The core of scientific theories are laws. These laws often make use of theoretical terms, linguistic entities which do not directly refer to observables. There is therefore no direct way of determining which theoretical assertions are true. This suggests that multiple theories may exist which are incompatible with each other but compatible with all possible observations. Since such theories make the same empirical claims, empirical tests cannot be used to differentiate or rank such theories. One property that has been suggested for evaluating rival theories is coherence. This was only understood qualitatively until we [Kwok, et.al. 98] introduced a coherence measure based on the average use of formulas in support sets for observations. The idea was to identify highly coherent theories with those whose formulas that are tightly coupled to account for observations, while low coherence theories contain many disjointed and isolated statements. Our current approach generalizes that insight to accommodate fundamental ideas from the philosophy of science and better mirrors scientific practice. Moreover, this new approach is neutral with respect to the philosophy and practice of science, and is able to explain notions like modularization using coherence.
|Number of pages||2|
|Journal||IJCAI International Joint Conference on Artificial Intelligence|
|Publication status||Published - 2003|