AI and Education · AI governance; higher education
Same Course, Different Rules: Instructor-Level Governance of Student–AI Interaction in Higher Education
1,836 syllabi show GenAI rules are set by instructor judgment more than by discipline or course.
- Stage
- Analysis complete · drafting
- Last activity
- 2026-08
Most universities responded to generative AI by delegating the decision downward: set your own course policy. That is a governance choice, and this project measures what it produced at one large public university across three years and ten departments.
Discipline turns out to govern whether an instructor says anything about AI at all, but almost nothing about what the rule permits once stated. The strictness of a policy is overwhelmingly an instructor-level property, and a stable one — the same instructor teaching the same course in consecutive years keeps the same stance far more often than chance. Two students in different sections of one course can therefore face genuinely different rules, not because the discipline or the material differs, but because their instructors hold durable and divergent views about what AI use means.
The record
- Idea
- Use within-discipline and within-course comparisons to locate where decentralized university AI policy variation originates: disciplinary context, course context, or stable instructor judgment.
- Research Question
- When universities delegate generative-AI policy to instructors, how much policy variation is explained by discipline and course context, and how much persists at the instructor level?
- Key Proposition
- Discipline shapes whether instructors state an AI rule but explains little about what the rule permits; courses constrain policy variation, yet stable instructor-level judgments sustain substantial differences even within the same course.
- Data
- 1,836 coded syllabi from ten departments at one large public university, Summer 2023–Winter 2026; instructors identified for 99.1%. Analyses include logistic and multinomial models, multilevel variance decomposition, pairwise disagreement measures, permutation nulls, and 508 within-instructor/course transitions.
- Analysis Result
- Discipline and department explained 19.5% of whether a rule was stated but only 1.4% of stated strictness; the instructor component was 68.4% of strictness variance, although it is an upper bound because course is not nested. Same-course instructors disagreed in 43.5–48.2 pairs per 100 versus a 70.0 corpus baseline, but substantial disagreement remained. When a rule appeared in consecutive years, the same instructor retained the same stance 76.3% of the time versus 30.0% under independence.