Computer Creativity in Games—How Much Knowledge Is Needed?

Abstract

In seeking to make computer systems that can be creative, there are several problems, including the definition of the concept of creativity, understanding how to design creative algorithms, encoding the domain knowledge needed for them, and evaluating the creative products of those algorithms. We make a case that games are in some ways a good domain in which to research and develop creative algorithms, in particular the design of computer programs that can play games in ways that human players would consider to be creative. Although the playing of games is not in itself a very important research aim, we argue that it is a good arena in which to test out attempts to make creative programs. We initially consider how the field of computer play may be said to have contributed to computational creativity so far, with games like chess that require a specialised knowledge base in order to perform well. As an example, we show the play of our program for a falling-blocks puzzle game, that uses simple search to play to a fairly good standard, and which a human player may well consider to be creative at times. In that case, we conclude that at least in some non-trivial domains, computational creativity might be achieved without the typically heavy requirement for a large or sophisticated knowledge-base or even machine learning.