A solver of single-agent stochastic puzzle: A case study with Minesweeper

摘要

People have enjoyed solving puzzles for decades because of the challenge and the satisfaction derived from solving problems. However, although previous researches focused on the puzzle’s complexity, strategy, and solving automatically, few studies worked on sorting out puzzles from a solvability way related to the stochastic elements among the solving process. The contribution of the study is twofold. Firstly, a single-agent stochastic puzzle definition is established via Minesweeper testbed, a well-known puzzle synonymous with Microsoft Windows. Secondly, this study proposes an artificial intelligence (AI) solver based on the obtained information on the board, called the ‘PAFG’ strategy, which stands for the primary reasoning, the advanced reasoning, the first action strategy, and the guessing strategy. The first two strategies take advantage of knowledge-based rules and linear system transformation (Gauss–Jordan elimination algorithms) to determine the probability of making a move independently. The last two strategies explore the beginning and ways to determine hidden puzzle states to enhance the winning rate of the AI solver. The experimental simulation of various configurations and the AI solver with PAFG strategy yielded a high-level winning rate of 96.4%, 86.3%, and 45.6% for the 9×9|10, 16×16|40, and 16×30|99 Minesweeper board configuration, which is comparable to the state of the art study. Thus, such an AI solver could contribute to classifying single-agent stochastic puzzles and establishing the boundary of the puzzle-solving and game-playing paradigm.