Memory Performance of Master Go Players

Our overall goal is to study the use of informa tion by Go players and the structure of their Go knowledge In particular in this paper we focus on human memory and conclude with a discussion of the bene ts of modelling human memory to AI and Go programs We report on two memory experiments The rst experiment used Japanese Go players to replicate earlier studies on Australian Go play ers The second experiment consists of three case studies of master Go players to dan amateurs The general task in the experiments was to re construct Go positions stone by stone in cor rect game order Two separate tasks were per formed in the episodic task the moves from a Go game were shown to the subjects in a se quential presentation in the inferential task the subjects had to reconstruct Go positions with no information about how the game was played In both tasks feedback was provided after placement of each stone The rst experiment replicated previous results for the experienced and beginner subjects The case studies showed that extremely high levels of memory performance by the master subjects extended even to very fast presentation rates