Kathryn Palmer’s recent Inside Higher Ed article “The ‘Nonexistent’ Research on AI’s Benefits for Education” raises an important concern. Colleges and universities are investing heavily in generative artificial intelligence even though evidence about when, how or whether these tools improve student learning remains limited. Some studies even suggest that unguided use could undermine learning.
We should take that uncertainty seriously. However, we would argue that the limitations of the research base on generative AI do not mean we are without relevant and useful evidence. Research on generative AI may be new, but the psychological science of human learning is not. (See, for example: Benassi, Overson and Hakala, 2014; Lang, 2021; Overson, Hakala, Kordonowy and Benassi, 2023; Zakrajsek, 2022.)
Decades of research in cognitive psychology and the science of learning tell us a great deal about the conditions that support durable knowledge, critical thinking and problem-solving. That research cannot tell us whether every new AI tool will improve learning, but it gives educators a rigorous and well-established framework for deciding when AI is likely to support learning and when it may interfere with it.
Psychologists know that learning requires thinking. Students learn by retrieving information, organizing ideas, connecting new material to existing knowledge, explaining their reasoning, recognizing gaps in understanding, correcting misconceptions and applying knowledge in new contexts. In short, students learn by engaging in targeted cognitive effort. The question, therefore, should not simply be “Does AI work?” A better question is “Does a particular use of AI strengthen the cognitive processes necessary for learning, or does it replace them?”
These questions are not unique to higher education. The American Psychological Association’s recently released expert report on children’s and adolescents’ learning with educational technology examines them in K–12 settings, and the same principles apply to college students.
One of the challenges Palmer identifies is that AI systems are evolving faster than traditional educational research can evaluate them. The article cites one expert’s recommendation to study features and uses across tools rather than attempting to establish whether each individual product “works.” Learning science provides a way to do that and shifts attention from technology adoption to instructional design and learning goals. Rather than starting with an AI tool and asking how to incorporate it, faculty should begin with a learning goal and ask:
- What must students know or be able to do?
- What thinking must they practice to achieve that goal?
- Which cognitive work must remain with the student?
- Is AI extending that thinking or doing it for them?
- Can students demonstrate the learning independently?
Learning science helps identify conditions under which AI use may support learning and conditions under which it may not. Students might, for example, produce an initial explanation or solve a problem independently and then use AI to generate questions that push their reasoning further. They might complete a first draft and then ask AI to identify an unclear claim, missing evidence or a potential counterargument. Students would subsequently decide which feedback to accept or reject and explain their reasoning.
Indeed, the research specifically demonstrating generative AI’s educational benefits remains limited. However, there is no doubt that AI is transforming the learning environment and foundational psychological science on learning provides the guidance needed to help design and evaluate the ways that generative AI impacts student learning.
Institutions should support smaller, iterative studies of specific AI uses. They should examine student work before and after AI assistance, assess whether students can transfer their learning to new situations without AI, and study how results differ across disciplines, levels of expertise and student populations. But educators do not need to wait decades before developing responsible approaches. We are not starting from zero. AI is new. The psychology of human learning is not. Until the AI-specific evidence catches up, the science of learning should guide the way forward.
Leave a Reply