Brief: Massive collection of game AIs for all known games
Jump To: Parent Description

  • Initially, the KB component of this project applies existing ai knowledge formation tools to games like chess and go. We are working in tandem with both the CMU Chess Club and the Pittsburgh Go Association to produce ontologies and domain axioms for both of these games. Development progresses mainly in two branches - the systems integration of existing text-to-knowledge software, and the translation of knowledge by the SMEs (subject matter experts). The systems integration operation is rewarding because it improves our understanding of existing resources, and the encoding operation because it is a form of rigorous study. Indeed, the author of Deep Thought did not know chess when he began programming it, but at the end had become skilled. Looking at chess and go from a KBS perspective will likely enhance and refine our understanding of these games.

    Our approach is as follows - we know that if we ignore the considerable literature when writing chess (and go) programs, our programs will be at best idiot savants - unlearned yet skilled. In the case of chess this may be sufficient to beat the current human world champion, however it has yet to be the case for Go. As computers advance, perhaps Go will also succumb to this style of brute force attack. Yet, we can know for certain that we are not taking advantage of the capabilities of computer systems when we write such simple programs. We can write better programs which when matured can beat existing systems.

    Results from AIT (algorithmic information theory) show that there are more complex, sophisticated programs, programs which exhibit greater intelligence in the sense that they can recognize and react to a larger number of conditions. Being able to reason logically about game invariants is essential to improve the quality of play. Our programs may not offer early success but they are provably more efficient and capable.

    So in other words its a simple question of weight ratios! (Monty Python reference)

    The tools involved are extremely mature (20 years). If you are interested in learning how this is done, there is a complete reference available online. Please refer to: "http://www.opencyc.org/doc".

    These systems are of extreme practical importance in ai and this effort can be viewed as calibration and verification of these technologies.