A Cognitive Systems Framework for Creative Problem Solving

Abstract

This thesis provides a theoretical framework for a wide variety of types of cognitively-inspired creative problem solving. The framework (CreaCogs) is formalized and its various creative processes detailed. The framework is put to the test in a few computational implementations: a solver to the Remote Associates Test - comRAT-C, its adaptation to the visual domain - comRAT-V, and an object replacement and object composition system in a household domain - OROC. The performance and process of these implementations are then (i) compared to human answers and performance in creativity tests or (ii) assessed with the same toolkit that would be used to assess human answers. A set of practical insight problems with objects are given to human participants in a think aloud protocol, which is then encoded and compared to the framework. The experiments and data analysis show that the framework is successful in computationally modeling creative problem solving across a wide variety of tasks.