This thesis investigates how search can be guided by knowledge in such a way that the search space is traversed efficiently and effectively. For this task we focus on the question how to combine knowledge with search. The more adequate the knowledge is, the better the search. If a search process is sufficiently endowed with knowledge and as a consequence the search is (rather) successful, we call the search process an informed-search process. In this chapter we provide a brief introduction on games and Artificial Intelligence (AI) (Section 1.1) and then discuss the notion of informed search (Section 1.2). Subsequently, we formulate our problem statement together with four research questions (Section 1.3). Section 1.4 provides a thesis overview.
Games and AIEver since humans achieved some degree of civilisation, they played games. The two most important reasons for games to be played are their intellectual challenge and their entertainment. For the first reason games are used as a testing ground for computational intelligence. Since the 1950s the AI community compares the computer performance with the human performance (Van den Herik and Iida, 2000), or otherwise stated: since the birth of AI computational intelligence is measured with respect to human intelligence. Shannon (1950) and Turing (1953) were the first to describe a chess-playing program, while Samuel (1959) wrote the first gameplaying program in the domain of Checkers. In the beginning most AI research in games concentrated on abstract games like Chess and Checkers. Later on (in the 1970s) Backgammon and Bridge were added to this list, in particular since they possessed additional features, viz. stochastic information and incomplete information, respectively. All four games offer a pure abstract competition, with an exact closed domain (i.e., well-defined rules). The game state is easy to represent and the possible actions are known. Less abstract games like football, of which the domain is not properly described and the rules are more vague, did not attract any attention in the beginning of AI.Since the 1950s there has been a steady improvement in the strength of gameplaying programs. The quality of these programs can be roughly categorised into five classes (Allis, Van den Herik, and Herschberg, 1991a).Research question 4: How can we use information gained during the search to improve move ordering?Move ordering (directing knowledge) is one of the main techniques to decrease the size of the αβ search tree. There exist several move-ordering techniques, which can be