Instance vs. Rule Based Learning in Controlling a Dynamic System

The question of whether human behavior could be better explained by assuming that people are capable of extracting from their experience some general principles (rules) or by supposing that they store in memory concrete, individual exemplars (instances) of the situations they deal with was examined in 2 experiments, adopting the Sugar Factory dynamic system control task, that contrasted the predictions of the computational model by Dienes and Fahey (1995) with those deriving from the ACT-R based model developed by Dieter Wallach and coworkers (Lebiere, Wallach, & Taatgen, 1998; Taatgen & Wallach, 2002). The first experiment produced findings that could not be explained by the Dienes & Fahey’s model while being consistent with the model of Wallach. The second experiment, however, obtained results that were at odds with the predictions of the latter. A new model is presented that is able to account for the results of both experiments by assuming that participants improve their performance in the Sugar Factory task by choosing, among a pool of very simple solution strategies, those that are judged increasingly more promising by the ACT-R procedural learning