Event Report: Beyond the Hype: Scholars Explore AI Use in Historical and Qualitative Research
On May 13th, 2026, a group of international historians, data scientists, and sociologists coalesced at the Munk School of Global Affairs and Public Policy.
The Data Sciences Institute and the Centre for the Study of the United States organized this day-long symposium in response to the rapidly changing field of LLMs (large language models) in research and the resulting hole in academic literature: how do we ethically approach AI use in a qualitative and historical academic setting?
As a group, the panel emphasized moving beyond the extreme polarization of AI as inherently good vs. inherently bad, instead advocating for a nuanced position, acknowledging the harms of privatized LLM corporations, while maintaining that LLMs with clear off-ramps, transparency, and reproducibility can assist in data collection.
As Ian Milligan, University of Waterloo Historian, remarked, “there is no opting out of this, there is only paying attention.”
The scholars also made clear that AI should not and does not have space in the interpretation of research. At the Campbell Conference Centre, a sigh of relief rippled through the audience when the scholars positioned the critical thinking and writing skills of historical and qualitative researchers not only as pertinent, but as a task that AI failed to replace.
S. Wright Kennedy made a salient case for becoming the LLM domain expert of your choosing. He made clear that technology companies operate on a different playing field and are not beholden to academic rigour, but rather their bottom line. The implication of such rapid development means that what gets built and what gets kept alive is always shifting. Kennedy, along with other speakers, noted how, in some instances, the LLM they were incorporating into their research became defunct in a matter of months, and with it, the data collected. Kennedy shared pathways to ensure trust, namely, auditing of AI-assisted outputs in software where the researcher is in control, rather than the AI.
Sergio Gabiel Petralia from Utrecht University used his knowledge of coding to position LLMs not as a story of replacement, but of learning new skills. While he acknowledged the challenges of diving into programming, he insisted that there is space for humans in this loop, asking the audience to make mistakes, “break something”, to advance capabilities for researchers across the world. His emphasis on creation with tools or infrastructure that are available to researchers regardless of monetary funding highlights the accessibility potential of LLMs in research.
Milligan delivered a presentation on the choices scholars face when they engage with or don’t engage with LLMs in their research. The title of his talk was “My Robot Research Assistant Occasionally Lies — And Soon the Archives Will Too.” Milligan stressed the importance of holding nuance in the way researchers approach AI in research, speaking plainly and directly of the concern over “hallucinations” while describing how to methodically incorporate AI into legible, trustworthy workflows with offramps.
Word embedding, a core process of LLMs, was at the centre of Alina Arseniev-Koehler’s presentation. The Purdue University researcher shared how she used the process of converting words into something tangible, mappable, which in turn could capture the cultural framing of diseases in the United States over time. Using these tools, Arseniev-Koehler saw how disease as criminalization vs medicalization gets imposed by cultural forces, with implications for further qualitative studies.
Yale University Professor Daniel Karell spoke of incorporating Synthetic Simulation into research practices. To Karell, this looked like using AI to create data that researchers don’t have but want. The synthetic environment of AI becomes a new space in which to probe synthetic political ideology as produced by AI chatbots. From human engagement with AI models, information about the biases of AI systems, their holes, and hallucinations becomes better understood.
Ethan Fosse, the Co-director of the Data Sciences Institute, gave a rousing summation of the day's takeaways.
Placing the history of AI as uniquely a result of Toronto-based innovation, Fosse was clear that “the objective is augmentation, not automation, namely, using computational tools to extend, rather than replace, the interpretive depth of traditional qualitative and historical methods.”
The scholars asserted their claims as one of curiosity, how should academic researchers engage in research, while pushing the envelope of the technology that we have access to?
The future is uncertain: together, we must set vigilant transdisciplinary standards through conversations like these while preserving critical human oversight in methods of research and knowledge production.
