Katherine Elkins studies how artificial intelligence reshapes human meaning: how models interpret language, generate stories, encode values, simulate creativity, alter institutions, and change the conditions under which people judge, teach, remember, and know.
Humanistic methods for evaluating frontier language models in ethically and linguistically complex settings — ethics-based audits, FATE evaluation through the Notre Dame–IBM Technology Ethics Lab, syntactic framing fragility, and CAISI/NIST standards work.
The institutional conditions under which AI can serve public ends — comparative regulation, open-source and open-weight model risk, public AI infrastructure, and cultural data governance, including work connected to UNESCO’s AI, IP & Culture Repository co-design process.
Methods for tracing emotional structure across narrative, from The Shapes of Stories through later work on narrative intelligence — emotional arcs that reveal structures plot and character analysis miss.
How knowing happens — consciousness, memory, perception, authorship, and interpretation across Proust, Wordsworth, Plato, Baudelaire, Kafka, Woolf, Maryse Condé, and contemporary AI.
What language models can generate, imitate, translate, and co-create — the writer’s Turing test, AI authorship, the AI Fiction Paradox, literary translation, and early live human-AI improvisation.
Human-centered AI as an educational and institutional project — curriculum, the AI CoLab, student research, faculty governance, and what large language models do to the university.
The AI curriculum article was cited by Tanya Klowden and Terence Tao in “Mathematical Methods and Human Thought in the Age of AI,” where they ask whether AI marks a fundamentally different technological moment for human thought. Archival Intelligence, a Schmidt Sciences Humanities and AI Virtual Institute project, extends this research into AI and cultural preservation. For detailed reception evidence, see Scholarly Reception.
The work moves across AI safety, computational humanities, cognitive science, and philosophy of information. It began in literature and philosophy, with research on memory, consciousness, authorship, perception, lyric authority, and interpretation, then moved into computational humanities through sentiment analysis, narrative emotion modeling, and the study of story shape. As large language models became public infrastructure, the same questions became AI safety, governance, creativity, and higher-education questions.
Across these areas, the work connects humanistic methods to AI safety, governance, narrative analysis, creativity, and higher education.
Several strands of this research are co-authored with Jon Chun, including work on GPT-3 and creative writing, AI and narrative studies, explainable AI for story analysis and generation, open-source generative AI risk, comparative AI regulation, semantic ethical auditing, and human-centered AI curriculum. See also Jon Chun’s research page.