Rutgers Engineers’ AI Brain Interface Triples Communication Speed

“Our work paves the way for faster and more accessible communication through noninvasive brain-computer interfaces,” said Laleh Najafizadeh, who directs the university’s Integrated Systems and NeuroImaging Laboratory and serves as a professor within the Rutgers School of Engineering’s Department of Electrical and Computer Engineering.

Engineers at Rutgers University developed a noninvasive brain-computer communication interface that uses artificial intelligence to translate brain activity into words nearly three times faster than standard systems.

The system, named ChatBCI-Assist, lets users select AI-generated phrases to convey their intended meaning without spelling out every individual letter. The findings were published in IEEE Transactions on Biomedical Engineering in the study ChatBCI-Assist: An Intent-Based P300 Speller With A Locally Deployed LLM and Adaptive Stopping Strategy Enabling Record Online Spelling Performance by Jiazhen Hong, Pradyumna Rao, Weinan Wang, and Najafizadeh.

Assisting Patients With ALS

The system could make communication considerably faster and less tiring for individuals with severe movement and speech impairments. This includes people with advanced amyotrophic lateral sclerosis, or ALS, a progressive disease attacking nerve cells controlling muscles throughout the body.

“For many individuals with advanced ALS, cognitive abilities remain largely preserved, but progressive loss of motor control can severely limit their ability to express their thoughts,” Najafizadeh said. Traditional communication relies on eye-gaze-based devices using specialized cameras to track eye movements, which become difficult to use if reliable eye control is lost.

Detecting Electrical Responses

ChatBCI-Assist does not read unrestricted thoughts. The system detects what a user deliberately focuses on among choices presented on a computer screen displaying a keyboard and suggested phrases.

Rows and columns flash rapidly on the screen. When the designated row or column containing what the user means to pick flashes, the brain generates a specific electrical activity known as the P300. Sensors in a cap worn by the user measure this electrical activity through the scalp about 300 milliseconds after the stimulus appears.

AI Suggestions Prioritize Intent

The system incorporates a large language model adjusted with communication messages reflecting the needs of people with ALS. Instead of spelling out “I want water” letter by letter, a user can select the AI-generated suggestion “Give me water.”

In copy spelling experiments where participants reproduced a sentence exactly, ChatBCI-Assist achieved nearly twice the communication speed of a conventional dictionary-based predictive system. In semantic spelling, performance reached nearly three times the speed of the baseline.

The language model operates locally on the computer rather than connecting to a cloud-based service. This keeps personal messages about health and private life off the internet.

Clinical Testing Plans

The researchers evaluated the system only with healthy participants so far. Clinical testing with individuals living with ALS is in the planning stages.

Shan Chen—a physician-scientist, medical director of the ALS Multidisciplinary Clinic within the Department of Neurology at Rutgers Robert Wood Johnson Medical School, and practitioner at Robert Wood Johnson University Hospital, which is an RWJBarnabas Health facility—is working jointly with Najafizadeh. “The rapid advances in AI technology in the health sciences are extremely exciting and offer new hope for our ALS patients,” Chen said.

Common Questions About ChatBCI-Assist

What is the P300 signal in brain-computer interfaces?

The P300 is an electrical response produced by the brain about 300 milliseconds after a person sees a stimulus they are specifically paying attention to. Sensors in the user’s cap detect this response and match its timing to flashing rows and columns on a screen.

How does semantic spelling differ from copy spelling?

In copy spelling, participants must reproduce a specific sentence exactly letter by letter. In semantic spelling, users communicate the same meaning while allowing the AI to suggest shorter or differently worded phrases.

Rutgers University, chartered in 1766 as Queen’s College, is the eighth-oldest college in the United States and one of nine colonial colleges founded before the American Revolution. It became the State University of New Jersey through legislative acts in 1945 and 1956, evolving from a private liberal arts college into a major public land-grant research institution.

Today, Rutgers operates across three campuses in New Brunswick, Newark, and Camden, enrolling over 71,000 students as of Fall 2026, making it the largest university in the state and a member of the Association of American Universities.

Where does the system run its AI model?

The large language model operates locally on the computer rather than connecting to a cloud-based service, allowing the system to function without a reliable internet connection.