Artificial intelligence and human behaviour in simulations - final report from FFI-project 722 Synthetic decision making

Abstract

The report describes the results achieved in the FFI-project 722 on synthetic decision making. It is observed how formalised combat simulations fit the game-theoretic framework of two-player zero-sum games. The games of Campaign, Operation Lucid and Operation Opaque are described. These games have been used for experiments with human decision making and artificial intelligence (AI). The human experiments indicate that the game-theoretic concept of randomisation in games of imperfect information fails to explain human decision making. Several AI techniques were utilised in the development of automatic agents playing our three games, including rule-based systems, fuzzy logic, genetic programming, neural nets and constraint satisfaction programming. The most significant contribution from the project was the development of reinforcement learning (including co-evolution) algorithms for games of imperfect information.