Abstract

The three basic kinds of theory development are expansion, contraction and revision by empirical evidence. Under empiricist assumptions, the history of scientific evidence can be represented by a sequence of true and cumulatively increasing evidence sets which in the limit determine the complete structure of the world. Under these assumptions it turns out that purely universal hypotheses are falsifiable with certainty, but verifiable only in the limit, ∀-∃-hypotheses are falsifiable in the limit but not verifiable in the limit, and ∀-∃-∀-hypotheses are neither nor. In the consequence, hypotheses with complex quantification structure can only be confirmed probabilistically. While these results are based on “empiricist” assumptions, the revision of theories which contain theoretical concepts requires either a given partition of possible hypotheses out of which the most promising one is chosen, or it requires steps of abductive belief revision. Revision of scientific theories is based on a Lakatosian preference structure, following the idea that in case of a conflict between theory and data, only peripherical parts of the theory are revised, while the theory’s core is saved from revision as long as possible. Surprisingly, the revision of a false theory by true empirical evidence does not necessarily increase the theory’s truthlikeness. Moreover, increase in empirical adequacy does not necessarily indicate progress in theoretical truthlikeness; a well-known attempt to justify this inference is Putnam’s no-miracles argument.