Proceedings of the Annual Meeting of the Cognitive Science Society

Humans have developed complex rule-based systems to explain and exploit the world around them. When a learner hasalready mastered a system’s core dynamicsidentifying its primitives and their interrelationsfurther learning can be effec-tively modeled as discovering useful compositions of these primitives. It nevertheless remains unclear how the dynamicsthemselves might initially be acquired. Composing primitives is no longer a viable strategy, as the primitives themselvesare what must be explained. To explore this problem, we introduce and assess a novel concept learning paradigm in whichparticipants use a two-alternative forced-choice task to learn an unfamiliar rule-based conceptual system: the MUI system(Hofstadter, 1980). We show that participants reliably learn this system given a few dozen examples of the systems rules,leaving open the mechanism by which novel conceptual systems are acquired but providing a useful paradigm for furtherstudy.