ganguly and his team are now refining the technology to make the robotic arm move faster and more smoothly. They also plan to test the BCI in a home environment, where it could help people with paralysis perform everyday tasks like feeding themselves or drinking water.

For the millions of people living with paralysis worldwide, the implications are immediately clear and exciting. A reliable BCI could restore a degree of independence, allowing users to perform tasks that most of us take for granted. It could also pave the way for more advanced applications, like restoring speech or enabling control of prosthetic limbs.

Still, this may not be for everyone. The way it’s set up, the system requires brain surgery to implant the sensors — and brain surgery is always risky. But the researchers remain optimistic.

The findings appeared in the journal Cell.