This semester, I enrolled in Visiting Fellow in International and Public Affairs Dany Bahar’s course IAPA 1002: “Foundations of Policy and Governance.” Before the class began, I read the first line under the course’s artificial intelligence policy: “Artificial intelligence is now a fact of professional life, and students will be expected to use AI tools as fluent collaborators throughout the semester.” I, like many of my peers, had to read the line again. It was the first time in my education that I had seen such a forward-looking AI policy, and, after years of using it on my own, I was excited to finally have a class that intended to use it deliberately.
For others in the course, this unconventional approach has been controversial — three weeks ago, The Herald reported some students’ pushback against Bahar’s requirement of AI usage. Those criticisms might be rooted in a valid nostalgia for analog learning, but they neglect the reality that AI has become a fixture of both the workforce and daily life — and an education which ignores that fact can no longer deliver what the traditional model once did. Any Brown policy that prohibits or neglects AI fails to recognize that students will turn to it regardless of what the syllabus allows. Refusing to engage with it hurts both parties — students lose the discipline-specific guidance they need to use AI well, which undercuts their learning, and professors forfeit any insight or influence into how the technology shapes their students’ education.
The only way to ensure AI is used to complement learning rather than obstruct it is through tailored implementation. In a 2025 survey, 85% of college students nationwide said they had used generative AI for coursework within the past year. At Brown, over 60% of students believe that AI use should be permitted for research support and helping with homework problems, according to The Herald’s Spring 2026 Poll. At that scale, students’ AI use is so widespread that it is clear guide rails and structure must be put in place. Studies suggest that such intentional uses of AI in the classroom can be beneficial. Researchers at Penn found that high schoolers in Turkey who used GPT-4 to practice math problems scored 17% worse than their AI-free peers when the tool was taken away. But in a second group that used a version of ChatGPT built to give hints instead of answers, students scored similarly to the students who used no AI. Similarly, in a randomized trial of 194 students in a Harvard physics course, a custom AI tutor that walked them through problems one step at a time produced more than double the learning gains of an active-learning class, and the typical student finished in 49 minutes instead of an hour. This is the purposeful integration we need. Rather than blaming AI for deficits in the classroom and undercutting the value of a resource that, used well, can dramatically improve how we learn, courses should either supply students with tools built to teach them, or design assignments that teach them how to use everyday AI tools effectively.
And the benefits of learning productive ways to use AI extends beyond students’ time on College Hill. According to a McKinsey survey, 89% of respondents said their organization uses AI in “at least one business function” in 2026, and in a 2026 international survey on the future of business education by the consultancy group CarringtonCrisp, 77% of employers said they expect new graduates to have experience with AI tools, but 58% said universities aren’t doing enough to build those skills. Brown’s goal to prepare students for the world beyond College Hill loses credibility if students graduate without knowing how to use a tool that the vast majority of employers rely on.
As it is, students themselves have to compensate for the gap in their AI education. Seven in ten college graduates surveyed by CarringtonCrisp said they taught themselves AI by downloading tools and experimenting. However, effective AI usage is difficult to learn from AI itself. This self teaching risks a deteriorating feedback loop — students who learn from AI have no outside reference of what good usage looks like so a habit that feels efficient, like asking for an answer and copying it down, faces no pushback. The more a student leans on AI, the less they understand the concepts they are exploring, and the less they understand, the more they lean on AI. Ironically, this is the same phenomenon that engineers try to avoid in AI, where models trained on AI generated data gradually degrade over time. Just as AI needs constant human input and outside reference points, students need professors to show them what proper use looks like.
College is the perfect setting to teach students how to use this technology properly. Professors have the resources and freedom to structure courses that use AI as a complement to their instruction. Students have the flexibility to experiment with the versatility of AI alongside peers and faculty across disciplines, each of whom adapts it to their own field. Sharing those approaches exposes students to the countless applications of AI, and that cross-pollination accelerates the growth and dissemination of AI literacy. And unlike a job, where stakes tend to be larger, college gives students the flexibility for more trial and error.
It is in this setting that IAPA 1002 is able to excel. While there are many assignments that encourage AI to aid writing and research, students are also graded on three in person debates and two in person exams, neither of which allow any technology. This shifts the focus to using AI to prepare, having it stress-test policy reports and digest information that you can later fact check or having it streamline part of the writing process for pre-assessment submissions. Ultimately, students still have to articulate their argument well and pull necessary information at appropriate times. The course may not use a custom tutor like the one in the Harvard study, but it follows the same model the researchers recommend, using AI for first contact with the material and class time to test whether we understand it.
Brown, of all places, should be willing and ready to adapt to the changing field of higher education. The Open Curriculum is built on the premise that students learn best when given the freedom to design their own education. Applied to AI, that principle means giving students room to experiment inside courses built to guide them. Brown students will go on to shape policy, build companies and lead institutions that are still figuring out how to use AI, and those institutions will need people who can use the tools responsibly. IAPA 1002 is one of the first to model this — and it shouldn’t be the last.
Angus Miller ’29 can be reached at angus_miller@brown.edu. Please send responses to this op-ed to letters@browndailyherald.com and other opinions to opinions@browndailyherald.com.
