AI through History, History through AI

Registration

Conference opening

Andreas Fickers, Frédéric Clavert & Sarah Oberbichler (C²DH, University of Luxembourg), Daniel Burckhardt (GHI Washington)

Keynote I

Chair: Sarah Oberbichler (C²DH, University of Luxembourg)

AI Literacy and the (new) skills historians need

Anna Neovesky (Universität Erfurt; Fachhochschule Erfurt)

From text recognition to source analysis and writing, AI is already embedded in historians’ workflows. But which skills are necessary to use its scholarly potential? This talk develops a practice-oriented understanding of AI literacy, situating it in the context of digital hermeneutics and the research process and activities.

Coffee break

Panel I: Knowledge Representation & Semantic Analysis in Historical Research

Chair: Caio Mello (C²DH, University of Luxembourg)

Augmenting Historical Research through AI: A Case Study on Norcia (Italy) 1859

Sara Alimenti, Regina Lupi, Giuseppe Liotta, Carla Binucci, Emilio di Giacomo, Giulio Biondi, Fabrizio Grosso, Marco Legittimo, Paola Venuti (University of Perugia)

The talk introduces a knowledge graph transformation of a newspaper corpus using AI techniques, enabling quantitative network analysis to identify patterns in heritage preservation debates during Italy’s transition from Papal States to Kingdom of Italy and the choices made following an earthquake.

Constructing Semantically Coherent and Interpretable Historical Vocabularies from Domain-Specific Corpora

Arlinde C.E. Vrooman (Huygens Institute, Royal Netherlands Academy of Arts and Sciences)

Vrooman demonstrates an inductive methodological approach to building the Dutch Historical Climate Vocabulary from VOC archives, balancing semantic coverage with historical interpretability through embedding-based clustering.

Embed and Seek: Tracking Quotations Across Sources

Jeri Wieringa, Bennett Nagtegaal, Rebecca Sutton Koeser, Laure Thompson, Hao Tan, Mary Naydan, Edward Baring (Princeton University)

The presentation introduces remarx, a quotation detection tool using multilingual embeddings to identify semantic similarities across language boundaries, enabling tracking of ideas across multilingual historical sources.

Lunch break

Panel II: Decolonizing AI and Marginalized Archives

Chair: Ferdaous Affan (C²DH, University of Luxembourg)

Adapting AI for Marginalised Archives: Developing Handwritten Text Recognition Models for Colonial Paraguayan Manuscripts

Guillaume Candela (Cardiff University)

Candela examines specialized HTR models for colonial Spanish-Guaraní documents that remain computationally inaccessible through existing “off-the-shelf” AI models, addressing epistemological erasure of Indigenous and Afro-Paraguayan histories.

Beyond ‘Dirty OCR’: Evaluating VLM-Driven Re-OCR Pipelines for Legacy Digital Collections in Latin America

Jairo Antonio Melo Flórez (University of California, Santa Barbara Library)

The talk addresses the “Problem of Done” with legacy digitizations from institutions like Hemeroteca Nacional de México, evaluating Vision-Language Models for democratizing OCR improvement while navigating epistemic risks of “hallucinations”.

Beyond Keywords: AI-Mediated Access to the Islam West Africa Collection through an MCP Server and Agent Skill2

Frédérick Madore (University of Bayreuth)

Madore presents two complementary approaches to overcome multilingual barriers in West African Islamic collections, offering an ethical framework that preserves institutional data sovereignty while enabling AI-enhanced access.

Panel III: Ethics, Epistemology & Critical AI Practice

Chair: Ellen Charlesworth (C²DH, University of Luxembourg)

Machine-Summarised Scholarship and its Limits: H-Soz-Kult as NotebookLM Podcasts

Aurel Daugs (Humboldt-Universität zu Berlin), Thomas Meyer (Clio-online e.V), Claudia Prinz (NFDI4memory, Humboldt-Universität zu Berlin)

The presentation evaluates AI-generated scholarly podcasts through four dimensions: content agreement, hallucinations, semantic shifts, and systematic biases, questioning where human editorial judgment remains indispensable.

A Sensitive Approach to Web Archives through AI: The Skyblog Experiment2

Emmanuelle Bermès (École nationale des chartes), Axel Chemla-Romeu-Santos (Université Paris), Marina Hervieu (Bibliothèque nationale de France)

The talk highlights generative AI as a creative tool for web archive exploration, using the “uncanny valley2 effect to reveal text message language patterns and transform fiction into an epistemological compass.

When to Generate? Comparing Computational Methods for Archival Inquiry

Samuel Backer (University of Maine)

Backer compares RAG-based archival exploration and structured LLM annotation through Baltimore Railroad and Early Hollywood case studies, arguing computational method selection must align with historical goals (flexibility vs. reproducibility) rather than universal “accuracy” standards, positioning AI as site-specific scholarly tools.

Break

Panel IV: History, Speculation, and Law

Chair: Torsten Hiltmann (Humboldt-Universität zu Berlin)

History, Speculation, and the Stochastic Parrot: A New Methodological Approach

Lukas P. Stuber, Phillip B. Ströbel, Felix K. Maier (University of Zurich)

This presentation proposes transforming speculative historical reasoning into a systematic method by analyzing distributions of variation rather than debating singular alternatives, making patterns of historical imagination observable and comparable.

