Digital History
Jo Guldi, Professor of Quantitative Theory and Methods at Emory University
Jo Guldi’s research outlines how Digital History merges historical perspectives with data science, enhancing the analysis of both long-term and immediate changes. This approach not only prevents trivial or mismatched research in data science but also broadens historical analysis to include significant global issues, thus informing more accurate AI developments.
“Because of its emphasis on testing the relationship between evidence and argument, digital history’s insights are crucial for developing trustworthy AI workflows.”
I. Digital history as the merger of humanities and science
In Guldi’s research, Digital History emerges from the application of historical concepts from the philosophy of history to statistical and machine learning methods developed across campus. Fusing these disciplines allows historians to navigate between long-term change over time – covering decades and centuries – and the dynamics of short-term change, for example, identifying exceptional days or weeks on which events and ideas change and new concepts are introduced.
In the absence of the guiding insights of history, Guldi shows, papers by data scientists and computer scientists frequently take on either matters so trivial as to be unremarkable or a mismatch between evidence and argument, leading to retracted papers. In the absence of insights from statistics and CS, meanwhile, History has tended to gravitate to smaller and smaller timescales of analysis, with a resulting concentration on small, human-level stories from remote archives – a “microhistory” that can be hard to connect to long-term, planetary issues of change.
II. Digital history’s impact on future for human wellbeing
Because it takes on issues of political and cultural change, the practice of digital history is crucial for “auditing” institutions and individuals, for example, in telling us which newspapers or congresspeople have systematically avoided discussions of climate over the past decade. In the case of the publishers of the modern American novel, digital history has helped us to systematically inspect the relationship between publishers and systemic sexism and racism. These same techniques can be applied to institutions like church, courts, corporations, individuals, and patterns of
Because of its emphasis on testing the relationship between evidence and argument, digital history’s insights are crucial for developing trustworthy AI workflows, where LLMs, for instance, present answers adequately grounded in primary sources and are trained to discuss the relationship between evidence and fact. In the absence of historians’ involvements in the design of AI, we get pictures of African-American Nazis and lawsuits thrown out of court because the footnoted citations point to nonexistent case law.
Further reading
- The Dangerous Art of Text Mining (Cambridge University Press, 2023)
- Jo Guldi, “Text Mining for Historical Analysis,” American Historical Review (accepted & forthcoming)
- Jo Guldi, “The Algorithm: Long-term trends and short-term change,” American Historical Review 127:2 (June 2022): 895-911 https://doi.org/10.1093/ahr/rhac160
- Jo Guldi, “The Climate Emergency Demands a New Kind of History,” Isis 113:2 (June 2022) https://www.journals.uchicago.edu/doi/abs/10.1086/719704
- Jo Guldi, “What Kind of Information Does the Era of Climate Change Require?,” Climatic Change 169, no. 1–2 (November 2021): 1-20, https://doi.org/10.1007/s10584-021-03243-5.