Theory Before Tools in AI Humanities Research

_akpiper · x · 2026-07-17

This article proposes a "theory-first, AI-second" approach for humanities research. While generative AI can expand the space of interpretation, lacking a theoretical framework turns this openness into arbitrariness.

The core argument is that AI is not an "alternative researcher" that automatically provides answers, but rather a measurement/analysis tool. Therefore, before use, researchers must clarify: the object of study, the unit of analysis, whether the model serves as the "object" or the "tool," and the specific concept being validated.

The article further emphasizes that in humanities and interpretive research, disagreement itself might be part of the phenomenon and shouldn't simply be eliminated as noise. Evaluations should differentiate between testing feasibility, consensus, or distribution fitting, rather than just checking if the output is fluent.

Original post →

More from Research

Research channel →