Theoretical studies of cooperative behavior have focused on decision strategies that depend on a partner's last choices. The findings from this work assume that players accurately remember past actions. The kind of memory that these strategies employ, however, does not reflect what we know about memory. Here, we show that human memory may not meet the requirements needed to use these strategies. When asked to recall the previous behavior of simulated partners in a cooperative memory task, participants performed poorly, making errors in 10-24% of the trials. Participants made more errors when required to track more partners. We conducted agent-based simulations to evaluate how well cooperative strategies cope with error. These simulations suggest that, even with few errors, cooperation could not be maintained at the error rates demonstrated by our participants. Our results indicate that the strategies typically used in the study of cooperation likely do not reflect the underlying cognitive capacities used by humans and other animals in social interactions. By including unrealistic assumptions about cognition, theoretical models may have overestimated the robustness of the existing cooperative strategies. To remedy this, future models should incorporate what we know about cognition.
Keywords: agent-based simulation; cooperation; decision strategy; forgetting; memory; prisoner's dilemma; tit-for-tat.