In the proposed project we aim at developing pattern-based machine-learning methods that can be applied to build a strong Go-playing system. In Go, patterns are the guidelines for ordering moves, or otherwise stated they classify the possible moves by their expected result. Since it is unlikely that a Go program using solely a pattern evaluator will yield good results, a shallow search has to be added. The main emphasis, however, will be on knowledge acquisition, representation and prioritisation from the patterns. Based on an ontology of the notions and concepts of the game, a hierarchy of concepts will be built with the help of the patterns. Our research question reads: Is it possible to generate Go-specific information from a given board position which enables a program to build a relevant connected network of strategic concepts, so that a Go player or Go program can make the right decision? The initial application focus will be on complex adaptive artificial neural network architectures. By combining reinforcement learning, back-propagation and evolutionary techniques we will be able to develop a system capable of both unsupervised learning from active play as well as supervised learning from expert games.