A computational model of backward reasoning in human problem solving

When solving a problem with a clear goal, people often break down the problem into subgoals, a form of reasoning known as backward reasoning. The cognitive mechanisms of backward reasoning are poorly understood. Inspired by classic ideas from Newell and Simon, we conceptualize backward reasoning as search through a space of subgoals and actions to attain them. We show that this space can be represented by an AND-OR tree, a structure first used in early artificial intelligence research to automate problem solving. We hypothesize that, when solving problems with a clear goal, people reason backward by searching through AND-OR trees. To test this hypothesis, we analyzed data from two problem solving tasks: Tower of London and Rush Hour. We developed a psychologically plausible, single-parameter computational process model, which we fit to this data on a per-participant basis. We found that our model successfully captured key trends in the data, unlike alternative models. Our results elucidate the underlying cognitive mechanisms of human problem solving and open the door to a precise characterization of individual differences in problem solving.