Fourth paradigm GIScience? Prospects for automated discovery and explanation from data

This article discusses the prospects for automated discovery of explanatory models directly from geospatial data. Rather than taking an approach based on machine learning, which generally leads to models that cannot be understood by humans or related to domain theory, the approach described here suggests we can instead construct models from fragments of domain understanding—such as commonly encountered equation forms, known constants and laws—resulting in discovered models that can both be understood by humans and directly compared with known theory. We then propose a conceptual model of the discovery process by which the various stages and components of discovery and explanation work together to learn models from data. The approach described weaves together ideas for describing models from Harvey's book 'Explanation in Geography' with current thinking on how explanatory models might be 'discovered' from data from Inductive Process modeling. On the way, we also highlight: (i) why it is important to have models that explain as well as predict, (ii) how such an approach contrasts with – and goes beyond – current work in deep learning, (iii) how the task of model discovery might be tackled computationally and (iv) how computational model discovery can play a valuable role in creating geographical explanations.