Guidance for Faculty on Addressing AI-Related Academic Integrity Issues

The following guidance is intended to help faculty prevent AI-related academic integrity issues in their courses and address them should they occur. Ideally, faculty can manage AI-related academic integrity by structuring course assignments and assessments that are in alignment with their policies. However, when academic integrity issues do arise, clear policies, expectations, and processes can help faculty engage in fair and effective communication and resolution of the issue.

 

Get Ahead of the Problem

Clearly State Course AI Policy in the Course Syllabus

The Office of Undergraduate Education AI Guidance provides several sample policies with different approaches to limiting/allowing AI use in coursework. 

In addition to setting the policy in the syllabus, discuss your policy in class and remind students of it particularly when they are working on assignments. 

Explain what counts as AI in your course all the way from fully generative AI tools like ChatGPT to Grammarly, predictive text in Google Docs or Microsoft Word. If any of these are allowed, encourage the student to cite the use of the tool. Encourage students to ask, at any point in the term, if they are unsure about whether a particular use of AI is in violation of the policy. 

Both in your syllabus and verbally in class, indicate that one purpose of the assignments is to assess student understanding of the material. As such, students may be asked to demonstrate their understanding of the work they have submitted, including via oral exam after submission. 

Match Your Assignments to Your AI Policy

If your AI policy is permissive: 

  1. design assignments where success requires demonstrable student learning beyond what AI can generate, and
  2. to the extent that output quality depends on skillful use of AI tools, provide resources that show students how to use the tools effectively in disciplinary context

The Bok Center offers concrete strategies for teaching in the age of AI that can help you adapt your assignments. 

If your AI policy is restrictive: 

  1. limit use of assignments such as take-home work that can be completed successfully with the aid of generative AI; and
  2. include other assignments that directly assess student mastery of content in contexts in which access to generative AI is not possible, for example, in-class assignments and/or tests/exams. 

The first is important because these assignments put students in a difficult position: students are expected to operate on the honor system in a context in which they perceive that their peers may be gaining an unfair advantage.

The second is important because it incentivizes students not to rely too heavily on outside help, from generative AI or other sources because they will have to individually perform on other assessments. Additionally, it provides a fuller picture of the mastery of a given student so that anomalous performance can more easily be detected. 

Design Your Assessments to Verify Student Understanding

Instructors are encouraged to design assignments that include measures that require students to demonstrate understanding of the work they have submitted. We recommend that instructors thoughtfully integrate this approach into their course design by linking an assignment’s grade not only to the work submitted but also to the student’s demonstrated understanding of the material.

For example, an instructor might require students to deliver an oral presentation or engage in a discussion about the content and process involved in completing an assignment. This practice becomes particularly effective if the grading weight of the assignment incorporates the post-submission check of understanding. So long as checks are equitably applied to students, they need not be applied to every student on every submission. 

 

Address Problems That Arise

Meet with Students to Assess Understanding

We recommend that instructors who think a student may have inappropriately used generative AI begin by meeting with the student to discuss their submitted work. It is recommended that this meeting take place soon after the work is submitted. In this meeting we suggest that instructors: 

  1. Share the concern they have about the assignment and ask the student whether they used generative AI or another aid to produce the work,
  2. discuss the reasons the work raised concerns and give the student an opportunity for explanation, and
  3. ask the student some questions about the content of the work and/or process of completing the work that give the instructor a sense of whether the student has a sufficient understanding of what they submitted to have plausibly authored it without aid of AI. Here are a few examples of the types of questions one might ask:
    • Can you walk me through your answer to question X?
    • Tell me about how/why you chose these sources.
    • You have used/applied X term/theory in your essay/answer. Can you explain it to me?

If the instructor believes AI (or other unauthorized means) were used in contravention of course policy, they are encouraged to refer the issue to the Honor Council along with an account of the meeting with the student. It is important to keep in mind that the Honor Council members are not disciplinary experts. The instructor should be as clear as possible about what in the student’s responses convinced them that the student submitted work authored by or with the aid of AI. Instructors who have questions about the Honor Council process are encouraged to contact the Honor Council directly.

Other potentially useful indicators of AI use

When submitting a case to the Honor Council, keep in mind these potentially useful indicators of AI use:

  • Evidence of a significant discrepancy in style or content mastery between different assignments submitted by a single student can be useful in determining whether AI or other inappropriate aid may have been used. For example, a student may submit an assignment that appears to show significant mastery of a subject but then demonstrate a lack of understanding of the same material on an in-class assessment that follows shortly after.
  • Concrete data such as the document revision history for a paper or timestamps from other educational technology systems used by a course can be helpful supporting evidence.
  • Citation of hallucinated sources or quotations. 

Unreliable/Non-Actionable Indicators of AI Use

  • AI “checkers” are not, to date, reliable and often over-identify AI use. At this time they do not produce reliable evidence.
  • Use of specific sentence structures, punctuation, or formatting in isolation are not necessarily sufficient to accurately indicate AI use. A significant pattern of or combination of these can be some evidence of AI use.
  • Citation of topics, sources, or techniques from outside the course materials are not necessarily conclusive evidence of use of AI. Some students use a wide range of sources beyond course materials to self-teach. This behavior should be directly discouraged if it is detrimental to student learning.  

Spot Other Academic Integrity Issues

When problematic work is submitted, identify whether the problem is solely that the student used generative AI in contravention of a policy or whether there are other issues with the work, such as non-existent sources or lack of citations, for instance, that raise academic integrity issues regardless of how the student produced the work.