Practicing cooperative skills shapes brain-wide networks

The article addresses how humans and nonhuman primates acquire and execute complex cooperative behaviors and which neural resources support these skills. It highlights that effective cooperation requires tracking others’ actions and identities, coordinating joint goals, and managing shared rewards while executing one’s own actions. The piece discusses new evidence from Franch et al. showing that practicing cooperative behaviors shapes activity in domain-general cortical networks—specifically visual area V4 and dorsolateral prefrontal cortex (dIPFC)—to highlight task-relevant social and sensory inputs.

The commentary situates the work within prior research on primate cooperation and social learning. Ethological and experimental studies show macaques can learn cooperative tasks, such as pulling strings or loops to obtain food (Barbary and Japanese macaques). Social gaze is known to facilitate learning by observation in rhesus macaques. Eye-tracking in freely moving nonhuman primates has been rare, previously achieved in ring-tailed lemurs. Beyond domain-general regions, specialized social brain areas (e.g., dorsal anterior cingulate cortex and middle superior temporal sulcus) contain neurons that signal decisions to cooperate and predictions of partners’ choices. The article also references long-term stability and plasticity of neural representations (e.g., hippocampal maps; progressive plasticity in visual cortex), framing questions about how practice reshapes neural circuits over time.

The piece summarizes methods used by Franch et al.: pairs of rhesus macaques were trained to cooperatively and simultaneously press buttons to pull a food tray. Behavioral performance was tracked across training sessions (simultaneity and delays). One monkey per pair wore two head-mounted cameras: a scene camera from the actor’s perspective and a mirror-based camera to record eye movements, enabling wireless eye tracking in freely moving macaques. A hidden Markov model estimated transition probabilities among key events (gazes, button presses) to infer how visual information guides cooperation. Wireless neural recordings were obtained from domain-general regions—visual area V4 and dorsolateral prefrontal cortex (dIPFC)—synchronized with eye-tracking. Neuronal selectivity and discrimination of cues were quantified with classifiers (support vector machine). Population decoding weights assessed distribution of information within areas, and spike-timing coordination analyses quantified functional coupling within and between V4 and dIPFC (feedforward and feedback directions).

• Behavioral training effects: simultaneous button presses increased from ~60% in the first session to ~80–90% in later sessions; the delay to initiate cooperation decreased by 93%.
• Gaze behavior: after the actor pressed the button, the most probable next event was looking at the partner, followed by the partner pressing their button. With training, the probability that watching the partner (or food) was followed by pressing increased by 220%.
• Neural coding: neurons in both V4 and dIPFC responded to gaze fixations and action choices during cooperation; dIPFC showed a higher proportion of mixed selectivity. Classification performance for social cues (viewing reward vs. partner) improved with practice in both V4 and dIPFC, while discrimination among non-social cues (self-button vs. random objects) did not improve. Discrimination between choices (self vs. partner choices) improved during training in dIPFC but not in V4.
• Network changes: within-area information about gazes and actions became more evenly distributed across neurons (more uniform decoder weights). Between areas, spike-timing coordination increased bidirectionally between V4 and dIPFC, suggesting reciprocal entrainment. V4 neurons that increased coordination with dIPFC contributed more to discriminating social cues, and dIPFC neurons that increased coordination with V4 contributed more to discriminating both social cues and social actions.

The findings indicate that practicing cooperative skills tunes domain-general cortical systems to prioritize task-relevant social information and actions. Social gaze emerges as a key driver guiding cooperative sequences, aligning behavioral improvements with enhanced neural discrimination of social cues. As training progresses, information is distributed more broadly within populations and coordination strengthens between sensory (V4) and executive (dIPFC) regions, supporting a brain-wide network perspective for cooperation. This addresses the broader question by showing that cooperation is not only represented in specialized social circuits but also shaped in general-purpose visual and prefrontal networks that integrate social and sensory signals during learning.

The commentary highlights that cooperative practice reshapes neural representations in domain-general networks (V4 and dIPFC), improving discrimination of social cues and strengthening interareal coordination, consistent with a brain-wide mechanism for social skill acquisition. It points to future directions: (1) elucidating interactions between domain-general regions and specialized social areas (e.g., dorsal anterior cingulate cortex, superior temporal sulcus) during cooperation, and (2) determining how neural coordination and selectivity evolve after mastery or periods without practice, including the stability or decay of learned representations. Continued development of wireless eye-tracking and chronic neural recording methods in freely moving primates will enable deeper, longitudinal investigations.

The summarized study focuses on domain-general regions (V4 and dIPFC) and does not directly measure specialized social circuits, leaving their contribution during training unresolved. Neural and behavioral dynamics are characterized during the training phase; how representations evolve after task mastery or inactivity remains unknown. Eye-tracking and neural recording were obtained from one member of each pair, potentially limiting coverage of dyadic neural interactions. Generalizability to humans is inferred but not tested directly.