Data and coherence theories of truth: Examples from a data-driven Geographical Information Science

The validity of information collections can be verified by their coherence, such as in the case of Volunteered Geographic Information. However, corresponding coherence theories of truth do not readily apply to collections of data if these consist of non-interpreted or virtually non-interpretable symbols, as is often the case with machine learning models and other black box systems. This paper argues why data-driven geography requires coherence theories, to then transfer the concept of coherence theories from the information to the data level. Finally, the relevant implications on the interpretation of data, especially in the context of black box systems and machine learning, are discussed.