A groundbreaking study from UCLA has shown that mice and artificial intelligence neural networks display remarkably similar patterns when learning to cooperate. Published in Science, the research offers fresh insight into the fundamental principles of cooperation that may transcend both biology and technology, reshaping how we understand social interaction and how we design collaborative AI systems.
Cooperation is one of the pillars of human society, essential for teamwork, diplomacy, and survival itself. When it breaks down, we often see conflict and instability. To explore the neural basis of cooperation, UCLA researchers designed a unique experiment where pairs of mice had to coordinate their actions within increasingly tight time windows—sometimes as short as 0.75 seconds—to earn rewards. Using advanced calcium imaging technology, the team monitored brain activity in the anterior cingulate cortex (ACC), a region critical for social decision-making.
The mice developed three main strategies: approaching their partner’s side, waiting before acting, and engaging in mutual interactions. These cooperative behaviors strengthened as the mice became more skilled, with interaction-based strategies more than doubling during training. When ACC activity was inhibited, cooperation sharply declined, proving this brain region plays a vital role in coordinated behavior.
To compare biological and artificial systems, researchers built AI agents trained through multi-agent reinforcement learning on a virtual version of the same cooperation task. The AI systems not only succeeded but also developed strikingly similar strategies to the mice, including waiting and precise coordination. Both biological brains and AI neural networks organized themselves into functional groups that prioritized partner-related information as cooperation improved. When specific cooperation-related circuits in AI were disrupted, performance dropped dramatically—mirroring the effects seen in mice.
This parallel suggests that computational principles of cooperation may be universal, spanning across species and even into artificial systems. According to senior author Weizhe Hong, both mice and AI independently developed similar strategies, hinting at a shared framework for social coordination. The findings build on Hong’s broader body of work, which includes studies on how animals display rescue-like behaviors and how brain dynamics synchronize during social interactions.
The implications extend far beyond the lab. By uncovering these common principles, researchers could not only deepen our understanding of human social behavior but also design more collaborative AI systems capable of working seamlessly with humans. Such insights could improve everything from workplace collaboration tools to autonomous systems that must interact cooperatively in real-world environments.
In conclusion, the study demonstrates that cooperation is not just a product of human culture but a principle that emerges naturally from both biological brains and artificial networks. By studying these parallels, we may uncover powerful new ways to foster collaboration—both among humans and between humans and machines.
