Scientists Gave Mice and AI the Same Task, And The Results Were Strangely Similar

Cooperation is considered one of the hallmarks of intelligent behavior. For instance, when animals cooperate, they can efficiently collect food, avoid predators, and raise their offspring. For a long time, scientists have tried to understand how cooperation evolves in social species. Recently, scientists conducted an unusual experiment by bringing together two different learners: mice and artificial intelligence. The experiment suggests that the fundamental principles of cooperation might be shared by biological and artificial intelligence.

The UCLA Health researchers tried to verify this assumption in an experiment, which was published in September 2025. The scientists trained two mice to perform a task that needed tight cooperation within a narrow time frame. At the same time, computer scientists tried to teach AI agents how to solve a similar coordination problem in a virtual world. Despite the huge differences between biological and artificial intelligence, both groups began to use similar strategies.

According to the findings, the mice gradually learned to approach their partner before attempting the task. They often paused briefly and appeared to wait for the other mouse to be ready. Researchers noticed that the animals also spent more time interacting socially as they improved at the task. In fact, the time spent engaging with a partner more than doubled during the training period.

These interactions appeared to help the animals synchronize their actions. Artificial intelligence agents trained in parallel experiments developed that were comparable to the mice. The AI systems learned to delay their actions until the right moment and coordinate their timing with another agent. Reports describing the research, including summaries published in Nature Computational Science, stated that both mice and AI systems independently discovered that waiting and mutual engagement increased their chances of success.

Another area of interest for the scientists was how the mouse brain supported this behavior. A region known as the anterior cingulate cortex (ACC) became highly active when the animals coordinated with a partner, as inferred from neural recordings. According to the UCLA research team, activity in this brain region seemed to represent information about the other mouse involved in the task. The animals’ ability to cooperate declined sharply when researchers experimentally suppressed ACC activity. The artificial intelligence systems showed a shockingly similar pattern: researchers saw that certain groups of artificial neurons inside the AI networks started specializing in coordinating actions with another agent. When those cooperation-related units were disabled, the AI's performance declined in the same way that cooperation declined in mice when the ACC was inhibited. These observations were described in supporting analyses, which are referenced in ScienceDirect studies on neural circuits and cooperation.

The parallel tracks of these two systems seem to imply that cooperation is governed by some general rules. “Whoever the learner is, whether a living brain or a computer network, the same moves seem to surface when people (or agents) need to coordinate a problem.” For the scientific community, the results have implications for both animal behavior and artificial intelligence. Observing how mice naturally learn to coordinate action allows researchers to gain more insight into the neurological basis of social behavior, and at the same time, observing how artificial intelligence systems naturally learn to coordinate can help us design more convenient ways for machines and people to work together. The experiment produced a surprising conclusion: cooperation is not unique to either biology or technology; rather, it appears that cooperation comes about when agents need to learn how to coordinate and work toward a goal.