AI Research in Games: The Role of StarCraft 2

Artificial Intelligence is an ill-defined term that can broadly refer to any kind of computer program that processes information. However, one unifying goal of such a poorly partitioned domain is the understanding of what may constitute intelligence through the construction of equivalent systems. It follows naturally that recent advances in various machine learning disciplines have caused a renewed interest in such topics. The main driver behind these advancements is a biologically inspired class of algorithms known generally as “deep learning”. The popularity of such systems have upended many expectations, notably the limitations of learning systems which have been classically understood to be weaker than manually designed expert systems. Though remarkable, this resurgence has not resulted in a direct improvement to the creation or understanding of intelligent systems, as most existing learning systems are still limited to narrow, highly engineered domains. Video games allow us to leverage the budding development of such systems due to several inherent features, including to the complexity that is usually demanded of them for human entertainment. With the examination of successful selflearning systems in the challenging domains of two player board games, as well as the progressing work in recreating those results in StarCraft 2, video games may well become the hotbed of advances in the quest to understand intelligence