Research areas
Five threads of inquiry spanning AI safety, computational narrative, archival intelligence, and the philosophy of knowing.
AI Safety & LLM Evaluation
How language models fail at negation, prohibition, and persuasion. Audited 16 models across 14 ethical scenarios — open-source models endorse prohibited actions 77% of the time.
Language, Narrative & Machine Intelligence
Co-developed the SentimentArcs methodology with Jon Chun: the first large-ensemble approach to diachronic sentiment analysis in full-length literary narratives.
Archival Intelligence & Cultural Heritage
Rescuing endangered New Orleans heritage archives using AI with community-governed data sovereignty.
AI Governance & Comparative Regulation
Comparative analysis of AI regulation across the EU, US, and China. Ethics-based audit methodology for LLM normative values.
Foundations: Embodied Experience, Memory & Representation
These essays establish the philosophical position that all the work above extends: mechanistic models of knowing fail to capture what consciousness, literature, and language actually do. AI has made this claim newly urgent and newly testable — but the claim itself is not new to this work.
A timeline of anticipation
-
2016
Co-founded Human-Centered AI Lab and world's first human-centered AI curriculum
-
2019
First transdisciplinary AI research, Modernist Studies Association
-
2020
"Can GPT-3 Pass a Writer's Turing Test?" — published months after GPT-3 API release. Now 382+ citations.
-
2022
Helix Center roundtable on NLGs with Ned Block, Francesca Rossi, Kyunghyun Cho — one month before ChatGPT
-
2022
The Shapes of Stories (Cambridge Element, CUP)
-
2023
NIST AI Safety Institute Consortium — appointed PI representing MLA
-
2024
ICML oral presentation (top 2%): open-source generative AI risks
-
2024
Schmidt Sciences HAVI grant — Archival Intelligence
-
2025
OpenAI Higher Education Forum, Weill Cornell Medicine-Qatar keynote
-
2026
ICML 2025 accepted (agentic AI decision bias), Chronicle of Higher Education Forum
Selected publications
"When Prohibitions Become Permissions"
Katherine Elkins and Jon Chun. Auditing negation sensitivity in language models.
"The Paradox of Robustness"
Jon Chun and Katherine Elkins. Decoupling rule-based logic from affective noise in high-stakes decision-making.
"Near to Mid-term Risks and Opportunities of Open-Source Generative AI"
Eiras, Petrov, Vidgen, Schroeder, Pizzati, Elkins, et al. Oral presentation (top 2% acceptance).
"Can GPT-3 Pass a Writer's Turing Test?"
Katherine Elkins and Jon Chun. Journal of Cultural Analytics, 2020.
"A(I) University in Ruins"
Katherine Elkins. What remains in a world with large language models.
"Comparative Global AI Regulation"
Jon Chun, Christian de Witt, Katherine Elkins. Policy perspectives from the EU, China, and the US.
"The Crisis of AI: A New DH Curriculum"
Jon Chun and Katherine Elkins. Human-centred AI curriculum for the humanities.
"Sentiment-XAI Greybox Ensemble"
Explainable AI with GPT models and interpretable white-box classifiers for sentiment analysis.