The prominent similarities between the biological brain and AI during social interactions suggest basic principles

Researchers at UCLA have made important findings that during social interactions the biological brain and artificial intelligence systems develop very similar neural patterns. This first study reveals that when mice interact socially, certain brain cell types synchronize in the “shared neural space,” and AI agents develop similar patterns when engaged in social behavior.

This study, “Interbrain Neurodynamics in Biological and Artificial Intelligence Systems,” is published in the journal Nature.

This new study represents the prominent convergence of two of the two most rapidly advancing fields today: neuroscience and artificial intelligence. By directly comparing how biological brains and AI systems process social information, scientists uncover the fundamental principles governing social cognition across different types of intelligent systems.

The findings may promote understanding of social disability such as autism, while simultaneously informing us of the development of socially perceived AI systems. This is at a critical time when AI systems are increasingly integrated into social contexts, allowing us to understand the social neural dynamics essential to both scientific and technological advances.

The interdisciplinary teams in the Department of Neurobiology, Biochemistry, Bioengineering, Electrical Engineering and Computer Science at the David Geffen School of Medicine and Henry Samueli School of Engineering UCLA’s departments of neurobiology, biological chemistry, bioengineering, electrical engineering and computer science used advanced brain imaging techniques to record activity from molecularly defined neurons in the prefrontal cortex of mice during social dialogue.

Mice act as an important model for understanding mammalian brain function, especially in brain regions involved in social behavior, as they share basic neural mechanisms with humans. Researchers have developed a new computational framework for identifying higher dimensions of “shared” and “unique” neural spaces across interacting individuals.

The team then trained artificial intelligence agents to socially interact with them, applied the same analytical framework to examine neural network patterns of AI systems that emerged during social and non-social tasks.

This study revealed significant similarities between biological and artificial systems during social interactions. In both mouse and AI systems, neural activity can be split into two different components, including a “shared neural space” that contains a pattern of synchronization between interacting entities, and a “unique neural subspace” that contains activities unique to each individual.

Surprisingly, GABAergic neurons (inhibitory brain cells that regulate neural activity) exhibited significantly greater shared neural spaces compared to glutamatergic neurons, the major excitatory cells of the brain. This represents the initial investigation of interbrain neuronal dynamics in molecularly defined cell types, revealing previously unknown differences in how certain neuronal types contribute to social synchronization.

When the same framework was applied to AI agents, shared neural dynamics also emerged as artificial systems developed social interaction capabilities. Most importantly, if researchers selectively disrupt these shared neural components in artificial systems, social behavior will be significantly reduced, providing direct evidence that synchronized neural patterns causally promote social interactions.

The study also revealed that shared neural dynamics do not simply reflect coordinated behaviors between individuals, but emerge from expressions of each other’s unique behavioral behavior during social interactions.

The research team will further explore the shared neural dynamics in different, potentially more complex social interactions. They also aim to explore how shared neural space disruption contributes to social disorders and whether therapeutic interventions can restore healthy patterns of interbrain synchronization.

Artificial intelligence frameworks may serve as platforms for testing hypotheses about social neural mechanisms that are difficult to examine directly in biological systems. They also aim to develop ways to train socially intelligent AI.

“This discovery fundamentally changes the way we think about social behavior across all intelligent systems,” says Dr. Weizhe Hong, professor of neurobiology, biochemistry and bioengineering at UCLA and lead author of the new work.

“For the first time, we have shown that the neural mechanisms driving social interactions are very similar between the biological brain and artificial intelligence systems, suggesting that we have identified the fundamental principles of how biological or artificial systems process social information, whether biological or artificial.

“The meaning is important both for the development of AI that can understand human social disabilities and truly understand and engage in social interactions.”