More than 80% of Middlebury College studentsUtilize generative AI for academic assignments, as indicated by a recent survey I carried out with my colleague and fellow economist Zara Contractor. This represents one of the quickest rates of technological adoption in history, far surpassing the previous ones.40% of U.S. adults have adopted it Forty percent of American adults have embraced it A 40% uptake among adult Americans 40% of the adult population in the United States has adopted it It is adopted by 40% of U.S. adults 40% of U.S. adults are using it The adoption rate among U.S. adults stands at 40% Forty percent of adults in the U.S. have adopted it 40% of American adults have taken it up It has been adopted by 40% of U.S. adultsand it occurred within two years of ChatGPT’s official release.
Subscribe to our newsletterfor the most recent science and technology news updates.
Even though we examined just one college, our findings correspond with other research, offering a growing understanding of how technology is being applied in higher education.
From December 2024 to February 2025, we conducted a survey with more than 20% of Middlebury College’s students—634 individuals—to gain insight into how students are utilizing artificial intelligence, andwe published our findings in a working paperwhich has not yet been reviewed by peers.
Our findings question the alarmist perspective on AI in higher education and imply that institutional policies should concentrate on the manner in which AI is applied, rather than whether it should be prohibited.
More than just a study tool
In contrast to sensational news reports claiming that “ChatGPT has decoded the entire academic project.” and “AI Plagiarism Is Becoming More PrevalentWe found that students mainly use AI to improve their learning instead of to skip tasks.
When we inquired of students regarding 10 distinct academic applications of AI—from clarifying ideas and summarizing texts to editing, generating code, and even composing essays—explaining concepts was the top choice. Students often referred to AI as an “on-demand tutor,” a tool that proved especially useful when office hours were not accessible or when they required quick assistance during late-night hours.
We categorized AI applications into two categories: “augmentation,” which refers to uses that improve learning, and “automation,” which involves tasks that generate work with little effort. Our findings showed that 61% of students who use AI utilize these tools for enhancement, whereas 42% use them for automation activities such as writing essays or creating code.
Even if students utilized AI to handle tasks, they demonstrated discernment. In open-ended answers, students shared that when they automated work, it was typically during busy times such as exam week, or for low-importance tasks like formatting references and writing standard emails, rather than as their usual method for handling significant academic work.
Of course, Middlebury is a small liberal arts institution with a comparativelymajority of affluent students. What about other locations? To discover, we analyzeddata from other researchersspanning over 130 universities in more than 50 countries. The findings align with our Middlebury research: internationally, students who utilize AI are more inclined to use it to enhance their studies, rather than to automate tasks.
But should we believe what students say about their use of AI? A clear issue with survey data is that students may understate uses they consider improper, such as using it for writing essays, while overstating proper uses like seeking explanations. To confirm our results, we compared them withdata from the AI company Anthropic, which examined real-world usage habits from university email addresses associated with their chatbot, Claude AI.
Data from Anthropic indicates that “technical explanations” are a primary application, consistent with our discovery that students frequently use AI to clarify ideas. In the same way, Anthropic identified that creating practice questions, revising essays, and condensing content make up a significant portion of student activity, which corresponds with our findings.
In other words, the data from our self-reported surveys aligns with real AI conversation records.
Why it matters
As writer and scholar Hua Hsurecently noted, “There are no accurate statistics on how many American students utilize AI, just anecdotes about how everyone is doing it.” These narratives often highlight extreme cases, such as a Columbia student who used AI “to falsify almost all assignments.”
However, these stories might mix broad usage with total dishonesty. Our data shows that AI is indeed widely used, but students mainly employ it to support their learning, not to substitute it. This difference is important: By labeling all AI use as cheating, sensational reporting could encourage academic misconduct, causing responsible students to feel foolish for adhering to rules when they think “everyone else is doing it.”
Additionally, this misleading image offers one-sided information to university officials, who require precise details regarding real student AI usage habits in order to develop successful, data-driven policies.
What’s next
Our results indicate that extreme approaches, such as complete bans or total freedom, come with potential dangers. Restrictions might unfairly affect students who gain the most from AI’s instructional capabilities, while giving an unfair edge to those who violate the rules. However, allowing unrestricted use could lead to harmful automated practices that may hinder the learning process.
Rather than implementing uniform policies, my research suggests that institutions should concentrate on assisting students in recognizing advantageous AI applications from those that could be detrimental. Regrettably, the study of AI’s real effects on learning is still in its early stages—there are no studies I know of that have thoroughly examined how various forms of AI usage influence student academic results, or if the effects of AI might be beneficial for certain students while harmful for others.
Until such evidence is present, anyone concerned about the impact of this technology on education must rely on their own discretion to assess how AI can enhance learning.
This piece is reprinted fromThe Conversationunder a Creative Commons license. Read theoriginal article.
Provided by The Conversation
This narrative was first released onMuara Digital Team.
