Categorization, Concept Learning, and Problem Solving: A Unifying View

Douglas Fisher
Department of Computer Science
Vanderbilt University
Nashville, TN 37235
Jungsoon Yoo
Department of Computer Science
Middle Tennessee State University
Murfreesboro, TN 37132

In G. Nakamura, R. Taraban, and D. Medin (Eds). The Psychology of Learning and Motivation, Vol. 29, San Diego, CA: Academic Press.

ABSTRACT: This paper illustrates the synergism between categorization and problem solving, and between artificial intelligence and cognitive psychology, with two computational models. The first, COBWEB, is a learning system for hierarchical categorization that provides good fits to experimental data on basic level, typicality, and fan effects. The second model, EXOR, learns to solve problems more effectively by extending the basic COBWEB learning and categorization strategies in a direction that is cognizant of background knowledge or preconceptions on the part of the learner. Our adaptation illustrates the importance of categorization in problem solving. It also promotes a unique perspective that unifies fan effects with typicality and basic-level phenomena, and identifies a phenomenon, basic levels of problem solving, which appears to be novel in the literature and speaks to issues of learning and training of problem-solving skills in humans and machines.

Our methodological viewpoint agrees with Anderson's ideas of rational analysis: if humans are bounded-rational agents, then a reasonable starting point for modeling their behavior is an objective function that describes ideal behavior conditioned on resource (e.g., memory) constraints. This view is consistent with traditional research at the interface of artificial intelligence and cognitive psychology, but we stress throughout that to be maximally useful, computational models should move beyond the known experimental data, thus providing guidance for further psychological study.

Related article: Fisher, D. and Langley, P. (1993). "The structure and formation of natural categories," The Psychology of Learning and Motivation, Vol. 25, G. Bower (Ed.), San Diego, CA: Academic Press.

Related article: Yoo, J. and Fisher, D. (1991). "Concept Formation over Explanations and Problem-Solving Experiences," International Joint Conference on Artificial Intelligence, Sydney, Australia: Morgan Kaufmann, pp. 630-636.

Related article: Yoo, J. and Fisher, D. (1991). "Concept Formation over Problem-Solving Experiences," In D. Fisher, M. Pazzani, and P. Langley (Eds.), Concept Formation: Knowledge and Experience in unsupervised Learning: Morgan Kaufmann.