AI-Assisted Hypothesis Generation: From Fragmented Manifestations to the Morphology of War Contamination

Abstract

War-related environmental contamination is a complex, long-term, and partially observable process. Researchers encounter manifestations, including damaged industrial facilities, ammunition disposal sites, abnormal soil concentrations, altered freshwater quality, unexploded ordnance, disturbed land, and delayed health effects. None reveals the complete process. Connections among sources, environmental transformation, transport, exposure, and consequences must therefore be reconstructed under uncertainty.

Hypotheses are necessary working instruments in this reconstruction. They temporarily connect manifestations, propose causal structures, and identify evidence expected if those structures are correct. They must remain inexpensive to formulate, comparable, revisable, and expendable. Traditional research often narrows hypotheses to satisfy disciplinary standards. Chemical, hydrological, and historical hypotheses gain methodological precision but may lose the capacity to represent relationships and development.

Scientific recognition creates a constraint. Publication, peer review, citation, institutional status, and paradigmatic compatibility are frequently conflated with validity. Yet silence, non-citation, or exclusion does not constitute logical refutation; recognition itself requires investigation.

The expert–AI system offers an alternative. The expert defines the object, evaluates plausibility, and retains responsibility, while AI compares disciplines, languages, cases, causal sequences, and explanations. A controlled portfolio of competing hypotheses can reveal the recurring morphology of long-term war contamination.