Title: Teamwork under extreme uncertainty: AI for Pokemon ranks 33rd in the world

Abstract: The highest grossing media franchise of all times, with over \$90 billion in total revenue, is Pokemon. The video games belong to the class of Japanese Role Playing Games (J-RPG). Developing a powerful AI agent for these games is very hard because they present big challenges to MinMax, Monte Carlo Tree Search and statistical Machine Learning, as they are vastly different from the well explored in AI literature games. An AI agent for one of these games means significant progress in AI agents for the entire class. Further, the key principles of such work can hopefully inspire approaches to several domains that require excellent teamwork under conditions of extreme uncertainty, including managing a team of doctors, robots or employees in an ever changing environment, like a pandemic stricken region or a war-zone. In this paper we first explain the mechanics of the game and we perform a game analysis. We continue by proposing unique AI algorithms based on our understanding that the two biggest challenges in the game are keeping a balanced team and dealing with three sources of uncertainty. Later on, we describe why evaluating the performance of such agents is challenging and we present the results of our approach. Our AI agent performed significantly better than all previous attempts and peaked at the 33rd place in the world, in one of the most popular battle formats, while running on only 4 single socket servers.