Stephen Blessing holds the Janet R. Matthews, PhD Endowed Chair in Psychology at the University of Tampa, where he’s taught for 22 years.

I was into AI before AI was cool. Back in the early 1990s, I was a graduate student at Carnegie Mellon University, part of a team building something called Intelligent Tutoring Systems—AI programs that helped high school students learn algebra and geometry through active, responsive interaction. That work eventually spun out into a company, Carnegie Learning, and I joined as one of its early employees. When ChatGPT went mainstream in late 2022 and my academic colleagues started debating whether AI belonged in the classroom, my reaction was a little different: I’d been here before. The question that interested me wasn’t whether to allow AI in my course, it was whether we could use it the way those early systems did: to make students more active with their learning, not less. 

At its core, active learning means students doing something with the material (e.g., retrieving it, applying it, connecting it to what they already know) instead of simply absorbing it. Back in my Carnegie Learning days, we used to talk about the instructor’s role shifting from “sage on the stage” to “guide on the side”—less time lecturing at students and more time designing experiences that get them to do the cognitive work themselves. That might mean having students build associations between a new concept and something familiar, or chunking complex material into smaller, manageable pieces they can comprehend one at a time. Decades of research back this up: a landmark 2014 meta-analysis in PNAS found that active learning significantly boosts exam performance and lowers failure rates compared to traditional lecturing (Freeman et al., 2014). The catch has always been that active learning takes more time and more creativity to design than a straightforward lecture. That’s exactly where generative AI has changed the equation for me. 

Given my early interest in AI and education, you may not be surprised that my research lab is now looking at how students can be guided into using active learning techniques with generative AI. This work is heavily influenced by Stephen Kosslyn’s book, Active Learning with AI: A Practical Guide, which offers a wealth of prompts instructors can use to help students study effectively. Drawing on those prompts, we built a 3-hour, self-paced Canvas course we used in some Psych 101 sections. After surveying 137 students who completed it, the results were clear: on a 7-point scale, students strongly agreed they’d use an AI agent to study in the future (M = 6.05) and, more importantly, that they’d learned how to better use one (M = 6.04). 

In my own classes, I’ve built assignments that would have been difficult, if not impossible, without generative AI. In my Sensation and Perception course, I provide numerous examples connecting perceptual phenomena to the real world. Some of these connections are genuinely hard to find on your own. The McGurk Effect, for instance, shows up in things like lip-reading or the strange mismatch of a dubbed foreign film, but most students have never noticed these sorts of connections and wouldn’t make them themselves. As part of a new assignment, I give students a starting prompt to explore these connections with AI, and they use what they find as a launching point for their own broader exploration, turning what could be a passive assignment into a genuine act of discovery. 

For my upper-level cognitive science seminar, I’m developing an assignment in which students interview a generative AI about whether it believes it has intelligence or emotions, then dig into and critically examine its answers. The idea grew out of my own experiments doing exactly this: asking AI systems pointed questions about their own processes and pressing on the responses. My hope is that it becomes a remarkably rich way to get students engaging with hard course concepts—consciousness, intelligence, the nature of emotion—not as abstractions in a textbook, but as live questions they’re forcing an AI to answer in real time. 

AI has also helped me up my own game as an instructor. This past summer, I had Claude work through my lecture slides with me, slide by slide, to sharpen how I present material. The exercise clarified something I already believed but hadn’t fully articulated: using minimal text, a slide should show the conclusion, not the reasoning that gets you there. But that conclusion itself needs to be a real, complete, self-contained sentence, because that’s the part a student will need to re-derive everything else when studying later. Having to defend that philosophy to an AI, over and over, forced me to actually apply it consistently. 

AI has also changed how I approach assessment. Once you’ve settled on your own philosophy of what a good exam question looks like, you can hand a generative AI some examples and have it help generate more. I now have a prompt I share directly with students, alongside instructions, that lets them develop AI-generated questions as they study for an exam, while having a dialogue with the AI during their review. While I have not formally done an assessment of this, informal student feedback has been positive, with many indicating it helped them to study for exams. 

Of course, using AI to promote active learning has its caveats, some endemic to any educational technology, and some perhaps unique to generative AI. Looking back at past technologies that critics believed would be the death of education—calculators, computers, even the internet itself—generative AI has plenty of company. As with those, I don’t believe AI is a death knell for education, but we do have to be deliberate about how we use it.  

Learning still requires the student to do the hard, active work of acquiring the material, and it requires instructors to spend real time understanding a tool’s strengths and weaknesses before building assignments and assessments around it. In my Sensation and Perception assignment, for instance, I explicitly tell students not to copy and paste what the AI gives them. They have to back up their claims with three real sources, and I point them to tools like Consensus to find actual peer-reviewed research, because an AI itself might hallucinate a citation that doesn’t exist. The AI’s initial response is just the start of an idea. The real work is verifying and building it out, as students write a fuller exploration backed by real research. Generative AI is not, nor should it be, a shortcut for the student’s learning or the teacher’s teaching. 

Generative AI is just the latest tool in educators’ toolbox for enabling active learning, though certainly one of the most powerful. Thirty years ago, those early Intelligent Tutoring Systems showed me what was possible when technology helps students engage actively with material, one carefully chunked piece at a time. Generative AI lets us do that at a scale and flexibility I couldn’t have imagined back then. As with any tool, the responsibility falls on us as educators to use it well, to design assignments and assessments that put students to work, rather than let them coast. If you haven’t already started experimenting, I’d encourage you to pick just one assignment this term and ask: where could AI help my students do more thinking, not less?

Resources mentioned in this post:

• Stephen M. Kosslyn, Active Learning with AI: A Practical Guide 
• My generative AI assessment-question prompt
• Sensation & Perception real-world connections assignment
• Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H., & Wenderoth, M. P. (2014). Active learning increases student performance in science, engineering, and mathematics. Proceedings of the National Academy of Sciences, 111(23), 8410–8415. https://doi.org/10.1073/pnas.1319030111 

MEET THE AUTHOR