Gaming the Known and the Unknown through Puzzle Solving with an Artificial Intelligence Agent

Researchers have devised multiple strategies for an artificial intelligence (AI) agent to solve a random puzzle like Minesweeper.

For decades, attempts at solving games have been unique in solving two-player games (ie, games such as checkers, chess, etc.), where game outcomes can be accurately and efficiently assessed using some artificial intelligence (AI). ) Search technology and collecting huge amounts of gameplay stats. However, such methods and techniques do not apply directly to the puzzle-solving domain, as puzzles are usually played alone (single player) and have special features (such as random or hidden information). So, the question arises as to how the AI ​​technique can retain its functionality to solve two-player games, but instead apply the same agent to the puzzle?

Over the years, puzzles and games have been considered interchangeable or interchangeable. In fact, it may not be all the time. From a real-world perspective, ‘play’ is something we encounter every day; Dealing with strangers. For example, not making the right decision (i.e., getting married) or making the wrong decision (i.e., quitting the job) or not making the right decision (i.e., remorse for ‘what if’). Meanwhile, the ‘puzzle’ is known to be there, and something that has not yet come out is also hidden. One such known context is, for example, the discovery of ‘amazing’ material AI boundary of solvability

The toy illustrates AI strategies that use knowledge-based strategies to deal with unknown information, while adopting data-based strategies to use the known information of the Minesweeper puzzle. The resulting explorations form the boundary for solvability in a single player random puzzle that is canonical to a wide range of real-world problems. Credit: Hiroyuki Iida from JAIST

The researchers adopted an AI agent with two science-based strategies and two data-based strategies to make the best use of the known and unknown information of the current decision to better evaluate the next decision. As a result, a single-agent can establish a boundary between puzzle-solving and game-playing pattern for random puzzle Minesweeper.

The boundary between the known and the unknown is usually blurred and such a situation plays a particularly important role in real-world problems that are very difficult to identify. As Professor Iida commented: “With the AI ​​agent’s ability to improve puzzle solving performance, the boundaries of solvability are clearly visible. Assessing the level of risk and so on. In essence, we all live within ourselves Minesweeper The world is trying to figure out a way forward for us, avoiding the ‘bomb’ in our lives.

There are many uncertainties with the face-paced advancement of existing technology and new models of computing available (i.e. IoT, cloud-based services, EDGE computing, neuromorphic computing, etc.). This situation may be true of individuals (i.e., affordability of technology), community (i.e., acceptance of technology), society (i.e., culture and standard) and national levels (i.e., changes in policy and rules). “Every day human activity involves a lot of ‘game’ and ‘puzzle’ situations. However, mapping the solvability pattern on a scale can create boundaries between the known and the unknown, reducing the risk of the unknown and increasing the usefulness of the known,” said Ms. Chang, lead author of the study. Liu explains: “Such a feat is achieved by keeping the puzzle fun and challenging with knowledge – driven techniques, AI technology and measurable uncertainty (winner rate, success rate, progress rate, etc.).”

Reference: Chang Liu, Shunky Huang, Gao Nying, Mohd Nor Akmal Khalid and Hiroyuki Ida, 28 March 2022, “A Soldier of Single-Agent Stochastic Puzzle: A Case Study with Minesweeper” Knowledge-based systems.
DOI: 10.1016 / j.knosys.2022.108630

About Japan Advanced Institute of Science and Technology, Japan

Established in 1990 in Ishikawa Prefecture, the Japan Advanced Institute of Science and Technology (JAIST) is the first independent national graduate school in Japan. Now, after 30 years of steady progress, JAIST has become one of Japan’s top universities. JAIST is calculated with multiple satellite campuses and seeks to promote competent leaders with a diversity-critical education system; About 40% of its alumni are international students. The University has a unique style of graduate education based on carefully designed course-based curricula to ensure that its students have a solid foundation for conducting cutting-edge research. JAIST also works closely with local and foreign communities by promoting industry-academy collaborative research.

About Mrs. Chang Liu from Japan, Japan Advanced Institute of Science and Technology

Ms. Chang Liu is a doctoral student at the School of Advanced Science and Technology (JAIST), Nomi, Japan. Her research focuses on researching appeal information about the evolution of puzzle games based on game mechanics and player experience, overseen by Professor Hiroyuki Ida in the Lab of Entertainment Technology. She analyzes important factors in the evolution of ancient puzzle games from ancient times and solves puzzles to find the line between puzzles and games and analyzes information in the process of playing games.

About Professor Hiroyuki Ida from Japan, Advanced Institute of Science and Technology, Japan

Dr. Hiroyuki Ida received her Ph.D. Heuristic Theories on Game-Tree Search from Tokyo University of Agriculture and Technology, Japan in 1994. Since 2005, he has been a professor at JAIST, where he is also Trustee and Vice President of Educational and Student Affairs. He is the head of the Iida Laboratory and has published over 300 papers, exhibitions and books. His research interests include Artificial Intelligence, Game Informatics, Game Theory, Mathematical Modeling, Search Algorithms, Game-Refinement Theory, Game Tree Search and Entertainment Science.

Funding information

The study was funded by the Framework for Grant-in-Aid for Challenging Exploratory Research (Grant No. 19K22893) from the Japanese Society for the Promotion of Science.

Leave a Comment