What Can Higher Ed Faculty Do About Student Misuse of chatGPT and Generative AI?

Faculty have reached out to me wondering what they can do about student misuse of chatGPT.?While some information has been published online suggesting steps educators can take to defend against cheating and other academically dishonest uses, many still feel lost for solutions.

I have been beta testing GPT-3 since January of 2021 and have been a member of the API community for the last two years.?My understanding of the technology was heightened through these experiences, so here are the suggestions I am currently providing faculty in response to the emergence of chatGPT and other forms of generative AI in the classroom.?This list represents my current thinking and opinions, it is not exhaustive, and my focus here is on the misuse of the technology in academic settings, and particularly, in online academic settings.?This article is not about the constructive academic uses of the technology, of which there are very, very many.

I expect my opinions to change as the technology evolves.?This technology is moving fast, and things that seem reasonable today may seem less reasonable tomorrow (and vice versa).?In any event, I hope there is something here that readers find helpful. All suggestions / opinions here are my own.

I separate my suggestions into:

• Immediate measures (actions everyone should take)
• Mitigating measures (actions that may help lessen the impact of student misuse of AI)
• Preventative measures (actions that may eliminate or nearly eliminate student misuse of AI)

Immediate Measures

1. If you have not already done so, go to https://chat.openai.com, sign up for an account, and begin experimenting with chatGPT today.?Until you have some experiential understanding of how this technology works, you will have little hope of mitigating or preventing its misuse.?Visit OpenAI's best practices page for some guidance, or do a Youtube video search for “beginner prompt engineering for chatGPT” and apply the methods you observe.
2. Once you understand how the technology works, update your syllabus to define appropriate and inappropriate use of generative AI and talk to your students about it.?You might feel inclined to define all uses as inappropriate.?I do not personally recommend this.?This technology is already moving into internet search, mobile assistant apps, career preparation services, grocery shopping, and all sorts of other areas where students will interact with it on a daily basis.?Prohibiting the use of generative AI will not be enforceable or realistic and it will create persistent stress and conflict for you and your students.?Alternatively, a policy that defines appropriate use of AI will signal to your students that you understand how this technology works and that any misuse of it will be both caught (hopefully) and addressed.?(Note: there are many beneficial uses for AI in student learning.?For a first dip in the pool, consider checking out this video forum from Bryan Alexander)
3. Avoid AI detection tools.?Generative AI has an infinitesimal ability to generate textual content, and it has that same infinitesimal ability to get around every AI checker.?Youtube is filled with evasive strategies that students can easily leverage, and it will continue to fill with more and more of them.?This is an arms race that plagiarism tech companies cannot win.?In my opinion, the risk of false positives and error in these tools is much too high for educators to tolerate, and though I cannot predict what the future might hold, I suspect it will always be this way.?Keep in mind that the algorithms in these checkers are not trying to outpace the development of generative AI like chatGPT.?Instead, they are trying to outpace the ingenuity and cleverness of students everywhere who are testing, sharing, and discovering workarounds every single day. They will never be able to move as quickly as tens of millions of students on social media can, but who knows - maybe there is a future where they will not need to.

Mitigating Measures

1. Assign your students a personal essay to establish an authentic sample of student writing.?The only requirement is that the essay must be about a personal experience to help ensure authenticity, but to make it relevant to your course, you could weave in a secondary topic of disciplinary significance.?Generative AI is trained on large datasets from the internet and other digital sources.?In general, it knows little or nothing of a student’s life, and hence, students will find it impossible to produce an authentic reflective essay using generative AI.? What we are betting on here is that students want to tell their own story and not let the AI misrepresent them. Regardless of how you get it, having an authentic sample of a student’s writing will allow you to use it for comparative purposes should you suspect dishonest use of generative AI.
2. If you make all class assignments personal this may dissuade students from using AI dishonestly.?The idea is that students will want to write about and discuss things that drive their passion or that they have a deep interest in.?If a student is passionate about dance or ballet, let them tie that into the course topic du jour.?If they are passionate about sports or eSports, let them indulge that.?Game of Thrones or Big Brother??Pack it and ship it to the inbox.?While this approach may incline many students toward authentic work, keep in mind that this is no guarantee.?Generative AI is perfectly ready to expound upon dance, sports, video games, Game of Thrones, and sadly, even Big Brother.
3. People are generally reluctant to engage in dishonest acts towards those they respect or are fond of.?If you build close and trusting relationships with your students (however you may define that), these connections can (in theory) dissuade academic dishonesty.?However, temper your expectations accordingly.?I have observed many educators online pitching this as a high impact practice for solving issues of academic dishonesty, but I remain skeptical.?While I do believe strong relationships with students can serve to mitigate issues of academic dishonesty, I also recognize that many of the reasons why students may exhibit dishonesty have nothing to do with how they feel about their teacher, and in desperate moments, that relationship may never factor into the student’s decision to be academically dishonest.?The allure of AI may be creating environments for dishonesty that we do not yet understand, and relationships may not be the factor it once was. Faculty who teach courses in large lecture halls with hundreds of students will naturally find this suggestion challenging.

Preventative Measures

1. If you integrate generative AI into your class assignments, this can potentially eliminate the misuse of AI in many situations.?This requires basing your assignments on engineered prompts, delivering these prompts to your students, and then instructing them to insert the prompt into a tool like chatGPT.?The resulting conversation can then be exported out of the tool once the session is over, placed into a Word document, and then be assessed as submitted work.?While students can refeed the AI its own responses, in most cases, it will probably be more expedient for the student to simply engage the AI in conversation if the intent is to get a smooth session.?This approach does have some significant limitations.?Liberating text from these tools is currently challenging, and in the case of some mobile devices, virtually impossible.?Students also need to have reliable internet access to effectively complete assignments like this, and that is not always guaranteed.?Lastly, some students may refuse to use a generative AI tool out of concern for privacy or another valid reason.
2. Generative AI models operate based on text.?One way to evade their misuse is to move away from text as the standard format for communicating learning.?Instead, faculty can consider a move to a video format of assessment.?Here written essays are converted to video essays recorded with a smart phone.?Interactions in an online class forum are facilitated through uploaded video instead of text.?Assignments based on short answer or multiple choice responses are tackled in real time while the camera is rolling.?Once again, there are significant limitations to consider.?Some students may not have access to a smartphone.?Other students may experience significant anxiety when recording themselves.?As we learned from the online proctoring fiascos of the last few years, asking students to record video in their homes can be fraught with unknown peril.?As if this all were not enough, video files can be quite large, your LMS may not support the file sizes or formats, and students may need to open Youtube, Vimeo, or similar accounts to make the videos available to you.?This is far from smooth sailing.
3. In my opinion, the best immediate solution to preventing the dishonest use of generative AI may also be the most controversial.?That solution is: return to campus and flip your classroom.?Unsurprisingly, when all assessment of learning and mastery is done in a face to face setting through activities, discussions, and exams, the misuse of generative AI is highly preventable.?While this suggestion may be helpful or even reassuring to faculty already engaged in blended or face-to-face learning, this will be of little comfort to those faculty teaching fully online who cannot change delivery formats.

Do you have suggestions on further measures that educators can take to mitigate or prevent the misuse of generative AI??I welcome those in the comment section below and am eager to learn more about the practices that might be working for you.