Yaodong Yang, Lantao Yu, Yiwei Bai, Jun Wang, Weinan Zhang, Ying Wen, Yong Yu

In this paper, we conduct an empirical study on discovering the orderedcollective dynamics obtained by a population of artificial intelligence (AI)agents. Our intention is to put AI agents into a simulated natural context, andthen to understand their induced dynamics at the population level. Inparticular, we aim to verify if the principles developed in the real worldcould also be used in understanding an artificially-created intelligentpopulation. To achieve this, we simulate a large-scale predator-prey world,where the laws of the world are designed by only the findings or logicalequivalence that have been discovered in nature. We endow the agents with theintelligence based on deep reinforcement learning, and scale the populationsize up to millions. Our results show that the population dynamics of AIagents, driven only by each agent’s individual self interest, reveals anordered pattern that is similar to the Lotka-Volterra model studied inpopulation biology. We further discover the emergent behaviors of collectiveadaptations in studying how the agents’ grouping behaviors will change with theenvironmental resources. Both of the two findings could be explained by theself-organization theory in nature.