Personalized AI decodes real-time emotions from brain signals for affective BCIs
Researchers trained personalized deep learning models to decode continuous valence and arousal from intracranial recordings in gray and white matter. The models generalized across different viewing tasks and achieved low-latency real-time decoding, advancing affective brain-computer interface applications.
The research applies personalized deep learning models to intracranial recordings from both gray and white matter, decoding continuous measures of valence and arousal — the two core dimensions of emotional experience. Because the models were trained per individual, they can account for how each person's brain represents feeling states. Their ability to generalize across different viewing tasks indicates the decoded signals capture emotion itself, not merely the specifics of a single activity.
Low-latency real-time decoding is essential for affective brain-computer interfaces, systems designed to sense and respond to a user's emotional state as it happens. By pairing personalization with speed, this approach moves the field closer to practical closed-loop devices. The work reflects a broader trend in neuroscience toward adaptive neural interfaces that interpret internal states continuously rather than through discrete commands.
Affective brain-computer interfaces could eventually help people with severe motor or communication disabilities express emotional states without speech or movement. Clinicians may also use such systems to monitor mood in real time for conditions like depression or anxiety, though intracranial recording requires invasive surgery, limiting near-term use to medical settings. Privacy concerns around decoding internal feelings will likely shape how these technologies are deployed, and regulatory frameworks may need to address who can access such intimate neural data.