Simulation of the hypothesis generation process

Abstract

Can the process by which scientific hypotheses are developed be made explicit? This paper investigates the issue by assuming hypotheses come from data, utilizing an information processing framework, and setting up a laboratory simulation in which subjects infer mathematical functions from plots of coordinate values. Protocol information collected in this setting provide the basis for a computer model of the sequence leading from the data to hypotheses. The model has five steps including the identification of the data's significant features and their organization into a coherent pattern, inferring of a general class of hypotheses, determination of a specific hypothesis from the class, testing by means of comparing hypotheses' predictions with the data, and iteration if necessary. Tests of the model indicate it gets the same answer as subjects most of the time and follows some of the process used by subjects in obtaining answers.