Center for Teaching & Educational Technology

• The CSU is committed to providing faculty, staff, and students with AI as a resource., It envisions AI as a critical workforce tool going forward. The CSU has invested significant resources in the past two years in helping faculty, staff, and students better understand generative AI, because it understands the disruptive potential of the technology. To support safe exploration of the technology, the CSU has acquired a license to ChatGPT Edu for all CSU community members. The benefit of this Edu license is that all CSU data, including your teaching materials, stays private to the CSU. OpenAI cannot read, monetize, or train their AIs on any of our content.
• At present, there is no senate policy on generative AI. Faculty members have the academic freedom to incorporate as much or as little AI into their courses at their discretion. Additionally, the Academic Technology and Instructional Spaces Subcommittee (ATISS) has created an excellent resource on guidance for AI in the classroom. 

As described above, generative AI policies in the classroom are up to the individual instructor. Instructors have the academic freedom to determine the extent to which, if any, that generative AI is incorporated into their courses. Because instructors vary on their stance on this issue, it is your responsibility to explicitly and thoroughly communicate your expectations to your students, regardless of whether or not you choose to allow the use of generative AI in your course. A statement in the syllabus is a good place to begin but should not be the only means of communication. We recommend the following:

1. Syllabus statement: This should clearly state your expectations around AI use in every relevant aspect of your course. It should state what is permitted, what is not permitted, and why. Be as specific as possible. For instance, can students use AI for brainstorming? Proofreading and citation formatting? Where is the use of AI absolutely off limits, and why? What would constitute academic misconduct and what will the process be if you suspect this? Here are some sample syllabus statements. We recommend that you adapt them and customize them to the specifics of your course. 

    

2. First day of class. Take a dedicated chunk of time to talk with the students about AI, regardless of your stance on its use. At the very least, you should go through the expectations in your syllabus statement and invite any and all questions. You may also invite students to share their experience with AI, both academically, personally, or even at their jobs. Emphasize that your goal is to learn about their experience, the extent of their use, and their knowledge base, not to determine if they misused it. CTET also has an AI use survey (LINK) that you can use and/or modify. 

3. Reminders on assignment prompts and online exams. This is where you draw specific connections between your syllabus statement and the nuts and bolts of your assignments. Including what use of AI is and is not permitted directly in the prompt will clarify your expectations and can prevent unintentional or intentional misuse and academic misconduct. 

For example: Use of Chatgpt for proofreading and generating citations in APA format is fine. Just know that there could be errors with citations, so check. For proofreading, it is still a very good idea to read through your paper for another round of revisions. This is the extent of AI use that is permitted for this assignment. You should not be using AI to generate the text of the paper, to conduct your analysis, or to search for articles that support your thesis statement. Because the goal of this assignment is for you to demonstrate independent critical thinking skills, use of AI constitutes cheating, plagiarism and academic misconduct. 

Currently, there are no foolproof tools to detect misuse of generative AI. You may have heard that detection tools such as Turnitin, GPTzero, and writer.com bolster accuracy rates of up to 98%. Unfortunately, these claims have yet to be substantiated. Independent studies show far lower accuracy rates, and pilot studies at individual institutions that use the tools also show low accuracy rates. These tools are also highly susceptible to false positives - where students’ original work is incorrectly flagged as AI generated - and false negatives - failing to detect AI generated text. These inaccuracies have led to a number of institutions disabling the tools after an initial pilot. For instance, Vanderbilt University, considered to be a global leader in generative AI implementation and research, disabled its Turnitin detection tool in 2023. Additionally, CTET at Sonoma State conducted its own pilot of Turnitin and Copyleaks in Fall 2023 and observed false positives during the process. As a result, ATISS recommended that CTET not license an “AI detection tool” until the SSU Academic Senate had developed a formal policy around AI usage.

In addition to their limited efficacy of these tools, there are additional problems with implications for equity and student success.  First, research shows that AI detectors have been found to incorrectly label text by non-native english speakers as AI generated (Myers, 2023). Second, because Sonoma State does not offer institutionally-supported tools, entering student assignments into third party detection tools may violate FERPA, privacy and copyright protections. 

Like the detection tools, Generative AI itself is not foolproof and is prone to errors and inaccuracies. If you suspect the use of AI in an assignment, consider the following:

1. Check for content that is inaccurate or just plain wrong. AI is prone to generating inaccurate information, as well as citing studies and research that did not occur. Look for multiple errors, unsupported claims, and inconsistencies with the research literature related to the assignment. 

    

2. Check for inconsistencies with what you have taught. If you are reading along and thinking “Wait, I never said that! Where is this coming from?”, it is possible that AI was used. Additionally, if the responses diverge from what was assigned in the prompt or the approach you took in class, this may also indicate AI use. That said, it is equally possible that the student misunderstood something you said in class, made errors when citing, or was confused about the assignment’s requirement. So the presence of these inaccuracies should be considered suspicious but not evidence of potential AI use.  

    

3. Compare with the student’s previous work. If you notice a significant change in tone or writing quality, or a student’s work seems to have suddenly shifted from beginner to advanced (especially if their attendance has been spotty), this may be a red flag.

    

4. Check the citations. As noted above, AI is prone to generate inaccurate or false citations. If a journal article, book chapter, or another source has incorrect dates, authors, or did not appear in a library database search, this may indicate an AI-generated reference. 

This is an area where CTET can provide instructors with resources and direct support. The best way to prevent the use of AI is to modify existing assignments to discourage and prevent its use. We are not suggesting that your existing assignments be scrapped and redesigned, only modified. You can schedule a consultation right now with us to help you with this. Or you can implement one or more of the strategies presented below.

Localize assignments

The more you require students to use and cite material that they learned in class, the less likely AI would be in generating text that would meet assignment expectations. Class lectures, asynchronous videos, discussions, and specific texts are not readily available online, and if students do use AI on the assignment, it will be easy to see that the material is not from class. This tactic also has an added bonus of encouraging attendance. 

Scaffold assignments. 

Breaking an assignment down into several smaller tasks carries several benefits. Lengthy papers and large projects tend to foster student procrastination, often resulting in a disappointing final product. Breaking the assignment into small tasks introduces structure, providing an opportunity to identify knowledge and skill gaps, while offering feedback for iterative improvement. It can also minimize the likelihood of AI use, especially if class time is used for these tasks. 

Incorporate in-class activities into assignments. One way to scaffold assignments is to devote class time to an aspect of the assignment, and have students elaborate and deepen their analysis in the next phase. For instance, “In class, your group was asked to jot down ideas for how to motivate people to eat more fruits and vegetables. First, summarize what your group brainstormed. Next, link these ideas to what you learned in class on the psychology of habit-building.”

Incorporate application and personal reflection. Personal examples and reflections generated by AI are likely to be general and bland, making it easier to distinguish from original work. Also, having students apply course concepts to their life experiences not only facilitates retention, it can dissuade students from turning to AI.