Treffer: Neural basis of cooperative behavior in biological and artificial intelligence systems.

Cooperation, the process through which individuals work together to achieve common goals, is fundamental to human and animal societies and increasingly critical in artificial intelligence (AI). In this study, we investigated cooperation in mice and AI systems, examining how they learn to actively coordinate their actions to obtain shared rewards. We identified key social behavioral strategies and decision-making processes in mice that facilitate successful cooperation. These processes are represented in the anterior cingulate cortex (ACC), and ACC activity causally contributes to cooperative behavior. We extended our findings to AI systems by training artificial agents in a similar cooperation task. The agents developed behavioral strategies and neural representations reminiscent of those observed in the biological brain, revealing parallels between cooperative behavior in biological and artificial systems. Editor's summary: Cooperation between individuals is a fundamental behavior in many animals. However, the neural correlates of cooperative behavior remain to be fully elucidated. Jiang et al. developed a behavioral paradigm to investigate cooperative behavior in mice and identified key behavioral elements, approach, waiting, and interaction, in the decision-making process during cooperation. In vivo imaging showed that the anterior cingulate cortex is a central hub for cooperative decision making. Finally, the authors modeled cooperation between artificial intelligence systems and showed that they developed behavioral strategies and neural representations reminiscent of those observed in mice. The study provides valuable insights into the mechanisms involved in the formation of groups and societies in the animal kingdom. —Mattia Maroso INTRODUCTION: Cooperation—the process by which individuals coordinate their actions to achieve shared benefits—is fundamental to human and animal societies and increasingly critical in artificial intelligence (AI). Cooperation often requires sophisticated integration of self-monitoring, partner observation, context-dependent decision-making, and precise temporal coordination. However, the neural mechanisms and computational principles that enable such coordination remain poorly understood in both domains. RATIONALE: We developed parallel experimental paradigms that allow direct comparison between biological and AI systems. For biological studies, we created a behavioral task in which pairs of mice must coordinate their nose poke actions within precise time windows to receive mutual rewards, combined with microendoscopic calcium imaging of the anterior cingulate cortex (ACC)—a region implicated in social cognition and decision-making. Simultaneously, we trained pairs of artificial agents using multiagent reinforcement learning to perform an analogous coordination task, which provided complete access to their "neural" computations and enabled precise experimental manipulations that are challenging in biological systems. This comparative approach allowed us to examine whether similar computational principles govern cooperation across biological and artificial systems. RESULTS: We found that mice can indeed learn to coordinate their actions for mutual benefits. Detailed behavioral analysis revealed that successful coordination depends on key social behavioral strategies that include approach, waiting, and interaction behaviors. These three preparatory behaviors increased substantially over the course of training and were more prominent during successful trials. Our neural recordings revealed that ACC neurons encoded different aspects of the cooperative process, including correct versus missed coordinative decisions, the three preparatory behaviors, and key decision-making processes. Furthermore, partner location was strongly represented in ACC activity, especially during moments requiring coordination decisions. Animals with stronger neural representations of behaviors and partner information showed better cooperative performance. Lastly, chemogenetic and optogenetic silencing experiments demonstrate that ACC activity causally contributes to cooperative behavior—inhibiting ACC neurons reduced cooperative success, impaired social behavioral strategies, and decreased the precision of coordinated actions. Artificial agents trained on an analogous mutual cooperation task developed strikingly similar behavioral strategies and neural representations. Like mice, successful artificial agents exhibited waiting behavior—pausing when their partner was far away and coordinating their movements to minimize distance differences. Analysis of the agents' recurrent neural networks revealed enhanced representations of partner-related information, paralleling our observations in mouse ACC. Furthermore, artificial neurons encoding key behavioral decisions—"hold" and "proceed"—emerged during training, and selectively disrupting these neurons impaired cooperative performance. CONCLUSION: Our findings reveal that successful cooperation in mice emerges from the combination of partner information, social behavioral strategies, and context-dependent decision-making processes. The ACC serves as a critical processing hub, representing partner information, behavioral strategies, and coordination decisions. The remarkable convergence between mouse ACC activity and artificial neural network dynamics suggests that these principles represent fundamental organizational requirements for any intelligent system engaged in real-time cooperation. These findings advance our understanding of the neural basis of social behavior and demonstrate the power of comparative approaches that bridge neuroscience and AI. Cooperation in biological and AI systems.: Both biological systems (left) and AI systems (right) engage in cooperative behavior. In biological systems, pairs of mice learned to coordinate actions within precise time windows to receive mutual rewards. In AI systems, pairs of agents navigated an artificial environment and learned to coordinate actions within specified time windows to receive mutual rewards. Through analyses of neural activities in the biological brain (the anterior cingulate cortex) and artificial brain (recurrent neural network), we reveal that both systems utilize similar computational principles underlying successful cooperation: perception of partner information, development of social behavioral strategies, and context-dependent decision-making processes that lead to cooperative outcomes with mutual benefits. [ABSTRACT FROM AUTHOR]

Copyright of Science is the property of American Association for the Advancement of Science and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)