AHA Session 58
German Historical Institute Washington 2
Central European History Society 2
German Historical Institute Washington 2
Central European History Society 2
Friday, January 9, 2026: 8:30 AM-10:00 AM
Continental A (Hilton Chicago, Lobby Level)
Chair:
Fabian Offert, University of California, Santa Barbara
Panel:
Jana Keck, German Historical Institute Washington
Todd Presner, University of California, Los Angeles
Fabian Offert, University of California, Santa Barbara
Joshua Sternfeld, independent scholar
Todd Presner, University of California, Los Angeles
Fabian Offert, University of California, Santa Barbara
Joshua Sternfeld, independent scholar
Session Abstract
AI models are increasingly intervening in the historical record – through generated images and texts, but even more so through modes of access. If the predominant mode of interacting with – and writing – the historical record, in the very near future, will be AI-mediated, we need a better grasp on AI’s forms of historiographical mediation. AI-mediated historiography must account for the convergence of traditional historical practice and epistemology with the algorithmic and probabilistic techniques that govern AI systems. Crucially, the main challenge of such an endeavor is methodological. It is a truism that AI systems are black boxes, but at the same time we are witnessing, on the one hand, a Cambrian explosion of models built upon variant (and frequently opaque) datasets, architectural approaches, training regimes, and inference techniques, and on the other hand an increasing proliferation of open-weight, open-source, and open-architecture models that allow for hands-on experimentation. It thus makes sense – at this exact historical moment – to interrogate specific and concrete implementations on aspects of historiographic work.
The proposed round table tackles the question of AI-mediated historiography from four different perspectives. Jana Keck’s contribution looks at generative image models in the context of museum work, questioning the role of speculative illustrations created with off-the-shelf AI tools. How much, and what kind, of curatorial contextualization – technical, ethical, and legal, do generated images require, especially if they are clearly speculative in nature, and contrasted with proper historical sources? Likewise, Fabian Offert and Thao Phan question the ethics of representation in contemporary uses of generative image models, focusing on the status of generated images in the documentation of political violence. What do models like Stable Diffusion “know” about political violence, they ask, and how is that knowledge aligned with human ideas about representations of the fundamentally unrepresentable? Joshua Sternfeld discusses the impact of “research” style models on historiography. Can models like Deep Research augment historiographic work, or does their tendency to hallucinate, combined with their enigmatic modes of aggregating and interpreting sources, foreclose any potential to facilitate our interaction with the historical record? Todd Presner, finally, discusses the way a group of Large Language Models (including GPT4-o, Llama-3, and Gemini) provide answers to questions about the Holocaust and proposes a methodology – grounded in the functioning of next token prediction – for analyzing the reliability, completeness, and accuracy of probabilistic, AI-generated history.
The four presentations will collectively address the shifting roles of the historian amid the rapid deployment of AI-mediated historiography across the commercial, academic, educational, and heritage sectors. More than ever, historians bear the responsibility for promoting the ethical stewardship and interpretation of the historical record grounded in the values of transparency, methodological rigor, tolerance, and curiosity.
The proposed round table tackles the question of AI-mediated historiography from four different perspectives. Jana Keck’s contribution looks at generative image models in the context of museum work, questioning the role of speculative illustrations created with off-the-shelf AI tools. How much, and what kind, of curatorial contextualization – technical, ethical, and legal, do generated images require, especially if they are clearly speculative in nature, and contrasted with proper historical sources? Likewise, Fabian Offert and Thao Phan question the ethics of representation in contemporary uses of generative image models, focusing on the status of generated images in the documentation of political violence. What do models like Stable Diffusion “know” about political violence, they ask, and how is that knowledge aligned with human ideas about representations of the fundamentally unrepresentable? Joshua Sternfeld discusses the impact of “research” style models on historiography. Can models like Deep Research augment historiographic work, or does their tendency to hallucinate, combined with their enigmatic modes of aggregating and interpreting sources, foreclose any potential to facilitate our interaction with the historical record? Todd Presner, finally, discusses the way a group of Large Language Models (including GPT4-o, Llama-3, and Gemini) provide answers to questions about the Holocaust and proposes a methodology – grounded in the functioning of next token prediction – for analyzing the reliability, completeness, and accuracy of probabilistic, AI-generated history.
The four presentations will collectively address the shifting roles of the historian amid the rapid deployment of AI-mediated historiography across the commercial, academic, educational, and heritage sectors. More than ever, historians bear the responsibility for promoting the ethical stewardship and interpretation of the historical record grounded in the values of transparency, methodological rigor, tolerance, and curiosity.
See more of: AHA Sessions