Cognitive modelling of the human Go player.
In order to understand human cognitive system, it is fundamental to uncover
intuitive knowledge that human beings use in the real world.
The game of Go, being a simplification of the real world
that keeps its spatial and temporal features, is a well-adapted domain.
The computer helps to explicit knowledge that is supposed to be mainly not
conscious. This hypothesis would explain the amazing difference between
man and machine in Go. We made usefull links between the game
of Go and the neighbouring domains of AI and mathematics: Distributed AI,
mathematical morphology, game theory, metaknowledge and fuzzy logic.
Computational model is object oriented, structured in levels.
It uses primarly Conway's games and involves approximately a half-thousand
rules. It includes an algorithm which recognizes statically
life and death of groups based on interaction with the neighbourhood.
It uses morphology tools to recognize territory in the manner of good
human players. It incrementally updates objects. Global decision uses
size and instability of objects as two main criterias. In building this model,
we revealed concepts we linked with human knowledge :
grouping, inside and outside, interaction, morphology. It reinforced us in our work,
based on the game of Go and the machine, to reveal intuitive knowledge.
The program MOOD INDIGO (My Object-Oriented Design Is Now Designed In Good
Objects), developped with C++, plays complete games. It plays a move between five
seconds and one minute on a Sunsparcstation. Its strength is evaluated between
15th and 20th kyu. We tested it against others Go programs on IGS as well as
against human players.
Cognitive and computational models, implicit knowledge, game of Go, Conway's
games, interaction, mathematical morphology, incrementality, objects.