Playing games is natural
Playing games is something that comes natural to humans. We easily understand the rules and by playing against more experienced players we pick up the subtleties and overcome di±culties for a particular game. In contrast, learning a computer to play a game is a considerable more di±cult process. Especially when chance moves and partial observability are involved, as is the case for games like poker, games quickly become intractable.
An often used solution for this problem is to have a computer play according to some heuristics that are defined by human nowledge about a particular game. This essentially comes down to programs playing a set of predetermined rules. The major downside of this approach is that these type of programs have a very limited capability to adjust their play and, herefore, are beaten rather easily by human players or other program designed specially to counter the heuristics behind the rules.
In this thesis we will examine frameworks that give a fundamental basis
for games and are less vulnerable than rule-based programs based on human
expertise.
