The cost of passing - using deep learning AIs to expand our understanding of the ancient game of Go

抄録

In addition to their playing skills, AI engines utilizing deep learning neural networks provide excellent tools for analyzing traditional board games if we define new measures based on their raw output. For the ancient game of Go, we develop a numerical tool for context-sensitive move-by-move performance evaluation and for automating the recognition of game features. We measure the urgency of a move by the cost of passing, which is the score value difference between the current configuration of stones and after a hypothetical pass in the same board position. In this paper, we investigate the properties of this measure and describe some applications for developing learning tools and analyzing a large number of games. As we use AI tools to gain new insights into the ancient game of Go and develop a more precise, quantified understanding, this work fits into the more significant and general project of utilizing superhuman AI engines for deepening human understanding and growing human knowledge.