Some of the recent ChatGPT panic in higher education is focused around academic integrity. ChatGPT seems to increase the ease of producing coherent writing with very little personal effort, and moreover to do so in a way that cannot be easily detected using pre-existing plagiarism-detection software. On that description, it seems worthy of panic: ‘easier to cheat and harder to detect’ seems a very dramatic shift in the academic integrity landscape.
But looking through the literature on academic integrity, there are at least 2 explanations of academic dishonesty that have not changed with the launch of ChatGPT.
First, the psychology of academic dishonesty seems likely to remain the same even if cost-benefit analysis might shift in the context of ChatGPT. Many studies over the years have asked students and researchers whether they have ever engaged in academically dishonest behaviour, and if so, why? Many of the reasons given in 2019 or 2021 are still the reasons given in the ChatGPT era, so they are worth delving into.
Prior to the launch of ChatGPT, the received wisdom was that conscious violations of academic integrity policies were most likely when students felt cornered. De Maio and Dixon (2022)‘s review finds that students who admit to academic misconduct converge on several reasons or explanations: students tend to cite time pressure, peer pressure, and grade pressure as reasons for their academically dishonest behaviours. Tindall & Curtis (2020) find that “negative emotions” such as stress, anxiety, and depression are correlated with positive attitudes towards plagiarism. In short, students are most likely to engage in academic dishonesty when they feel under pressure, one way or another. Some of these pressures may arise for reasons internal to course design, while others might arise because students are three dimensional people with lives (and pressures) outside of any particular course they may take. Syllabus design can potentially mitigate both.
Secondly, unknowing or ignorant violations of academic integrity requirements represent a significant portion of academic misconduct cases. Ignorance of the nuance of the requirements of academic integrity is frequent amongst students (and not nonexistent among faculty and researchers), so mistaken or accidental or ignorant violations of academic integrity requirements are to be expected. ChatGPT might add to a pre-existing confusion, but only by a matter of degree.
Proceeding with caution
These two features (stress and confusion) are not new with ChatGPT, and they frame my two pronged approach to academic integrity in the era of ChatGPT.
First, in recognition of the proportion of students and academics who are confused by what ChatGPT is or does, take an educative rather than a punitive approach to the existence of ChatGPT. Teach what academic integrity requires for your educational context and explicitly mention ChatGPT. De Maio and Dixon (2022) emphasize that clear academic integrity statements and policies are important for maintaining a culture of academic integrity. McGee (2013) describes how cases of academic dishonesty are higher when expectations are unclear, and points to improvements in academic integrity when academic integrity expectations are made explicit at both the course and institutional level. If your institution has any specific resources on academic integrity or Honor Codes, show your students how to access them.
In the ChatGPT era, this might be as simple as including an explicit ChatGPT statement in the syllabus or in each assignment description. “For this assignment, use of ChatGPT and other LLM AI is not permitted for either generating the assignment nor for editing and improving the assignment.” But, if you go this route, be consistent throughout your course and assessment designs. Include an explicit ChatGPT statement in your syllabus and in each assignment description, because its omission in only one place might be noticed.
If your preference or context allows, you may wish to redesign some assignments to explicitly allow ChatGPT. There are some good ideas for reimagining assessment using ChatGPT out there, with some key points being to keep ChatGPT components explicit and optional, and keep the credit proportional to the effort involved (i.e. likely low-stakes). If you are in a course context where assignment redesign is possible (e.g. small enrolment, low prep load, TA support, full control over assignment and syllabus design, etc.) your colleagues who are not so positioned will appreciate it if you explicitly mention that your ChatGPT-friendly policy applies only to your course, and does not extend to other courses.
But while we are on the subject of assessment design, the second component of my approach to ChatGPT is to acknowledge and work to minimize the psychological stressors known to be used as explanations of explicitly academically dishonest behaviour. Low-stakes and flexible assessments help students mitigate grade-pressure and time-pressure. Offering students choices of prompts, choices of material, choices of occasions, or flexibility with deadlines can all empower students in ways that help them manage their stress. Flexibility in course and assessment design – such as flexible deadlines, flexible forms of engagement, and flexibility in modes of delivery – are part of a Universal Design for Learning approach. Fovet (2020) explains how flexibility and choice serve as components of UDL, and Amrane-Cooper et al (2021) point to a connection between flexibility in assessment design and easing of anxiety.
Sotiriadou et al (2020) suggest that authentic assessment design can reduce incidents of academic dishonesty. Similarly, De Maio and Dixon (2022) point to authentic assessment design and authentic curricula as reducing incidents of academic dishonesty.
Finally, Bretag et al (2019) point to personalized and reflective assignments as less associated with academically dishonest behaviour. Bretag et al (2019) also point to in-class assignments and vivas (oral examinations) as less associated with academically dishonest behaviour, although these assessment styles are more likely to be limited to smaller class sizes.
Over the coming months and years, ChatGPT will get better at doing what it does, and AI detectors may also become more effective. There will be various iterations of large language model AI, and each will have its foibles and its virtues. But if the circumstances surrounding your teaching do not permit a complete course re-design every semester, educating about academic integrity requirements will help students better understand the contours of the academic integrity terrain, while using assessment and course design to acknowledge and support students navigating the pressures and anxieties related to assessment will make it easier for students to continue to choose academic integrity.
In short, not quite everything has changed with the advent of ChatGPT, and not everything can change in response to ChatGPT.