Original title: Evolution of cooperation with Q-learning: how much information do we need?
Yile Ku, Xin Ou, Jiqiang Zhang, Shengfeng Deng +2 · August 2026
Why do individuals cooperate, and does knowing more help them do it? Researchers tested this using Q-learning, a method where computer agents learn through trial and error to get the largest total reward. They set up groups of agents arranged in networks and changed the size of each agent's neighborhood. This neighborhood size stood for how much information an agent received. Surprisingly, having the most information did not create the most cooperation. Instead, cooperation followed an inverted U-shaped pattern. Cooperation peaked at a moderate neighborhood size and dropped when neighborhoods were too small or too large. A medium amount of information gives agents enough detail to spot chances to cooperate back and forth. At the same time, it prevents information overload from ruining their decisions. Having just the right amount of information works best.
You are at a party deciding whom to talk to. If you only see one person, you might miss better conversations. If you see everyone, you get overwhelmed and talk to no one. Here, the number of visible neighbors is like the number of people you can see. A moderate number helps you find good partners without confusion.
This matters because it challenges the common assumption that more information always leads to better decisions. In real-world settings like team collaboration or online communities, providing too much information might actually reduce cooperation. The finding suggests that designing systems with moderate information flow could foster more cooperative behavior.
The abstract doesn't say. It covers the finding of an optimal information level in two network types using Q-learning, but does not discuss limitations such as generalizability to other learning algorithms or network structures.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Evolution of cooperation with Q-learning: how much information do we need? — Yile Ku, Xin Ou, Jiqiang Zhang et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.