Researchers at UCLA have found that mice and artificial intelligence systems develop similar strategies when learning to cooperate. The study compared how pairs of mice and AI agents coordinated their actions to achieve shared goals, revealing parallels in both behavior and neural activity.
The research team designed a task where two mice had to act together within increasingly short time frames to receive rewards. Using calcium imaging, scientists observed brain cell activity in the anterior cingulate cortex (ACC) as the animals performed the task. At the same time, artificial intelligence agents were trained using multi-agent reinforcement learning on a comparable cooperation challenge in a virtual setting.
Mice learned to approach their partner’s side, wait for their partner before acting, and interact mutually before making decisions. These behaviors became more frequent as training progressed. The ACC was shown to encode these cooperative behaviors; when its activity was inhibited, cooperation dropped significantly.
Artificial intelligence agents developed similar approaches, including waiting for partners and precise action coordination. Both biological brains and AI networks formed functional groups that responded more strongly to cooperative cues as performance improved. Disrupting specific neurons linked to cooperation in the AI models also reduced their ability to work together.
“We found striking parallels between how mice and AI agents learn to cooperate,” said Weizhe Hong, senior author of the study and professor at UCLA’s Departments of Neurobiology and Biological Chemistry. “Both systems independently developed similar behavioral strategies and neural representations, suggesting there are fundamental computational principles underlying cooperation that transcend the boundary between biological and artificial intelligence.”
Hong’s research program has previously examined prosocial behavior across biological and artificial systems. Earlier studies by his group showed that both mice and AI can form “shared neural spaces” during social interactions, offering insight into how different forms of prosocial behavior arise.
“Understanding cooperation is crucial for addressing some of society’s biggest challenges,” Hong added. “By studying how both biological brains and AI systems learn to work together, we can better understand the neural basis of human social behavior while also creating more collaborative artificial intelligence.”
The research was led by Weizhe Hong and Jonathan C. Kao at UCLA with contributions from Mengping Jiang, Linfan Gu, Mingyi Ma, Qin Li, among others from various departments at UCLA including Neurobiology, Biological Chemistry, Bioengineering, Electrical Engineering, Computer Engineering, and Computer Science.
Funding came from several National Institutes of Health grants as well as support from organizations such as the Packard Foundation in Science & Engineering.
###
