ABSTRACT
A recent trend in computer-based learning has been to set up systems that the learner explores, rather than setting very specific goals to reach. Our previous research on complex problem solving has supported this approach (Vollmeyer, Burns, & Holyoak, 1996). When learning how to control a system with a set of inputs linked to a set of outputs, participants learned more about the system when they were given a nonspecific goal rather than a specific goal. These results could be explained using Simon and Lea's (1974) dual-space framework (or that of Klahr & Dunbar, 1988, who extend this framework to scientific discovery) in which induction is seen as a search of instance space (i.e., examining states of the system), integrated with search of rule space (i.e., formulating and testing rules that might govern the system's behavior).
