A Unified Cognitive Framework for Artificial Creativity: Modeling Divergent Thinking and Conceptual Blending

Abstract

Artificial creativity has been a difficult research issue because of the inability to reduce the high-level cognitive theories to computationally testable models. This paper hypothesizes a universal cognitive approach to artificial creativity combining ideas of divergent thinking, conscious/unconscious processing, and blending of concepts at a layered computational architecture. The framework is inspired by the literature on cognitive and neuroscience and separates exploratory (divergent) and evaluative (convergent) processes and models these processes in terms of semantic expansion, concept blending based on similarity and concept convergence filtering. A proof-of-concept implementation of the method is provided using large-scale textual knowledge in the form of the Wikipedia encyclopedia and classical semantic representations such as TF-IDF, Latent Semantic Analysis, and Explicit Semantic Analysis to show that it is feasible. The outputs depict how the intended architecture is capable of producing and optimizing new combinations of concepts in a directed and understandable way. Instead of proposing complete validation of creativity, this contribution creates a rigorous computational basis upon which cognitive theory and empirically testable models are connected, which can serve as the foundation of future empirical testing, multimodal extrapolation, and human-in-the-loop creativity testing.