DIRECT CONNECTIONISTIC METHODS FOR SCIENTIFIC THEORY FORMATION

Abstract

Thagard’s theory of explanatory coherence (TEC) is a conceptual and computational framework that is used to show how new scientific theories can be judged to be superior to previous ones. In Structures in Science (SiS), Kuipers criticizes TEC as a model that does not faithfully reflect scientific practice. This article tries to explain the machinery behind TEC, and tries to indicate where TEC falls short (conceptually speaking) and where it can be improved. The main idea proposed in this article is not to derive a coherence network from the input (a la TEC), but to construct a coherence network right from the input itself.