The Davis Family Library at Middlebury College in Middlebury in November 2025. File photo by Glenn Russell/VTDigger This story by Jing Williams was first published in the Addison Independent on June 25, 2026. MIDDLEBURY — Artificial intelligence is a powerful force in education today. It ca n accurately summarize novels, write essays, solve math problems and code — in seconds. And a popular narrative online now is that most college students use AI to automate their work and to cheat, Middlebury College assistant economics professor Germán Reyes said. Yet in their research, Reyes and colleague Zara Contractor discovered a more nuanced story about student AI use. They found that some uses of artificial intelligence only helped in the short term to boost grades, but using AI in a different way actually improved learning in the long term. Contractor and Reyes conducted a two-part research study to understand how students are using AI, what they think they’re learning from it and what they’re actually learning from it. The first part of their research was surveying Middlebury students from December 2024 to February 2025 about their AI use at school. They found 80% of Middlebury students use AI for academic work, but contradicting the common narrative, the majority use it as an augmentation tool rather than automation. Reyes describes augmentation as working with the tool to enhance learning, versus automation, where AI does the work for you.  To verify their data was truthful and applicable in other educational scenarios, the research team asked Anthropic, the company that developed the AI software called Claude, for data about how students with the college’s email addresses were using the technology. Anthropic confirmed: Most students were using AI for “technical explanations,” to explain concepts rather than automate work. Comparison to global data from over 50 countries also verified that college students aren’t, for the most part, using AI to just do their work for them. What the survey results revealed to Reyes is that higher education institutions need to understand how AI is being used by students before they create policies about its use. “If some students are using it to learn more and you decide to ban AI because you think it has no room in college to help students learn, then you’re inadvertently going to be harming students who benefit from the tool,” Reyes said. “So I think it’s important to have a good understanding of ground level facts to make policy decisions that are consistent with how students are using the tools.” Going deeper than just how students were using AI, Reyes and Contractor wanted to understand how AI impacts learning in the second part of their research, conducted in spring of 2025. They thought AI would have a negative impact on learning, a hypothesis supported by other papers published at the time. So Reyes was surprised again when the findings revealed a different story. Participants were randomly split into two groups: one with access to only online tools and one with online tools and AI. Then they researched a topic such as CRISPR gene-editing technology and wrote an essay on it. Around 70% of participants in the AI group adopted the tool and could use it however they wished. Participants came back a week later to write a second essay on the same topic and complete a test on it, this time with no tools.  What the researchers found surprised them: Those who had used AI to learn about the topic generally performed better on both essays, assessed by online graders, with a small difference on the first essay and a larger one on the second. A chart showing how often Middlebury College students use chatbots for various academic tasks. Courtesy of the Addison Independent. But it wasn’t simply using AI that determined how students did — it was how they used it, a factor Reyes and his colleagues hadn’t even considered. Those who automated AI to write their essays did much better than the others on the first essay, but much worse on the second one. Reyes said he and his colleagues went through the conversations students had with the AI to classify prompts as augmentation or automation. “Augmentation users had a much smaller effect on the essay in week one, but a much larger effect in week two, so one interpretation is, when you have access to the tool and use it to do the work for you, that provides large, short-term gains,” Reyes said. “But that comes at the cost of lower learning in the long run.” Although automation users did worse in the second essay, augmentation users did better on average than those who didn’t use AI tools at all. Reyes said an important lesson from the experiment is that the effects of AI largely depend on how students use the tools, explaining why similar studies can have such different outcomes. It also confirmed AI usage in higher education is more complicated than people might think. “I think many people tend to be a bit pessimistic about how students use the tool, but when we talk to students, they have a very nuanced and sophisticated view of the ways in which it can help or hurt.” There is no perfect policy that addresses all the concerns of the ever-adapting technology, Reyes said, but having data to back up policy decisions is extremely important. Conversations about AI often consider it in isolation, but Reyes believes the technology is deeply connected to students’ value of their education and factors like grade inflation. For example, if there is high grade inflation at an institution, “there is less value to augmentation because you need to differentiate yourself from all other students who also have a very high GPA. But how do you do that? Well, presumably by doing more activities … and how do you find the time for those other activities? Perhaps by using AI to automate some of the academic work.” Reyes said he and Contractor are finishing their paper on the experimentation portion of their study right now, but the research won’t stop there. He secured funding to conduct the survey again, planning to keep many questions the same, while adding new ones for 2026, such as whether students are paying for premium versions of AI. An economics expert by training, Reyes is involved in many projects about AI, from employing it as a personalized tutor in Peruvian high schools to studying how the technology affects the labor market. He said all the research he’s conducting overlaps in some way. “There’s all these interesting questions that everyone is interested in and there’s so much to do, so it’s a very exciting area,” Reyes said. “At the same time, it all feels a little bit urgent. There’s an inclination to join as many projects as possible, so it can be overwhelming, but it is certainly the case that a lot of my research agenda has been shifted to AI, partly because it has some of the most exciting questions of today.” Read the story on VTDigger here: Research at Middlebury College reveals nuanced story about artificial intelligence use. ...read more read less