AI in Teaching & Learning

This may include adding language into a syllabus or for individual assignments to address explicitly how and when students may use GenAI for successful assignment completion.

It may benefit instructors to review and revise their course or assignment-level learning outcomes to anticipate whether students will engage with GenAI and, if so, what they will learn from engaging with GenAI. Alternatively, instructors may want to revise or review their learning outcomes to clarify what skills or competencies they hope their students will gain by not using GenAI, emphasizing what students should be able to do in their courses independent of GenAI usage.

Certain course readings or materials may need to be updated to reflect changes in professional or disciplinary practices that have been affected by GenAI usage.

GenAI can be very effective at summarizing large swaths of information and generating output. However, GenAI output is not always accurate, and students may need to learn how to cross-check GenAI output with information from other sources, such as research databases and library-supported search engines.

Educators and students who've experimented with GenAI report a range of uses: walking through a concept a second way, generating practice problems, getting oriented on unfamiliar material, etc. Students themselves report mixed but often positive experiences: in a 2025 Inside Higher Ed student survey, about a quarter of students who use AI for coursework said it's helping them learn better, though half said it's having mixed effects on their critical thinking. Whether GenAI actually improves learning outcomes more broadly is still disputed among researchers: a 2025 Harvard study found real gains from a carefully-designed AI tutor in a physics course (Kestin et al., 2025), while another study found a significant association between heavy AI tool use and reduced critical thinking scores, attributing it to "cognitive offloading" (Gerlich, 2025). 

Some instructors have found narrower, more discipline-specific uses worth exploring. In computational courses for example, GenAI can help students debug code or work through syntax errors. Others have taken a different approach; they treat GenAI less as a tool to adopt and more as an object of study, having students critically examine AI outputs for bias, factual errors, or reasoning gaps as a media-literacy exercise. 

GenAI can also help multilingual and second-language writers strengthen arguments or catch grammatical issues, and some students describe using it that way. It lets them see what a cleaner version of an argument looks like, catch grammar or structure errors, or get unstuck when the mechanics of English are getting in the way of an idea they already understand well. For these students, the appeal usually isn't that GenAI does the thinking for them. It's that it lowers the friction between having something to say and actually getting it onto the page in a form that reads clearly.

Cheating and academic dishonesty come up often in conversations about GenAI. A 2026 study led by researchers at Berkeley's own Center for Studies in Higher Education surveyed more than 95,000 students across 20 public research universities and found that both GenAI use and GenAI-related misuse vary considerably by field of study. The study suggests this isn't one uniform problem with a single fix, and that what counts as appropriate use in one discipline, or even one assignment, may not translate cleanly to another. 

Some instructors are also concerned that students may lean on GenAI in ways that get in the way of learning rather than support it, whether that's submitting AI-written work as their own or using AI heavily enough that they never fully engage with the material themselves. As noted in the Opportunities section, the research on this is genuinely mixed. Some studies point to real learning gains under carefully designed conditions, while others report an association between heavy AI use and weaker critical thinking performance. Neither finding cancels the other out, and it's reasonable for an instructor to take both seriously while deciding what's right for their own course. 

Bias and hallucination are worth naming as well. These models are trained on very large amounts of existing text and images, and that training data can reflect real-world biases, including around gender and race, which the models can then reproduce in their output. They can also generate inaccurate information with a lot of apparent confidence, something often called "hallucination," and it isn't always obvious when this is happening. A few bigger questions also remain unresolved. Who owns AI-generated content? How copyright law applies to material used in training these models? What the environmental footprint of that computing looks like? And many more. All of these are all still active areas of debate, legally and otherwise. There's no settled answer to any of them yet, and it seems fair to be upfront with students that these are open questions rather than solved ones.