How do you measure learning? Is it the object produced at the end—the term paper, correct test answer, or final presentation? Or is the act of production itself—the revisions, the scribbles toward solving the equation, or the presentation drafts discarded? If you use refrigerated chocolate chip cookie dough to bake cookies, have you learned to make chocolate chip cookies or just produced the cookies? These are the questions our sector is currently grappling with since the introduction of AI a few years ago. In this week’s issue, we look at how one edtech company proposes to address these questions and discuss what that means for higher education.  

Is It Actually Clarity? 

From To Catch AI Cheating, Turnitin Wants to Monitor Students’ Keystrokes | The Chronicle of Higher Education 

To detect AI use in student writing, Turnitin has released Clarity, which monitors student typing, revisions, use of the copy-and-paste function and time spent writing.  

Our Thoughts  

Ick! Yuck! Honestly, those were my first reactions when I read this piece. I’m not going to say that higher education is currently not challenged with assessing learning due to AI because that would be untrue. However, I’m not willing to concede that keystroke logging is the answer because the only question it answers is “did the student type this?” It doesn’t actually tell us whether the student understood the assignment, if the prose is actually theirs, or if they are actually learning anything. Yes, I know you’re probably reading that and saying, “But Allen, them submitting a term paper doesn’t answer those questions either!” You would be correct and maybe that’s my point. We’re simply trying to find the next best AI-proof alternative to our original assignments without facing the reality that perhaps we struggled to accurately assess or measure learning in the first place.   

Before I go further, it's worth being precise about what the tool does, because "keystroke logging" undersells it. Clarity is a writing space inside Turnitin's Feedback Studio, and while a student drafts in it, the software captures the keystrokes, the deletions, the pauses, the paste events, and the total time on task, then assembles all of it into a replay that an instructor can scrub through afterward, watching sentences appear, get cut, and reappear. Now, I might argue that if instructors feel they don’t have time to redesign assignments, they might not have time to watch the drafting videos for 100 students in their course. But that would be dismissing what is a very real problem for faculty across the sector, who are looking for alternatives for poorly conceived and proven ineffective AI detectors. Turnitin is at least trying to provide something new to meet this challenge.  

However, I still think there are other ways. Interestingly enough, Clay Shirky made a similar argument the day before the Turnitin article ran. His observation is that our pre-AI assignments didn't work because we designed them brilliantly; they worked because the alternatives to actually doing the assignment were inconvenient, expensive, or easy to catch, which meant we could treat what students handed in as a rough proxy for the effort they'd put in.   

But Shirky’s larger point was not to disparage these proxy assignments that AI broke. Rather, he wanted to highlight that AI in the classroom is a wicked problem, meaning one that can't be pinned down cleanly, that looks different depending on who you ask, and that has no finish line where you get to declare it handled. His solution is not a campus-wide policy, since he doesn't think one is possible. Instead, it’s disciplinary variation, faculty forming informed opinions grounded in their own expertise, spelling those out assignment by assignment, and then actually comparing notes about what worked and what didn't. I might add including some folks from your teaching and learning center in the conversation, but even still, I agree with him. This is what higher education does; it experiments and offers a variety of solutions. We don’t wait for knowledge to come to us. We are the knowledge makers.  

Shirky concludes his piece by saying that waiting around for the confusion to sort itself out is an abdication, and I would extend his argument to Turnitin’s Clarity software. It, too, is an abdication. Not because surveillance is uniquely sinister (well, maybe), but because it ends the questioning. If the institutional answer to how we assess learning is that we watch students type, there's nothing left to experiment with, and we’ve ceded our expertise and ability to address this wicked problem to yet another piece of educational technology. It also effectively ends the conversation about how we could better assess learning because someone would reasonably ask why we’re still debating this problem now that we have this solution. 

Additionally, while I support Shirky’s ideas for experimentation, I am concerned that there’s a gap in our infrastructure for disseminating that knowledge more broadly. We have built enormous machinery for announcing what worked: conference sessions, teaching center showcases, journal articles, campus newsletters, and a reliable stream of vendor case studies. We have almost nothing for circulating what a colleague tried in good faith and then abandoned. For example, take the faculty member from this Chronicle piece who tried a new technique to help his students learn while combating AI. He moved to in-class writing but found the format didn’t work as he expected because it gave students too little time to develop an argument and him too little text to respond to in a way that would help students grow as writers. His experiment is a finding, but beyond that article, he would have little way to distribute these ideas beyond his campus or maybe even beyond his academic department. Without a way to distribute and discuss these ideas more broadly, we’re left with institutions independently rediscovering the same dead ends every semester.  

Finally, I would be remiss if I didn’t put my registrar hat back on for this. If I were in the room and purchasing Clarity was being discussed, I’d be asking myself if these recordings, which Turnitin calls Writing Process Data, are educational records under FERPA. If so, what protections are required once they are created? How long do we have to keep them? Can a student request their own? If academic dishonesty charges are brought against a student using these recordings, do they become part of the conduct file? What happens to this data if we try it for a year and then decide not to renew? While these details may seem trivial, they’re an important part of institutional record keeping and it’s better to decide them now before you have a legal reason to understand your answers.  

Ultimately, I think solving this problem falls to each individual working in the classroom or supporting teaching and learning at your institution. If you tried something last year that didn't work, take it to your next department meeting, or to your teaching and learning center, or write the conference proposal that isn't a success story. Turnitin will happily fill the silence if we leave it there, and the more we talk to each other about what failed, the less appealing a per-seat license looks by comparison. If we do that often enough, the next person who tries something...anything...to address learning in the age of AI will at least start from a stronger foundation of what has and hasn’t worked so far. We may not all be giants, but that doesn’t mean that someone can’t stand on our collective shoulders for a slightly better view.   

Sparks 

• Trump Officials Try Again to Reform Colleges, With Higher Ed’s Help (Inside Higher Ed) - Developed with input from the Association of American Universities and other university leaders, the Trump administration tries a new tactic to reform higher education. I have hopes that this is the beginning of better conversations between the current administration and higher education...maybe.
• Hey, Claude, Should I Go To College? (Education Next) - Michael J. Petrilli, president of the Thomas B. Fordham Institute, looks at how AI answers questions about college enrollment. This is an interesting read with good statistics that also forces you to ask whether you’ve been as honest as you should be with people considering postsecondary education.
• 7 ways college admissions season is changing this year (The Washington Post) -  As the admissions cycle for Fall 2027 opens, The Washington Post looks at seven ways college admissions is changing for better or worse. If you’re looking for an article that provides a quick summary of the differences we can expect during this cycle, this is a good place to start.