Rules discovery: Transfer and generalization

This paper presents a study about human transfer of learning between isomorphs and on the conditions of rule discovery that could be of help for machine learning. When faced with a new problem, the human learner uses the knowledge s/he already possesses in order to mentally represent and manipulate the objects s/he has to deal within the process of problem solving. We propose that familiar domain knowledge provide concepts as useful biases for discovering general rules when solving isomorphic problems as well as problems which entail a larger problem space. Results of two experiments using isomorphic versions of the ldquorule discoveryrdquo Nim game, and versions that entail a larger problem space, suggest that participants use the even concept in the familiar domain according to external representations that can either predictably favor or impair learning and transfer in a foreseeable way.