Learning From Mah Jong: Towards a Multi-Agent System that can Recognize Graphic Units

摘要

Sketching is a major means of exploiting the first conceptual developments in architectural design. If we want to support the architect in the ideas-developing phase of design, then we need to understand the conventions of depiction and encoding in drawings. The theory of graphic units provides an extended list of such conventions that are widely used. We propose that a multi-agent system for recognition of graphic units in drawings is fruitful: agents can specialize in graphic units, a multi-agent system can deal with ambiguity through negotiation and conflict resolution, and multi-agent systems function in dynamically changing environments. We first make a multi-agent system that can do something simpler: playing Mah Jong solitary. The Mah Jong solitary system shares the following important features with a multi-agent system that can recognize graphic units: (1) specialized agents for moves; (2) negotiation between agents to establish the best move; (3) dynamically changing environment; and (4) search activity in more advanced strategies. The paper presents the theoretical basis of graphic units and multi-agents systems. The multi-agent framework and its implementation is presented. Various levels of game play are distinguished, and these are correlated to the multi-agent system. The paper shows how the findings form the basis for graphic unit recognition.