For decades, the only way to solve games was to solve two-player board games such as checkers and chess-like ones. The game outcome can be accurately and efficiently predicted using an artificial intelligence (AI), search technique, and collecting massive amounts of gameplay statistics. This method and technique can’t be directly applied to puzzle-solving because puzzles are usually played by one player (single-player) with unique characteristics such as stochastic and hidden information. The question was then raised as to whether the AI technique could be used to solve single-agent puzzles instead of solving two-player games.
Puzzles and games have been considered interchangeable for years. This may not always be true. From a real-world perspective, “game” is something we all face every day. It’s dealing with uncertainty. The unknowns of making the right or wrong decision (i.e. getting married) or not (i.e. regretting the ‘what if’). While ‘puzzle’ was already known, there is still much to discover. For example, graphene, a ‘wonder material’, was discovered and is still being commercialized. But how do you define the boundary between “puzzle” and “game” in a puzzle-solving context.
These questions were addressed by Professor Hiroyuki Iida and his colleagues at the Japan Advanced Institute of Science and Technology, Japan (JAIST), in their most recent study, published in the journal Knowledge based Systems. This research study is focused on two key contributions: (1) the definition of the solveability of a puzzle within a single-agent context via Minesweeper and (2) the creation of an artificial intelligence (AI), agent that uses the unified combination of four strategies known as PAFG solver. The proposed solver was able to solve the puzzle with a higher level of accuracy than the state-of the-art. It took advantage of both the known and unknown information in the Minesweeper puzzle.
Researchers created an AI agent that combines two knowledge-driven strategies with two data-driven techniques to maximize the use of the available and unknown information to determine the next decision. The boundary between puzzle-solving/game-playing paradigms can be determined for single-agent stochastic puzzles like the Minesweeper.
This condition is crucial in real-world problems, where the line between the unknown and known is often blurred and difficult to define. Professor Iida says that the AI agent’s ability to improve puzzle solving performance has made the boundary of solveability clear. This allowed for the definition of “puzzle” and “game”, which are common in real-life situations such as determining high stake investment and assessing the risk of making an important decision.
There were many uncertainties associated with the rapid advancement of technology and new paradigms of computing (i.e. IoT, cloud services, edge computing or neuromorphic computing). This could apply to people (i.e. technological affordance), communities (i.e. technology acceptance), society, (i.e. culture and norm) and even national levels (i.e. policy and rules modifications). Every day, human activity is a mix of ‘game’ or ‘puzzle’ conditions. The study’s lead author, Ms. Chang Liu, says that by mapping the solvability paradigm at large, it is possible to establish boundary conditions between known and unknown. This minimizes the risk of unknown and maximizes the benefit of what is known.” This feat can be achieved by combining knowledge-driven techniques, AI technology and measurable uncertainty (such a winning rate, success rate or progress rate). While keeping the puzzle challenging and fun, it can be done with knowledge-driven techniques.