AI in Teaching: What the Science of Learning Already Tells Us

Few topics in higher education generate as much debate right now as the role artificial intelligence in all aspects of teaching and learning. Faculty listservs, teaching centers and national conferences are awash in urgent questions: Is AI undermining learning? Is it encouraging cheating? Is it decreasing critical thinking skills? Should it be banned, regulated, or embraced? Students, meanwhile, are already using AI tools extensively, and often more extensively than their instructors realize.

What is striking about the current discussion is how much of it is driven by perceptions, anecdotes and survey-based opinion data, rather than by a robust body of evidence demonstrating how AI actually affects student learning. That gap matters. But it should not leave us paralyzed. While we wait for the sound scholarship that can answer those questions directly, we already possess something powerful: decades of well-established psychological science on how students learn. That foundation offers clear guidance for how AI can and cannot support learning in the college classroom while we gather data that allow us to have more specific recommendations for the subtle uses of AI in the classroom.

What We Think We Know About AI in Higher Education

Recent surveys paint a picture of widespread AI use paired with deep unease among both faculty and students. Faculty report high levels of concern that generative AI undermines original writing, critical thinking and academic integrity. Students report frequent use of AI for brainstorming, studying and tutoring-like support, but also express uncertainty about whether it is helping or hurting their learning.

These perceptions are not trivial. They shape institutional policy, classroom norms and faculty behavior. But perception is not the same as evidence. Most of the data driving these conversations are self-reports, which serve as, in some cases, knee-jerk snapshots of attitudes or usage patterns rather than studies that isolate AI’s causal impact on learning. Even large-scale surveys cannot tell us whether students who use AI learn more, learn less, or simply learn differently.

The existing empirical literature on AI in higher education is growing, but it remains uneven. Much of it focuses on engagement, efficiency, or attitudes rather than durable learning outcomes. Many studies examine short-term effects, specific tools or highly contextualized implementations. As several recent systematic reviews have noted, the field lacks a coherent, replicable body of evidence that allows us to draw confident conclusions about AI’s effects on deep learning across disciplines and institutions. In other words, the confidence with which we talk about AI’s educational harms or benefits often exceeds what the research on AI can currently support.

Waiting for Evidence Does Not Mean Waiting to Act

Acknowledging the limits of the existing scholarship does not mean educators should stop making decisions about AI. Classrooms are not laboratories that can be placed on pause. Students and faculty are using these tools now, whether or not instructors design work that includes acceptable AI use. To provide guidance for faculty and students, many institutions are scrambling to write policies now, whether or not the evidence base is complete. The question, then, is not whether to act, but what knowledge we should rely on while we wait for more direct evidence.

Here is where psychology offers a crucial contribution. Long before generative AI entered the classroom, psychological science had already identified core principles of learning that are remarkably stable across contexts, technologies and generations of students. These principles do not tell us exactly how to use every new tool. But they do tell us what must be true for learning to occur.

What Psychological Science Tells Us About Learning

Decades of research converge on several foundational insights. First, learning requires a great deal of cognitive effort. Students learn more when they actively retrieve information, generate explanations, and struggle productively with material. Tools that replace thinking rather than supporting it are likely to impair learning, regardless of whether they are textbooks, calculators or AI chatbots.

Second, feedback matters, but its timing and nature are critical. Feedback is most effective when it is specific, explanatory and connected to opportunities for revision. AI systems can generate feedback quickly, but speed alone does not guarantee quality or learning value. Feedback needs to respond to the subtle context of a course, and often, AI cannot do that in the same manner that a faculty member can.

Third, metacognition, meaning students’ awareness of their own understanding of content and skills as well as their own thinking, plays a critical role in learning. When tools create illusions of understanding or fluency, students may feel confident without actually mastering the material. This risk is not unique to AI, but AI can amplify it if used improperly. None of these principles are controversial. They are among the most replicated findings in the learning sciences. And importantly, they are technology- and discipline-

Using Psychological Science as a Design Filter for AI

Rather than asking whether AI is “good” or “bad” for learning, faculty and institutions would be better served by asking a different set of questions grounded in psychological science:

• Does this use of AI increase or decrease students’ cognitive engagement with course material?
• Does it support formative feedback and revision, or does it short-circuit those processes?
• Does it promote metacognitive awareness, helping students understand what they know and do not know?

When AI is used to scaffold learning by prompting students to explain their reasoning, offering feedback on drafts that students must then revise, or modeling problem-solving strategies, it can align well with what we know about effective learning. When it is used to generate final answers, bypass practice or obscure students’ own thinking, it violates those same principles. In short, when AI sidesteps effort and leads to a strong product, the process suffers, and that process is where learning happens. The distinction is not subtle, and it does not require waiting for a new generation of AI-specific studies to recognize.

A More Productive Path Forward

Higher education does need more rigorous scholarship on teaching and learning focused on AI and learning outcomes. We need studies that move beyond attitudes, perceptions, and adoption in order to examine transfer, retention and long-term skill development. That work will take time, and it should.

But in the meantime, we should resist the temptation to let uncertainty push us toward extremes whether uncritical enthusiasm or blanket prohibition. Psychological science offers a middle path: a way to evaluate AI practices against what we already know about how learning works. If faculty and institutions anchor AI decisions in these foundational principles, they can act thoughtfully now while remaining open to revising their approaches as the evidence base grows. In a moment defined by rapid technological change, that combination of scientific humility paired with theoretical grounding may be one of higher education’s greatest strengths.

Beth Schwartz, PhD, a cognitive psychologist by training, is the Senior Director of Precollege and Undergraduate Education at the American Psychological Association (APA). Beth came to APA after 30 years of experience in higher education as a professor of psychology and a chief academic officer. Drawing on her expertise in cognitive and behavioral science, her current work focuses on advancing education through evidence-based teaching, with expertise that helps educators understand the connection of learning science principles to pedagogical strategies, including how AI impacts student learning. Through workshops, presentations, and publications, Beth collaborates with educators and disciplinary leaders to translate research into practical strategies that improve learning, equity, and well-being. Beth is leading efforts at APA to provide educators with guidance on how to conduct scholarship that addresses the impact of AI on student learning to help develop sound research that fills this gap in the literature, and to help educators use learning science findings when deciding when and how AI can be used in their classrooms. 

Chris Hakala has over 20 years’ experience as a professor and Center Director in higher education institutions. Over the years, his research has focused on reading comprehension, teaching and learning, effective faculty development and assessment. In addition, Chris has been invited to present at many conferences around the country as well as dozens of colleges and universities on topics ranging from reading narrative text to how to effectively manage large classes, or how to engage students in ways that maximize student learning. At Springfield, Chris runs the Center for Excellence on Teaching, Learning and Scholarship and teaches courses in the psychology department.