In a groundbreaking study published in Nature, a team of researchers from UCLA has uncovered remarkable parallels between the neural dynamics of biological brains and artificial intelligence systems during social interactions. This pioneering work bridges the gap between neuroscience and AI, revealing that both living brains and AI agents synchronize their neural activity when engaged in social engagement, suggesting a shared set of underlying principles governing social cognition across fundamentally different types of intelligences.
The research focused on the dorsomedial prefrontal cortex (dmPFC) of mice, a brain region intricately involved in processing social information. Utilizing cutting-edge brain imaging methods that capture activity from molecularly specific neuron types, the team observed how individual neurons coordinate during real-time social exchanges. These experiments spotlighted a complex interplay of neuronal activity, demonstrating that socially interacting mice exhibit synchronized patterns within what the team defines as “shared neural spaces,” while simultaneously maintaining unique neural signatures reflective of individual behavior.
To complement and extend these biological insights, the researchers engineered artificial intelligence agents to perform social tasks and applied the same analytical framework developed for the mice brain data. This approach was novel in its examination of high-dimensional neural subspaces—conceptual mathematical spaces where patterns of neural activity can be decomposed into shared and unique components. In the AI networks, these subspaces similarly revealed synchronized neural patterns when the agents engaged socially, juxtaposed with distinct individual behaviors during non-social tasks.
One of the most striking discoveries was related to the role of specific neuron types in social synchronization. The study distinguished between GABAergic neurons, which are inhibitory and modulate overall neural excitability, and glutamatergic neurons, the primary excitatory cells of the brain. The inhibitory GABAergic population demonstrated significantly larger shared neural spaces during social interaction compared to glutamatergic neurons, pointing to a previously unrecognized contribution of inhibitory neurons in driving inter-brain social synchronization. This nuance in cell-type activity adds a new layer of complexity to our understanding of neural communication during social behavior.
Equally compelling was the causal demonstration achieved within the artificial intelligence systems. By selectively disrupting the shared neural components in AI agents, the scientists observed a marked reduction in social behavior capabilities. This experimental manipulation provides the first direct evidence that synchronized neural patterns are not merely correlative but functionally indispensable for social interactions, highlighting a causal mechanism analogous to biological systems.
These findings shed light on the longstanding question of how complex social behaviors emerge from neural circuits. Importantly, the shared neural dynamics do not simply mirror coordinated actions among individuals but rather encode detailed representations of each other’s unique behaviors during interactions. This insight implies that the neural basis of social cognition involves a dynamic, mutual encoding of social partners’ behaviors, fostering a richer and more flexible communication framework within interacting intelligences.
The interdisciplinary nature of the work, combining neurobiology, bioengineering, computer science, and electrical engineering expertise, exemplifies a new paradigm in the study of social cognition. By harnessing computational models informed by biological data, the study opens a promising avenue for understanding social disorders such as autism spectrum disorders (ASD), where social synchronization is known to be disrupted. This convergence could inform the development of targeted therapeutic strategies that restore or enhance neural synchronization in affected individuals.
Beyond the biological and clinical implications, the study propels the field of artificial intelligence into a new era of social awareness. Current AI systems, while increasingly embedded in social contexts like virtual assistants and autonomous agents, often lack a robust foundation in social neural dynamics. This research provides a scaffold for designing AI that can understand and genuinely engage in complex social interactions, moving closer to the goal of creating socially intelligent machines capable of nuanced human-like communication.
Future research announced by the team aims to delve deeper into more complex social scenarios, moving beyond dyadic interactions to group and hierarchical social dynamics. The complexity of natural social behavior demands understanding how these shared neural spaces evolve in larger networks and whether similar cell-type-specific dynamics persist in more elaborate social contexts. By exploring these dimensions, scientists hope to unravel the mechanistic principles that scale from simple to complex social cognition.
Moreover, the team plans to investigate pathological disruptions in shared neural spaces and their potential roles in social disorders. Understanding how breakdowns in neural synchronization relate to behavioral impairments could pave the way for interventional approaches, possibly using novel neurotechnologies to modulate neural circuits and restore healthy social function. This translational aspect emphasizes the bidirectional benefits between neuroscience and AI, where advances in one domain can catalyze breakthroughs in the other.
From a technical perspective, the novel computational framework developed to dissect shared and unique neural subspaces represents a significant methodological advance. This approach transcends traditional population neural analysis by enabling high-dimensional, cell-type-specific characterizations of interbrain synchrony and inter-agent network dynamics. Such tools are invaluable for dissecting the complexity of neural communication, not only in social interactions but potentially in other cognitive processes that require integrated neural coordination.
The implications of this research reverberate far beyond the academic sphere. As AI systems become more deeply embedded into everyday human environments—from education to healthcare and social platforms—the ability for machines to interpret and adapt to nuanced social cues rooted in neural dynamics will become essential. This work indicates that successful social AI may require architectures that mimic not just behavioral output but the underlying synchronized neural frameworks found in biological brains.
Lead author Weizhe Hong encapsulated the significance of this discovery, stating, “This discovery fundamentally changes how we think about social behavior across all intelligent systems. We’ve shown for the first time that the neural mechanisms driving social interaction are remarkably similar between biological brains and artificial intelligence systems. This suggests we’ve identified a fundamental principle of how any intelligent system—whether biological or artificial—processes social information.”
In sum, the UCLA team’s findings illuminate a shared neural architecture underpinning social cognition across life and machine. This landmark study not only bridges disparate research fields but also lays the foundation for revolutionary advances in understanding, diagnosing, and potentially treating social dysfunction, while simultaneously paving the road toward socially competent artificial intelligence capable of genuine interaction in human-centered environments.
Subject of Research: Animals
Article Title: Inter-brain neural dynamics in biological and artificial intelligence systems
News Publication Date: 2-Jul-2025
Web References: 10.1038/s41586-025-09196-4
References: Inter-brain neural dynamics in biological and artificial intelligence systems, Nature 2025; DOI: 10.1038/s41586-025-09196-4
Keywords: Artificial intelligence
