by Soren Johnson
Feel free to download my adaptive turn-based strategy game... Share it with your friends!
Play the game (for Win95/98): Advanced_Protection.exe (164 K)
Read the project description (in Microsoft Word format): Advanced_Protection.doc (438 K)
Read the rules: Rules.txt (9 K)
Turn-based strategy games have been a popular form of computer entertainment over the last two decades, from early wargames like "Lords of Conquest" to contemporary civilization-building games like "Alpha Centauri." In nearly all of the programs, the human player is able to match strategies with a computerized opponent, controlled by some form of AI. However, because these AI's were relatively static, the games often included multiple "difficulty levels" to challenge the users. Usually, each level had different parameters and rules to allow the computer's AI to perform better- to match the human's experience and ability.
While these methods did allow the user manually to choose between levels of competition, a better alternative might be for the system to adapt its AI automatically to match the user's ability and strategy. In other words, if the system had a limited search space of strategies, they could all be tested against the user's moves. Then, the program could determine which strategies worked best against the user and which strategies worked worst. This approach would provide three distinct advantages over systems with difficulty levels.
First, the system could dynamically switch between strategies depending on how well or poorly the user has been performing- a struggling user would be faced with weakly-performing strategies while an expert could be faced with the strategies which best countered his or her moves. This method would require a system which could rate the users' abilities, to distinguish more successful users from less successful ones. The second advantage of this system would be that the game does not need to change its rules to allow the computer's AI to perform better. The game will always function the same- only the computer's strategy would change. In other words, the computer will not "cheat" through rules and parameters unique to higher levels. The final advantage of this system is that the AI would truly know which strategies are most effective (and which are least effective) against each specific user because the strategies will have been tested against the user's previous moves. Games with difficulty levels are inferior in this regard because although the higher levels will probably challenge the user to a greater degree, the program has no way of guaranteeing that fact.
Thus, my project involves constructing a game which allows such an adaptive AI to be developed. The game, "Advanced Protection," was carefully designed to allow for such an adaptive AI system to be created. In other words, the game has only a few decision points for the user, so that the system has a limited amount of data on which to train. This fact is important because the AI needs to switch quickly between strategies- few users will enjoy a game which takes hours, minutes, or even many seconds to train between turns.