Evaluate models
Test language-model behavior, ethical reasoning, persuasion, bias, and emerging capabilities.
Research / AI in Higher Education
AI in Higher Education
Katherine Elkins’s work asks what universities should teach when students can increasingly delegate traditional academic tasks to machines. Her answer moves students upstream: from users of AI systems to researchers who evaluate, build, interpret, and question them.
Humanities students should help investigate and shape AI, not simply learn how to use it.
Since 2016, Elkins’s human-centered AI curriculum at Kenyon has combined computational methods, interpretation, ethics, and model evaluation. The educational model treats humanities and liberal-arts students as participants in experimental AI research.
The central model
Students as AI Researchers
Students do not encounter AI only through prompting, software demonstrations, or abstract ethics discussions. They ask research questions whose answers are not already known, then design methods for testing them.
This is AI education as research participation rather than software training: students work directly with model behavior, cultural data, narrative, translation, ethical reasoning, and emerging systems.
Developed beginning in 2016
A Decade of Human-Centered AI Education
The curriculum began when AI was still largely encountered in liberal-arts education as something built elsewhere. Elkins and Jon Chun instead created courses in which students worked directly with computational methods and emerging AI systems.
The model evolved from computational humanities and narrative analysis through machine learning, model evaluation, generative AI, multimodal systems, and agents. The Kenyon AI CoLab is the student research environment supporting this work; the human-centered AI curriculum is the curricular program. Neither is the Human-Centered AI Lab, Elkins’s separate nonprofit organization.
Research in practice
What Students Do
Analyze language and culture
Use computational methods to study narrative, sentiment, translation, interpretation, and cultural data.
Build experiments
Design evaluations, fine-tune or customize systems where appropriate, and investigate multimodal and agent-based AI.
Interpret results
Bring disciplinary knowledge to ambiguous evidence, hidden assumptions, uncertainty, and questions of human responsibility.
Public scholarship
Learning Through Original Research
More than 190 undergraduate research projects have generated over 125,000 downloads across 4,700+ institutions in 198 countries. Mentored Research maintains the current counts and the full project record.
190+research projects
125,000+downloads
4,700+institutions
The educational response
Beyond AI Literacy
Knowing what AI is, how to prompt it, and how to detect errors matters. But as systems perform more traditional academic tasks, higher education must move beyond tool literacy toward judgment.
Students need to learn how to frame problems, evaluate outputs, interpret evidence, understand assumptions, manage uncertainty, and decide when disciplinary expertise and human reasoning should remain authoritative.
The institutional question
What Is the University For?
A(I) University in Ruins: What Remains in a World with Large Language Models?
Generative AI does not simply create a cheating problem. It exposes a deeper one: many forms of academic work were designed around tasks that machines can now increasingly perform. The response is to place more weight on interpretation, research design, disciplinary reasoning, original inquiry, and human responsibility.
The Crisis of Artificial Intelligence: A New Digital Humanities Curriculum for Human-Centred AI
The article articulates AI Digital Humanities as an educational framework in which computational systems support inquiry into knowledge, interpretation, meaning, and public consequence—not merely answer production.
Disciplinary expertise
Why the Humanities Matter
Language models make interpretation technically consequential. Ambiguity, framing, narrative, culture, persuasion, and value are not peripheral to these systems; they shape how the systems behave.
Humanities training contributes rhetorical analysis, cultural context, theories of authorship, ethical reasoning, and evaluative judgment—the expertise needed to understand what model outputs assume and when they should be trusted.
Teaching, experimentation, publication
Curriculum, Lab, Archive
The curriculum leads into work at the Kenyon AI CoLab; mentored projects then become public repositories that later students can question, extend, or overturn. IPHS provides the academic home for part of this work at Kenyon.