Recognizing Emergent Subshapes in Design Problem Solving: a Connectionist Investigation

Human problem-solving behaviour has been modelled as a search through the space as defined as problem states, within which earlier states move to subsequent ones by applying rules in the human mind until the goal state is found. This cognitive model of problem-solving has been broadly accepted and has become dominant in both cognitive psychology and artificial intelligence (AI). In the field of computeraided architectural design (CAAD), search models are also widely used for solving design problems, although various foci of design knowledge are differently represented by shape grammars, graphs, and knowledge-based systems using predicate logic for different purports. In design search, design evolves from one state to another by exhaustively or heuristically applying proper rules. Each rule application involves, first, pattern-matching the antecedent of a rule to the current state and, second, transforming the matched portion of that state into the consequence of the rule. However pattern-matching techniques of current CAAD systems are still limited. In current CAAD systems, only those two squares can be dealt with by patternmatching for further development. However, a human designer can effortlessly recognize not only those two but other emergent subshapes, for example a smaller square in the middle where the two squares overlap and two L-shapes in the corners. Therefore a human designer can thoroughly deliberate all these alternatives before making a decision. In other words, human designer is capable of restructuring shapes in terms of emergent subshapes in any step of designing