AI and Learning

This page highlights emerging findings in educational research on AI and learning, along with their implications for teaching.

One idea to keep in mind as you read: unguided AI use can pull students into a vicious cycle that harms their learning. Instead of spending time trying to learn, some students turn to AI right away, using it as a magic answer box. They skip the productive struggle of working through a problem themselves, their skills don't develop, and they rely on AI even more the next time. Each pass through the cycle makes the next one more likely. Our Course & Assignment Design page offers practical strategies for applying these findings, and breaking the cycle, in your own teaching. 

What the Research Shows

Finding 1: AI can feel like learning, without being learning

GenAI tools are good at sounding fluent and confident, and that articulateness is easy to mistake for understanding--what researchers call the illusion of competence (Lodge & Loble, 2026). But learning requires effortful work and taking the time to struggle with material before it sticks (Bjork, 1994; Bjork & Bjork, 2011).

GenAI makes it tempting and easy to skip that effort entirely. Its outputs look finished and correct, creating a feeling that one has learned, but without doing the thinking. That's cognitive offloading: handing off work that was supposed to build a skill, not just produce the output (Risko & Gilbert, 2016). Some tasks are fine to offload and can save time, but educators should be wary of students offloading the learning itself.

What this means for your teaching: Build in moments to see students' process, not just their final product, through short explanations in their own words, drafts, or quick checkpoints. That's often where you can tell whether the thinking behind an assignment actually happened.

Finding 2: AI only helps with clear scaffolding, otherwise it can deskill

Guided AI use, built around clear learning goals and checkpoints, produces meaningfully better outcomes than open-ended, unguided use (Qu et al., 2025; Ma & Zhong, 2025; Wu et al., 2026). The main idea here is that if AI is used as a tool with clear structure, modeling, and guidance, it can support learning.

What this means for your teaching: If AI is part of your course, pair it with clear guidelines and room for students to reflect on their process, so they're still putting in the effort and work needed to build the skills you want them to learn.

Finding 3: AI's impacts on learning are inequitable 

Students with weaker prior knowledge or metacognitive skills are most vulnerable to the vicious cycle described above (Darvishi et al., 2024). Cognitive offloading is most tempting exactly where foundational skills are thinnest, so unstructured AI use tends to widen existing gaps rather than close them (Gerlich, 2025). This can especially harm students from under-resourced communities.

What this means for your teaching: Students bring different levels of prior knowledge and skill into your course. Keep that diversity in mind as you set expectations and guidance around AI use.

Finding 4: 'Every student is using AI' is a myth

Students come to class with existing habits and strong opinions about AI already formed. And they are not a single, homogenous group. The most dominant theme in student responses to the AI-related questions in UO's 2025 Rising and Flourishing Together (RAFT) survey was a call to ban or strongly restrict AI in coursework. TEP's survey data shows a similar heterogeneity: roughly a quarter of UO students surveyed placed low value on AI and were highly aware of its limits (skeptics), roughly another quarter used it readily and were least equipped to catch its errors (enthusiasts), and just over half decided whether or not to use AI on a case-by-case basis, depending on the situation (pragmatists) (TEP GenAI CAIT 25-26). Instructors have multiple student perspectives in the room: while some don't engage with AI, some use it pragmatically, and some use it regardless of your course policy.

What this means for your teaching: Your classroom likely holds a mix of skeptics, enthusiasts, and pragmatists. Keeping these different student profiles in mind can help you make your teaching more responsive to the range of students actually in the room.

Finding 5: Open conversation with students helps

Only 1 in 10 UO students interviewed for TEP's GenAI CAIT project felt comfortable asking an instructor to clarify their policy on AI use (TEP GenAI CAIT qualitative interviews). Both students who said they make values-based decisions about when to use AI, and students who make context-based, case-by-case decisions said they respond best when an instructor explains their reasoning, not just their rule. Students are working through the same uncertainty instructors are, and open, honest conversation about it does more than a policy statement alone.

What this means for your teaching: Students are more likely to follow a policy they understand the reasoning behind than one they're just handed. Explaining your “why” is meaningful.

Finding 6: Faculty responses to GenAI are evolving

At one large U.S. research university, AI policy across 31,000 syllabi has shifted steadily away from blanket restrictions and toward task-specific guidance between 2022 and 2025 (Chirikov, 2026). And those syllabi are communicating less about AI and academic integrity and more about AI's impacts on learning. These findings are a single-institution signal, not a confirmed national trend, but point at something real: faculty are adjusting their approaches as AI tools evolve and research on AI's impacts on learning develops.

What this means for your teaching: Your policy doesn't need to be right on the first try, and you're not alone in revising it. Treat it as a living document, not a one-time decision.

Conclusion

In many studies, two things show up as protection against skill erosion: AI literacy, a conceptual understanding of AI, not a technical skill (Chiu, 2025), and metacognition, a student's habit of actively noticing their own thinking. Our Student Metacognition Guide, coming soon, gives students a direct way to practice metacognition themselves.

Taken together, these findings point to the same throughline: thoughtful course and assignment design determines whether AI use supports learning or undermines it. Course & Assignment Design turns this research into concrete teaching moves you can put into practice in your own course. The UO Faculty Examples pages show what that looks like in practice, with real examples from instructors across the UO community already doing this work.

Are you researching AI and learning?

If you're doing your own research on AI and teaching, or you know a colleague with findings worth sharing, let us know. We'd love to spotlight it here.

References