Colleges grapple with AI use in the STEM classroom amid uncertain impacts

The Wild West of artificial intelligence in the classroom may soon be seeing regulation in the US.

Rep. Suzanne Bonamici (D-OR) introduced a bill Sept. 22 aiming to regulate the technology across all levels of education. Dubbed the Artificial Intelligence Education and Workforce Readiness Act, the bill aims to establish safeguards against potentially damaging use of AI in schools by requiring risk assessments for classroom technology, preventing the use of student data to train AI models, and funding research on the effects of AI use on education.

“AI is already shaping how students learn and how people earn a living, but federal rules and regulations to protect students, support workers, and enforce ethical standards do not exist,” Bonamici writes in a press release announcing the bill. “We need clear rules and accountability for how AI is used in classrooms and workplaces.”

Bonamici isn’t the only one who thinks so. In fact, in the 2025–26 state legislative session alone, 146 bills pertaining to particular aspects of AI have been introduced, suggesting an increase in worry among policymakers. Up to now, regulating AI use in the classroom has been mostly left to local school districts and individual universities.

For example, at the local level, the two largest school districts in the US, the Los Angeles Unified School District and New York City Public Schools, recently implemented AI moratoriums, banning students in certain grade levels from using generative AI platforms.

Other school districts, in contrast, are starting to test out AI-based tutoring tools. The Aldine Independent School District in Harris County, Texas, for example, introduced Amira, an AI platform whose makers claim it helps students improve their reading skills by analyzing their accuracy and pacing to give them personalized feedback.

Meanwhile, in higher education, which Bonamici’s bill would cover, AI policies have been largely left up to faculty. According to one survey (PDF) conducted by the American Association of Colleges and Universities (AAC&U) and Elon University in November 2025, 87% of the 1,057 surveyed faculty said that they had created their own AI policies, but only around half said an institution-wide policy existed at their schools.

“Sometimes the policies are just, ‘Ask your faculty member. See what they have on their syllabi,’ ” says C. Edward Watson, coauthor of the survey report and vice president of digital innovation at AAC&U.

According to one collection of AI policies for college-level courses across different schools, these individual classroom policies can run the gambit. They vary widely based on factors that include the AI literacy of a professor and their personal feelings toward the technology. Classroom AI policy can also depend heavily on what a professor sees as a “legitimate use” of generative AI tools, Watson says. “It varies not just from one discipline to the next, but from one faculty member to the next in the same department.”

Regardless, as governments and educators try to figure out how generative AI fits into the classroom, students are likely already being affected by the technology. The jury’s still out on whether that impact has so far been positive, negative, or somewhere in between.

“This is not a story of simple resistance to change,” writes Lynn Pasquerella, AA&U president, in the forward to the organization’s survey report. “It is, instead, a portrait of a profession grappling seriously with how to uphold educational values in a rapidly shifting technological landscape.”

Talking to students about AI harms

Anecdotally at least, some professors tell C&EN that the main issue they’re seeing with AI tools is their students overrelying on them, which early studies suggest could shrink critical thinking skills and diminish motivation. That observation is also backed by surveys such as the 2026 Digital Education Council’s AI in Higher Education Global Survey, where one-fifth of surveyed students said they found it harder to work through problems without the help of AI.

In John Trant’s chemistry classrooms at Windsor University, this dependence on generative AI is especially apparent whenever he gives out assignments that have a heavy writing component, such as laboratory reports. “They get a whole lot less writing in high schools now than they used to,” Trant says. “So, they’re a whole lot more uncomfortable with it.”

Prior to the end of 2022—when large language models first became available to the public—most students would inevitably struggle through these assignments. In the process, they would ideally improve their writing skills. Alternatively, some students would resort to cheating and pay someone to write up their lab reports on writing service websites. But the cost of these services seemed to act like a big enough barrier that it was never too widespread a problem, Trant says.

AI platforms such as ChatGPT, on the other hand, are free, at least on some levels. They’re also extremely easy to use. With a simple prompt, a student can ask an AI application to generate a detailed outline to use in writing their lab or other school report. They can even ask an AI tool to write the entire thing.

In either case, the students have offloaded at least part of the task to AI, Watson says. They shortcut “all that productive struggle that’s associated with the process of learning.”

“Cognitive offloading” was a phrase that came up over and over again in C&EN’s conversations with professors. It also appeared in the responses to the AAC&U survey. That said, AI isn’t the first technology for which offloading has been a concern—the same thing was often brought up as calculators became commonplace in schools.

But generative AI feels different, says Chuck Pearson, a professor of chemistry at Tusculum University.

Pearson first noticed generative AI infiltrating his classroom a couple years back when, all of a sudden, his carefully crafted problem sets were taking students minutes, not hours, to finish online. More answers than usual were also being submitted correctly on the very first attempt.

“Something technological had to be helping,” he wrote in a blog post about the experience.

Pearson tested this hunch by plugging one of his problem set questions into ChatGPT. He got a nearly perfect answer in return. AI use “doesn’t just remove small elements of the problem-solving practice from our students,” he says. “It removes the entirety of the problem-solving practice.”

