[1Win4] Poster session 1

[1Win4-26] Analysis of the interaction between learning and evolution in communication emergence using evolutionary reinforcement learning.

〇Takuma Fujiwara¹, Kei Wakabayashi¹ (1.University of Tsukuba)

Keywords:AI, Evolutionary Computation, Learning Opportunities

Understanding the interaction between genetic evolution and learning in the evolution of communication is an important issue in evolutionary biology and artificial intelligence. In this study, we propose an evolutionary reinforcement learning model that introduces the effect of genetic behavioral bias into a reinforcement learning model based on Pollia's vase in order to elucidate this interaction. In the proposed model, genetic factors form the initial conditions for behavior, and learning is conducted to improve the success rate of communication based on these conditions. In experiments, we analyzed the effect of the amount of learning opportunities for each individual on genetic evolution by simulating a signaling game using the proposed model. As a result, the Baldwin effect, which is believed to promote evolution through learning, was not observed, and on the contrary, a phenomenon was observed in which an increase in learning opportunities decreased the rate of evolution. This result suggests that the effect of learning on evolution is not simple and depends on the situation.