AI-Aided Brain-Computer Interface Improves Speed and Task Accuracy

A team of engineers at the University of California Los Angeles (UCLA) has developed a noninvasive brain-computer interface (BCI) that incorporates artificial intelligence (AI) to support users in completing physical and virtual tasks. The wearable system interprets signals from the brain and uses a camera-based AI “co-pilot” to assist with cursor movements and robotic arm control.

The system, described in Nature Machine Intelligence, combines electroencephalography (EEG) signal decoding with real-time visual input. The study assessed the technology in four individuals, including one person with paralysis, and demonstrated faster task completion when AI assistance was active.

Developing a shared control system

BCI technologies aim to convert neural activity into control signals for devices, but many existing systems require surgical implantation of electrodes. While implanted BCIs have shown precise control, their use has remained limited due to the invasive nature of neurosurgery. By contrast, external BCIs such as those based on EEG are less risky but often lack sufficient accuracy and responsiveness.

The UCLA system addresses this gap by decoding users’ movement intentions from EEG signals. These decoded commands are then interpreted by an AI model, trained to understand user goals using video input. This combination allows for shared control, where the AI assists the user in refining or completing actions.

Participants wore an EEG cap and performed two tasks. In the first, they moved a computer cursor to various targets, maintaining its position over each target briefly. In the second, they used a robotic arm to move blocks on a table to designated positions.

AI assistance improved performance

All participants, including the individual with paralysis, completed the tasks more quickly when the AI system was active. Without AI, the participant with paralysis was unable to complete the robotic task. With AI assistance, they finished in approximately six-and-a-half minutes.

The results suggest that AI-enhanced decoding of EEG signals can increase both speed and task accuracy. Importantly, the AI system inferred user intent independently of eye-tracking data, relying instead on combined neural and visual inputs.

The study was conducted in the Neural Engineering and Computation Lab at UCLA and supported by the National Institutes of Health and the Science Hub for Humanity and Artificial Intelligence, a joint initiative between UCLA and Amazon. A patent application has been filed for the BCI system.

Reference: Lee JY, Lee S, Mishra A, et al. Brain–computer interface control with artificial intelligence copilots. Nat Mach Intell. 2025. doi:10.1038/s42256-025-01090-y

This content includes text that has been generated with the assistance of AI. Technology Networks' AI policy can be found here.