To Pearson and many other educators, learning how to problem solve is one of the most valuable benefits of a college education. “We teach problem-solving in a more structured form at the undergraduate level so that students can deal with the unstructured problems they’re going to encounter in research and in the workforce,” Pearson says.

He’s begun emphasizing this idea to his students. “You have to make a decision as a student that the process of learning is worth putting aside [generative AI] initially,” he says.

But according to Scott Reed, a chemistry professor at University of Colorado Denver, and the founder of ChemIllusion, a company that offers generative AI–based image tools for chemists and chemistry educators, there is a time and a place where offloading part of a task to technology can help students learn. The tech assist just has to be employed on a task that doesn’t involve building new skills, he says. For example, as he tells his students, performing mental arithmetic might distract them while they are trying to learn how to do a more complex math problem. An AI tool can help eliminate that distraction. “I like giving them the options and the empowerment to make their own choices,” Reed says.

Having these types of discussions with students is precisely the advice that Watson gives to faculty, especially given that AI’s still a relatively new technology and classroom policies about its use vary widely across different schools, and even across a student’s course load in the same school.

“Lean into the conversation,” Watson says. That requires not only making it clear what the rules are around using AI tools in a specific class but also explaining the reasoning behind a particular decision, he adds.

When policies haven’t been properly articulated, students feel like “it’s the Wild West,” Watson says, based on what he’s learned from student focus groups. According to a 2025 survey by Gallup and the Lumina Foundation, 52% of nearly 4,000 associate and bachelor’s degree students said that at least some of their classes don’t have clear policies on AI.

These same students were also more likely to say that they weren’t getting enough education about the technology.

“Part of AI literacy that students should learn before they leave college is an understanding of all of the reasons why they might want to reject AI, but [they should] also understand how to use it, and also recognize how it’s influencing careers and the career path that they are currently on,” Watson says.

AI is already reshaping the chemical sciences, with new tools coming out all the time. According to a report published by the United Nations Educational, Scientific, and Cultural Organization (UNESCO), all AI users will need technical knowledge, such as how to write AI prompts, and the ability to evaluate AI outputs and judge any ethical concerns that pop up when training AI models.

“We need the most creative and most dynamic thinkers in the face of this new technology,” Pearson says. “They are prepared [to use AI] because they’ve learned how to think in the first place.”

Weighing the good over the bad

Perhaps what makes conversations around AI in the classroom especially difficult is that there are often two camps that garner the most visibility. Those like Trant are very open about being against using the technology, at least against using it uncritically. Then there are professors such as Reed who are cautiously embracing AI and foresee a future where its use is ubiquitous in education.

The reality, however, is that most educators lie somewhere in the middle. Although over 90% of the AAC&U survey respondents expressed concerns about students’ overreliance on AI tools and their potentially diminishing critical thinking skills, 61% also said that they saw the benefits of AI use when it comes to enhancing and customizing learning.

An interest in personalized learning is what led Reed to publish an organic chemistry textbook generated using AI. Unlike standard textbooks, which Reed says students barely use these days, his textbook offers a more personal experience. For example, there’s an AI chat interface that can generate additional examples if the student wants more practice on a particular topic. The textbook can also let students and instructors fine-tune the content, allowing them to change the depth, depending on what they need to learn.

All these features are completely optional, Reed says. “Even though everything I’ve built was built using AI, I understand that students might prefer to avoid using it directly.”

Given AI’s propensity for outputting inaccurate information, Reed says that the majority of people who have come across his textbook have been, at best, skeptical of its quality.

He understands the raised eyebrows. “It’s really easy at a first pass to get lots of chemical errors,” he says. But the method he used to create the book, which he plans to describe in an upcoming paper, let him iteratively correct those errors. On top of that, he also got feedback from human reviewers.

“I can’t promise there’s zero errors in the 600-page document, but I would put it on par with any textbook I’ve ever reviewed in the past,” Reed says.

It’s too early to tell whether Reed’s organic chemistry textbook can improve learning outcomes, but other researchers have also been exploring the benefits of AI use in the classroom. One systematic review of 19 studies on AI use in science, technology, engineering, and math (STEM) education found that AI can increase student interest in a topic by tailoring content to different learning styles.

The sample size is small, and more research is needed, the researchers admit. But the findings are promising.

For now, it’s impossible to know how the future of education is going to look as AI continues to rapidly evolve. Rather than living up to its hype or bringing the downfall of critical thought, it will probably be “a little bit from column A and a little bit from column B,” Watson says.

But he warns that educators shouldn’t wait until the dust settles to figure out what to do about AI. “The dust may not settle,” he says. “You have to lean into the moment and do the best you can.”

Pearson agrees, adding that it’s OK if one’s view of AI changes over time. His opinion has. “I think that we need to be reflective on the implications of how different students, academics, and professionals are interfacing with this technology, so that we can refine our own responses over time,” he says.

“The technology is going fast, and we’re academics,” he adds. “If we’re skilled at anything, it’s going slow. I don’t think that’s entirely a bad thing.”