Geographia Technica, Vol 21(2), Special Issue: Artificial Intelligence Applications in Geography, 2026, pp. i-xiii
FROM POSTFACTUM TO ANTEFACTUM GEOGRAPHY: WHAT ARTIFICIAL INTELLIGENCE WOULD HAVE CHANGED IN SEVEN EARLIER STUDIES
A Retrospective Human–AI Experiment in Technical Geography
Ionel HAIDU 
ABSTRACT: Discussions of artificial intelligence in Geography frequently take the form of inventories of possible applications. This Letter adopts a different strategy. Instead of asking what AI might theoretically do, it asks what contemporary AI would have changed in seven geographical studies previously authored or co-authored by the editor and completed without access to present-day AI technologies. The studies concern forest-cover change, storm-induced windthrows, floodplain delineation, urban imperviousness and runoff, trend-signal detection, road-accident distribution, and atmospheric pollution during the COVID-19 lockdown. Each paper is examined through the same retrospective protocol: the original geographical question and technical workflow are reconstructed; possible AI entry points are identified; the potential gains in speed, accuracy, uncertainty assessment, knowledge discovery, prediction and decision support are evaluated; and the data and validation requirements of the proposed redesign are discussed. The exercise does not claim that AI would necessarily have produced superior empirical results, because the original datasets were not reprocessed. It demonstrates, however, that AI could have enlarged the range of questions that could be formulated and tested. The seven cases show a common transition from fixed workflows to adaptive learning, from deterministic maps to probability fields, from separate datasets to multimodal integration, and from postfactum explanation to antefactum prediction, counterfactual analysis and intervention design. AI is therefore proposed not as a substitute for geographical reasoning, but as the learning layer of Technical Geography.
Keywords: GeoAI; spatial reasoning; hybrid modelling; uncertainty; counterfactual analysis; decision support; human–machine collaboration
Full article here