"Establishing Rigorous Digital and AI Methodologies for Business History Research"

The increased sophistication of AI and digital tools and the growth of digital databases of value for business history research necessitate an urgent discussion of which tools and methods are suitable for use by modern researchers. This paper discusses challenges and promises with such techniques and tools.
The rigorous application of digital and AI tools will allow for a larger and more sophisticated research field, one which can organize and process larger amounts of source material, more clearly trace and showcase how narratives derive from source material, and enable comparative studies and collegial review. The unrigorous or unprincipled application of AI technologies will instead undermine the desired dual veracity, extract data from datasets on an unprincipled and inconsistent basis, obfuscate historical source analysis, and impede or preclude reproducibility of empirical claims, the antithesis of methodological principles outlined by Harvey & Maclean (2023).
Of key importance for the researcher wishing to increase their productivity or to push the frontiers of research without compromising scientific integrity is developing an understanding of the differences inherent in tools that are either deterministic (regular expression analysis, indexed searches, OCR) or probabilistic (generative AI, large language models such as chatGPT, supervised and unsupervised machine learning on datasets). Deterministic tools and their transparent application will help a researcher more clearly showcase the application of methodologies, establish claims of causal inference, and set requirements for falsification. Careful application of probabilistic tools helps the researcher create hypotheses, find promising areas of inquiry, and increase productivity in semi-skilled aspects of dataset construction. In contrast, a naïve and general application of them engenders precipitous risks for the research field, as it would create impermeable methodologies.