Faculty adapt teaching strategies to address AI use
Departments across MIT rethink course policies as AI recommendations roll out
Following the Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training at MIT’s report, The Tech surveyed professors to understand their stances on AI in their courses and how they incorporated the report’s recommendations into their AI policies.
Many students worry about how professors use AI tools and whether they receive adequate guidance on using the tools in an educational context. This ambiguity is a friction point for many students, a sentiment prevalent in the report; it found that AI overuse in the classrooms makes students feel like “teaching is not a priority for their instructors.” To this end, the report urged instructors to ensure every MIT subject has a clear policy on generative AI use in the syllabus and on the course website.
The report’s appendix included sample AI policies for course syllabi, drawing on survey feedback, existing AI policies submitted by faculty and instructors, and similar AI use policies on other campuses, Co-Director of the Ad Hoc Committee Eric Klopfer said.
Faculty members across departments are implementing these recommendations in various ways. Professor of Applied Mathematics Laurent Demanet emphasizes the instructor’s role as a “human shepherd,” guiding students to use AI as a tool for learning, not as a crutch.
Regulating AI use outside the classroom is a challenge. Still, it requires special consideration, since most graded homework in technical classes at MIT has traditionally been take-home written problem sets. Demanet teaches 18.03 Differential Equations with a fairly liberal AI policy.
According to the syllabus, course staff will not police AI use beyond requiring disclosure. In fact, Demanet recommends using AI in cases like reviewing class notes, but she also warns that over-reliance can damage exam scores.
Real World Computation with Julia is taught by Professor of Applied Mathematics Alan Edelman; the class’s grading and structure have been completely restructured compared to past semesters, said Mary Feliz ’28, a computer science (Course 6-3) and mathematics (Course 18) double major currently enrolled in the course. From a copy of the syllabus that Professor Edelman shared with The Tech, the AI policy opens with 2 statements: “1. Humans like to learn, at least when it is not too stressful. 2. LLMs [Large Language Models] can help us leapfrog our learning.” The guidelines go on to explain that LLMs can help students learn concepts required for a deep understanding of course material that they may not have learned in prerequisite courses. For everything else, the syllabus advises students to remain “engaged and asking questions”.
Some faculty say the Ad Hoc statement’s specific recommendations did not necessarily change their AI policy. Still, these professors believe there has been improvement in the communication of these expectations between faculty and students.
8.01 Classical Mechanics co-leaders Krishna Rajagopal and Michelle Tomasik wrote via email, “The 8.01 policies this year are not different in their intention than they were last year, but we hope and think that they now provide our students with greater clarity about our expectations.” They cited the Ad Hoc statement as helping them improve how they communicate their expectations around AI use to students. As 8.01 is a class designed with a collaborative approach to learning, the professors recognize that AI is now a tool that many students use to learn and study — as a result, they require any AI use cited as a collaborator alongside a brief description of the specific use on each problem set.
Some professors, such as Associate Professor of Science, Technology, and Society Oliver Rollins have opted not to outline an overarching AI policy in the course syllabus, but rather approach use on a task-by-task basis.
“Everyone’s using it in everything else, right?” Rollins said. “We draft emails with it, we do a bunch of other things with it, so if you’re trying to limit [AI usage] within the classroom, I just think it’s an impossible task.”
Rollins acknowledges that as AI continues to evolve and expand, more regulation and discussion will be required, but cautions that it must be done with care and consideration. AI can help bridge gaps in understanding when access to human instruction is limited, or for students who may have had less prior instruction in reading and writing, or as a translation tool for students who may not have English as their first language, he said.
“If we police AI, the people that we’re going to attack are folks who are already in these really precarious, marginalized positions,” Rollins added.
It is difficult for both faculty and students to understand and adapt to rapid AI advances in real time, let alone create and agree on policies for its use. Demanet captured the uncertainty well: “We’re not sure how to meet the moment in our classes.”