Can We Infer Move Sequences in Go from Stone Arrangements?

Inference commonly happens in our daily lives and is also a hot topic for AI research. In this paper, we infer move sequences in Go, i.e., the order in which moves are played, from stone arrangements on the board. We formulate the problem as likelihood maximization and employ a general optimization algorithm, simulated annealing, to solve it. Our experiments on professional and amateur games show that the proposed approach sometimes produces more natural move sequences than those played by humans.