Shared problems, different kinds of expertise

Why the work needs more than one field

Jon Chun works with humanists, social scientists, engineers, technical researchers, archivists and other domain experts, students, policymakers, and practitioners because the problems themselves cross those boundaries. Evaluating a model's judgment requires both technical tests and a defensible account of judgment. Building for a cultural archive requires computational infrastructure and knowledge of language, history, evidence, and community context. Teaching new AI methods requires students to enter the research as contributors rather than spectators.

These collaborations are organized around shared work: defining the problem, deciding what evidence matters, building or testing a method, and learning from the failures that each form of expertise can see. The institutions matter because they make that work possible, not as a list of affiliations.

Archival Intelligence

Archival Intelligence is part of the Schmidt Sciences Humanities and AI Virtual Institute. Jon Chun serves as Co-Principal Investigator. His role includes computational methods, workflows, evaluation design, and open research infrastructure. He works with Katherine Elkins, humanities experts, engineers, archivists, and institutional partners on endangered cultural archives in New Orleans, including Creole and Cajun multilingual newspapers and early jazz materials.

The collaboration joins archival and humanistic knowledge with computational methods, AI workflows, and evaluation design. Domain experts help determine what must be represented, which histories and language differences matter, what counts as relevant evidence, and how provenance and ambiguity should remain visible. Technical development turns those requirements into workflows that can work with difficult source materials, including smartphone photographs, without allowing convenience to erase context.

The project is a working model for collaboration between interpretation and engineering: questions discovered in the archive change the technical design, and system behavior gives historians and domain experts new objects to examine. See how the archival system is being designed →

Standards, reasoning, and human judgment

Jon Chun serves as Co-PI representing the Modern Language Association in the NIST CAISI consortium. The collaboration brings humanistic methods into standards-facing work on LLM evaluation, red-teaming, and ethical auditing. The team's results were presented during the opening keynote at the consortium's first plenary at the University of Maryland.

A related Notre Dame–IBM Technology Ethics Lab project joined Jon and Katherine Elkins with Notre Dame's Yong Suk Lee to audit how large language models reason in high-stakes decisions. The team used juvenile recidivism as a test case and evaluated fairness, accuracy, transparency, and explainability together rather than treating model output as the only result.

With Christian Schroeder de Witt and Katherine Elkins, Jon also co-authored a comparative study of AI regulation across the EU, China, and the United States, extending the collaborative frame from model behavior to the institutional choices that shape deployment. The work connects researchers at Kenyon and Oxford around a concrete governance question. Read the frontier-evaluation research →

Testing openness in practice

Through the Meta Open Innovation AI Research Community, Jon has participated since 2023 in research conversations about open AI, model access, evaluation, and deployment. His co-authored position paper on the near- and mid-term risks and opportunities of open-source generative AI brought technical researchers and humanistic analysis into the same argument and was presented as an ICML 2024 oral.

Multi-agent research and teaching provide another shared workspace. Students and collaborators build systems, test agent behavior, and examine how deployment choices alter risk. Practitioner conversations, including BWG Global forums on enterprise adoption, supply a modest reality check on how organizations use the same models outside academic settings. See the open-AI and agent research →

Students as research collaborators

At Kenyon College, collaboration includes building people and research capacity. Jon designed the technical curriculum, computing courses, and technical research environment behind the AI CoLab; he and Katherine Elkins developed the broader interdisciplinary Human-Centered AI model together. Students learn enough programming, machine learning, generative AI, and evaluation to bring their own disciplinary questions into technical work.

Original research is the organizing principle. Student teams formulate questions, implement computational approaches, test assumptions, and publish their strongest work through Digital Kenyon. This connects teaching to Jon and Elkins's longer joint research program in computational narrative, where technical instruments and interpretive questions continually reshape one another.

Since 2019, their shared work has included Middle Reading, the Writer's Turing Test, explainable story analysis, translation, and cross-cultural narrative comparison. The full dated chronology is available at Narrative Translation & Transmission →. Explore the teaching and mentoring model →

Putting methods into conversation

Some collaborations are designed to test ideas in public rather than produce a single system or paper. At the Helix Center for Interdisciplinary Investigation, Jon has joined conversations on AI, language, and emotion. His work has also entered UNESCO and United Nations conversations on education and global AI governance, while the nonprofit Human-Centered AI Lab helps distributed researchers and domain experts collaborate across institutional boundaries.