Machine Generalization and Human Categorization: An Information-Theoretic View (original) (raw)

This paper explores how insights from psychological studies on categorization can enhance the development of intelligent systems that effectively communicate generalizations to human users. It introduces the concept of "Category Utility," which quantitatively predicts the success of various categorization tasks better than traditional metrics by measuring the informational value of categories. The findings suggest that fostering an understanding of the psychological preferences for category levels, particularly the basic level, allows for a more intuitive human-machine interaction by aligning machine generalization with human categorizational abilities.