By J. D. Wright

I’ve spent most of my career in higher education as one of the academic-integrity true believers — a zero-tolerance hardliner who prosecuted apparent violations with a zeal that might have made you wonder whether I was acting on a staunch commitment to the principles of academic honesty or just perceiving lapses in ethical judgment as personal affronts. In many ways, I’m the last person you’d expect to suggest that an enforcement-first approach to academic integrity, especially in recent years, undermines the work of teaching and learning.

And yet, here I am, making precisely that opening argument in support of my case.

What changed? Generative AI. It forced me to ask a question that I’d never had to face when dealing with cheating: What can I realistically prove? Our old toolkit had a satisfying logic: Feed a paper into a plagiarism detector, watch the highlighted similarities appear as the detector scoured the internet for matches, and build an airtight case with an inevitable conclusion. There was a forensic clarity to the enterprise that felt like a warrant for holding wrongdoers accountable for the highest of all academic crimes, one with existential stakes for the academy. AI broke that logic open.

An LLM doesn’t copy existing material, and its output isn’t quite like plagiarism. It finds patterns in vast bodies of text and uses them to predict, byte by probabilistic byte, a plausible response to a given inquiry. It doesn’t leave fingerprints at the scene, and there’s no smoking gun, at least not in the ways we used to see. So the certainty I once had when I saw a paper as the product of misconduct has had to yield to the complicated realities of indefiniteness and reasonable doubt. I can still analyze evidence, construct a theory, and advocate for a specific outcome, but except in a few idiosyncratic cases, I can’t be sure. “Proof” looks different now.

The surprise was that, instead of feeling thwarted, I began thinking about what I should be doing instead of strict enforcement to serve the core reason we’re all here — to help students learn.

In “Cheating Lessons: Learning from Academic Dishonesty” (Harvard University Press, 2013), James Lang argues that cheating is primarily a course-design problem rather than a student-character problem. Ineffective course-design choices, he convincingly asserts, can incentivize cheating. For example, he identifies common design failures that make cheating feel rational to students: assignments with such high stakes that a single submission can tilt an entire grade; an emphasis on final products over the learning process that should produce them; and tasks that feel like arbitrary credential hurdles rather than genuine intellectual work. Each one shifts the student's calculation in the wrong direction. 

Learning science has addressed these problems for decades, and as it happens, our choices can disincentivize cheating, too. Better still, adopting approaches based on rigorous evidence turns out to be a key factor in demonstrably enhancing education outcomes.

Consider self-determination theory, the framework developed by psychologists Edward L. Deci and Richard M. Ryan. Research on this idea consistently shows that intrinsic motivation is one of the most reliable predictors of academic honesty. Students who are invested in an assignment by virtue of having had some agency and autonomy in developing it (and in showing how they will demonstrate mastery) will be less likely to cheat than students who feel alienated from the process. Surveillance and punishment, contrastingly, tend to reinforce the external-motivation dynamic in which the grade is the entire point and we incentivize cheating from the very start.

Translating these precepts into practices is much less daunting than it sounds. In Lang’s “Small Teaching: Everyday Lessons from the Science of Learning” (Jossey-Bass 2nd ed., 2021), he argues that meaningfully better outcomes don’t require a complete course overhaul — incremental changes, often simple to adopt, yield surprisingly significant improvements in learning. 

“How Learning Works: Eight Research-Based Principles for Smart Teaching” (Jossey-Bass, 2nd ed., 2023) reinforces the same idea and outlines specific measures that reduce cheating while enhancing student learning. Consider, for instance, an assignment that emphasizes process over product (perhaps by requiring rough notes and drafts, reflection components, interim checkpoints, and in-class work in groups). Adopting that approach not only supports students in achieving a more sophisticated metacognitive awareness of their own learning; it reduces the benefits of cheating by spreading the assignment’s value across multiple submissions and increases the costs of cheating by multiplying the number of opportunities for detection and the effort involved.

Similarly, authentic assessments that genuinely matter to students — whether by virtue of them having had a voice in assignment development, having been part of an explicit discussion about the assignment’s purpose, and having spent time with written reflections on their personal goals — increase intrinsic motivation and incentivize the work that supports learning.

You might well ask: What happens when (not if) something still looks, feels or sounds wrong about a submission? Even though we might not be able to detect AI usage, we can design assignments and processes that can guide us toward productive conversations with our students. When a submission includes the document’s version history, for example, and I happen to see that it was completed in only a few minutes, straight through from start to finish, with no in-line edits for typos, errors or substantive changes, it’s reasonable for me to talk to that student about the writing process —including how my own unfolds — and how essays just don’t get written that way. 

My goal? Not to win a conviction or secure a confession (although students sometimes admit that AI was the primary author). My goal is to secure an opportunity to teach — to talk about the writing process, productive struggle, and the connections between writing and thinking; to help students own their choices and understand why they’re significant; to give students a way to identify and work through the challenges that accompany writing, honestly and (let’s just say it) with integrity, to the endeavor of learning, to themselves and their values, and to their futures.

Of course, you should impose a penalty where warranted (doing so in accordance with governing procedures), but it should be integrated into the larger work of teaching and learning. 

AI prompted me (if you can indulge the pun) to make the kinds of changes that learning science has been recommending for longer than I’ve been teaching. The evidence-based approach isn’t softer than the zero-tolerance one. And it’s not easier. It demands more from everyone — more honest conversations; more willingness to see academic integrity as a means rather than an end; and more intentional thinking about our assignments, how we design them, and what purposes they actually serve. But for the record, it’s the better approach.

The prosecution rests.

J. D. Wright, JD, PhD, is a teaching consultant in the University Center for Teaching and Learning and has been teaching literature at Pitt over the span of more than 20 years.

Sources

Lang, J. M. Cheating Lessons: Learning from Academic Dishonesty. Harvard University Press, 2013.

Lang, J. M. Small Teaching: Everyday Lessons from the Science of Learning (2nd ed.). Jossey-Bass, 2021.

Lovett, M. C., Bridges, M. W., DiPietro, M., Ambrose, S. A., & Norman, M. K. How learning works: Eight research-based principles for smart teaching (2nd ed.). Jossey-Bass, 2023.

Ryan, R. M., & Deci, E. L. “Self-Determination Theory and the Facilitation of Intrinsic Motivation, Social Development, and Well-Being.” American Psychologist, 55(1), 68–78, 2000.