History between Technology, Progress and Legality: Transparency of AI Use under the EU AI Act

Urban Makorič (Milko Kos Historical Institute, Ljubljana)

Makorič examines how the EU AI Act’s transparency provisions, designed for commercial contexts, align with historical epistemology’s demands for traceability, interpretability, and reproducibility in historiographical work.

Keynote II

Chair: Andreas Fickers (C²DH, University of Luxembourg)

Uncritical Fabulation: Historical Emplotment and Narrative Theory after the Transformer

Todd Samuel Presner (University of California)

While historical narratives have always been shaped by discursive formations, archival technologies, and inherited representations, Large Language Models generate narratives from a vast probabilistic space in which statistical frequency and linguistic fluency function as proxies for truth and understanding. In contrast to conventional historical narratives that reflect epistemological, aesthetic, or ethical choices, LLMs produce narratives through “uncritical fabulation,” a predictive process in which statistically likely patterns are derived from what has already been ‘said’ and ‘represented.’ This talk historicizes pattern-based reasoning and argues that we need a new approach to historiography and narrative theory in light of the transformer.

Break

Conference dinner

Panel V: Infrastructure, Workflow & Hermeneutic AI

Chair: Tugce Karatas (C²DH, University of Luxembourg)

AI as Infrastructure – Building a Reproducible Research Data Cycle for Historical Sources

Clemens Beck, Daniel Motz (Friedrich Schiller University Jena)

Demonstration of the HisQu project infrastructure, showing how context-sensitive LLMs combined with symbolic modeling techniques create a closed research data cycle for historical sources. Participants will see the iterative human-in-the-loop workflow, the RAG architecture with multiple AI agents, and the graphical “lab notebook” that documents each modeling and processing step for transparency and reproducibility.

Augmented Hermeneutics: An LLM-Based Evaluative Workflow for Semantic Annotation in Historical Texts

Torsten Hiltmann (Chair of Digital History, Humboldt-Universität zu Berlin; NFDI4memory)

Hiltmann presents a three-step workflow integrating LLMs into historical scholarship while preserving epistemic agency: (1) rigorous evaluation through ground truth construction, (2) epistemic infrastructure design principles for reasoned arguments, and (3) hermeneutic circle refinement. Using medieval hagiography as a case study, it demonstrates how LLMs can reliably annotate deeply interpretive phenomena when framed as hermeneutic undertakings.

Towards an Interactive ‘Evidence-RAG’ Peer-Review Workspace for the Journal of Digital History

Élisabeth Guérard, Mehrdad Almasi, Marion Salaün, Frédéric Clavert (C²DH, University of Luxembourg)

This presentation proposes a workflow that opens the AI-based peer-review black box through claim-evidence mapping, methods auditing, and reproducibility verification while maintaining human scholarly judgment.

Coffee break

Parallel workshops

Hail the other side: To which extent does dataset documentation inform both human and machine worlds?”

Mari Wigham (Huygens Institute for Dutch History and Culture); Giulia Osti (University College Dublin); Maria Eskevich (Huygens Institute/Royal Netherlands Academy of Arts and Sciences); Jörg Lehmann (Staatsbibliothek zu Berlin)

The workshop critically examines how DCH dataset documentation templates expose epistemic fractures between historical contextualization and ML’s task-driven determinism, reframing “bias” as an interpretive signal rather than an error to mitigate.

Algorithmic Historical Fiction and Instructional Design: Redefining History Learning in the Age of AI

Elias Stouraitis (University of Thrace)

This workshop examines “algorithmic historical imagination” as a pedagogical framework where AI generates hypothetical scenarios for students to evaluate against historical evidence, cultivating critical literacy and moving learning from memorization to participatory, methodologically self-aware inquiry.

Impresso: (Re)Thinking Historical Sources and Collections Retrieved with AI Tools in Large Digital Infrastructures

Cao Vy; Daniele Guido; Caio Mello; Ferdaous; Marten Düring (C²DH, University of Luxembourg)

The workshop demonstrates the Impresso Datalab’s AI tools for multilingual historical media exploration (including NER/NEL and embedding-based search) while critically reflecting on how researchers construct, evaluate, and interpret datasets retrieved through AI-driven pipelines.

Lunch break

Panel VI: History of AI and Simulation

Chair: Finola Finn (C²DH, University of Luxembourg)

A History of Simulation in the European Union: Politics of Simulation or Simulation of Politics?

Nele Elina Prinz (Copenhagen Business School)

Prinz traces how simulation emerged as a political technique in EU governance, examining the shift from political representation toward simulation-based decision-making informed by data-driven computational predictions.

From Archive to Algorithm: Generative AI and the Transformation of Collective Visual Memory

Anders Ravn Sørensen (Copenhagen Business School)

Sørensen analyzes how AI-generated historical imagery destabilizes the epistemic status of historical photographs, marking a transition from archive-based authority to algorithmically mediated plausibility in collective remembrance cultures.

Conference Conclusions & Farewell

End of the conference