NIST CAISI · Schmidt Sciences HAVI · Cambridge University Press · Kenyon AI CoLab
AI safety cannot be solved by engineering alone. Katherine Elkins works on the parts that require culture, narrative, judgment, memory, and institutions.
Elkins’s work connects literary and cultural analysis with computational methods, human-centered AI, and model evaluation. Current projects include AI safety research with the MLA team at NIST CAISI and Archival Intelligence through Schmidt Sciences HAVI; at Kenyon, she and Jon Chun began developing a human-centered AI curriculum in 2016.
Signature articles
Journal of Cultural Analytics · 2020
The most-cited article in the history of the Journal of Cultural Analytics.
Published within months of GPT-3’s release, the article tested literary imitation, reasoning, and authorship before those questions became central to public debate.
International Journal of Humanities and Arts Computing · 2023
The most-read and most-cited article in the International Journal of Humanities and Arts Computing.
The article argues for a humanities-led model of human-centered AI and has since been cited across education, medicine, science, and digital humanities.
Selected work
Cambridge University Press · 2022
The most-cited Element in Cambridge’s Digital Literary Studies series. The book develops a methodology for sentiment analysis of narrative and shows how emotional arcs reveal patterns not captured by plot or character alone.
ICML oral · 2024
The position paper developed a benefit-risk framework for open-source generative AI; subsequent work at FAccT, TMLR, and in policy research has used that framework to analyze openness, accountability, and open-weight model risk.
Co-leads the team representing the Modern Language Association at the NIST AI Safety Institute Consortium — the only humanities-led team in the federal consortium.
Co-PI of a Schmidt Sciences HAVI project building open AI tools to rescue endangered cultural archives, beginning in New Orleans.
Current research examines narrative and emotion in model evaluation and generative AI, alongside translation and cultural transmission across languages.
In others' words
“An ‘objective’ model, according to Katherine Elkins, cannot exist.”
— Cao et al., 2025
“Debating the system of values we wish these tools to align with is the first step.”
— Klowden and Tao, International Journal of Humanities and Arts Computing, 2023
Speaking, media, research, and collaboration inquiries → Contact