Roberto Serrano had taught the same economics course at Brown University for nearly two decades. Class sizes stayed small. Scores followed predictable patterns. Then came spring 2026. Enrollment jumped to 86 students. A campus shooting the previous December left many anxious. Serrano relented. He offered a take-home midterm. Closed book in name. Unlimited time in practice. The results stunned him.
Average score hit 96 percent. Historical range for the course sat between 65 and 80 percent. Even with a harder exam designed for the format, something felt wrong. Serrano ran sample answers through ChatGPT. Matches appeared. Convoluted phrasing. Arguments built on contradiction rather than direct proof. He told students he would not void the midterm yet. They had a chance to prove him wrong on the in-person final.
The final delivered its verdict. Average plunged to 48.6 percent. A record low. Three students scored zero. Eighteen dropped the class after the midterm. Nine more skipped the final. Nineteen failed overall. Serrano voided the midterm and weighted the final at 80 percent. “We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is OK,” he said. “That leads to a declining society, to a failed society. We cannot choose to become idiots.”
His warning, reported by Inside Higher Ed, landed just as Brown’s Generative AI in Teaching and Learning committee released its findings. The timing sharpened an already uneasy conversation. Faculty and students alike see generative AI reshaping higher education. Not always for the better.
The GAITL report, drawn from 697 community responses, paints a clear picture. Fifty-six percent of undergraduates use AI tools daily or weekly. That climbs to 67 percent among graduate and medical students and 85 percent for those in master’s programs. Students turn to these systems for tasks once considered core to learning. Explaining solutions to difficult problems. Debugging code. Revising drafts. Assessing their own understanding. Forty-one percent of undergraduates and 54 percent of graduate students report frequent use for learning new material and skills.
Yet concern runs deep. Eighty-eight percent of Brown undergraduates and 73 percent of graduate and medical students worry that AI carries negative effects on their cognitive capacities. Faculty express even stronger doubts. Ninety-five percent fear it reduces students’ long-term learning. Eighty percent expect cognitive capabilities to decline. Seventy-five percent anticipate more cheating cases. Those numbers mirror national trends cited in the report.
Brown’s committee reviewed existing studies from the US and UK. Roughly 25 percent of students already submit AI-assisted assignments. The share grows sharply each year. Literature on the subject remains sparse but consistent in one respect. Over-reliance on generative tools risks decreasing higher-order thinking and metacognition. Students become passive recipients of AI-generated content rather than active participants in their own education. “Without active learning strategies, students risk becoming passive recipients of AI-generated content rather than active participants,” the report states.
The same document offers a narrow path forward. AI can supplement student cognition. Evidence supports that possibility. But only with specific guidance from teachers. The committee stresses instructor discretion. Faculty should set clear policies in syllabi. Some ban AI outright. Others permit tightly structured use with attribution and reflection. The key lies in communication. Instructors must explain their reasoning whether they restrict the tools or require them.
Brown has committed to acting on several recommendations. University-wide baseline guidelines will appear soon. Individual departments will adapt them. Investment in AI literacy training for teaching staff follows. The goal is informed restrictions that prevent students from compromising their own development. Detection tools draw skepticism. False positives and negatives abound. The focus shifts to redesigning assessments around process, authentic tasks and human judgment.
This episode at Brown fits a larger pattern gaining attention. A June 2026 NPR/Ipsos poll found more than half of K-12 teachers believe AI makes it harder for students to learn critical thinking skills. Seventy percent worry it weakens those skills along with research abilities. NPR reported that nearly three in four teachers see AI carrying bigger implications for education than the internet or computers. Many already use the technology to save time on lesson plans and materials. Yet the majority voice reservations about its effect on young minds.
Similar unease surfaces in higher education surveys. A January 2026 poll by the American Association of Colleges and Universities showed 95 percent of college faculty expect increased student overreliance on generative AI. Ninety percent predict diminished critical thinking. Eighty-three percent foresee shorter attention spans. These figures come as institutions scramble to update policies.
Broader research points to cognitive offloading as a central mechanism. When students outsource thinking to AI, they skip the productive struggle that builds mastery and resilience. A Brookings Institution analysis from early 2026 described the threats as primarily cognitive, emotional and social. AI prioritizes speed and engagement over the slower work of genuine learning. Students develop unrealistic expectations about how easy knowledge acquisition should feel. They grow less willing to persist through difficulty.
And yet the technology shows promise when deployed with care. Stanford’s Human-Centered AI group has explored ways AI might personalize instruction and free teachers for higher-value interactions. Some educators experiment with AI as a co-pilot rather than a crutch. The difference hinges on deliberate design. Teachers who treat AI as a supplement rather than a substitute report better outcomes. Those who allow unchecked use see the opposite.
At Brown the GAITL committee avoided overly restrictive rules. It called for experimentation within shared principles. Academic integrity remains non-negotiable. Assessments should emphasize process over product. Students must demonstrate their own knowledge and abilities. “The existence of GenAI does not change that,” the report notes. Instructors support intellectual development only when students engage sincerely.
Serrano’s experience offers a cautionary data point. His class split along lines of understanding. A small group grasped the material deeply enough to perform on the final. Many others did not. The discrepancy suggested some had used AI to bypass learning rather than accelerate it. Serrano submitted his evidence to Brown’s academic code committee in May. Initial silence followed. After the story gained traction, the committee requested individual complaints against each student. He found the process cumbersome for a case involving dozens of suspected violations.
Tricia Bertram Gallant, director of academic integrity at UC San Diego, commented on the structural challenges. Faculty rarely receive compensation or recognition for pursuing large-scale cheating cases. Committees need better resources and streamlined procedures for group allegations. Brown officials said they treat such matters seriously and have reached out to Serrano. He was scheduled to meet with the dean this week.
The incident has fueled discussion across platforms. On X, users debated whether AI makes smart students smarter and others less capable. One analysis of Serrano’s data suggested top performers used the tool efficiently to fill minor gaps while weaker students offloaded core understanding. The pattern echoes warnings from educators who see AI amplifying existing disparities in study habits and metacognitive skills.
Recent coverage reinforces the tension. An Education Week report highlighted downsides including reduced connections between students and teachers. Half of students in one survey said AI use in class left them feeling less connected to instructors. Teachers and parents also noted declines in peer-to-peer interaction. Seventy percent of educators expressed worry over weakened critical thinking and research abilities.
UNESCO has issued guidance for policymakers on AI in education. It stresses the need for competency frameworks that help students and teachers understand both potential and risks. The organization warns that rapid adoption has outpaced regulation. Without deliberate frameworks, unintended consequences multiply.
Brown’s response attempts to thread the needle. It acknowledges real harms while refusing to reject the technology. The committee’s literature review found that well-structured use with instructor guidance can support learning. Poorly managed adoption leads to de-skilling. The difference often comes down to whether students treat AI as an oracle or as a sparring partner that demands justification and critique.
Faculty at other institutions report parallel experiences. Some humanities professors have abandoned traditional papers. Others require students to write in class or submit process documentation including prompts and revision histories. A few integrate AI explicitly, asking students to evaluate its output, identify errors and improve upon it. These approaches demand more from instructors. They also force students to remain active participants.
The Serrano case may accelerate policy changes at Brown and beyond. The university plans to publish clear guidelines soon. Departments will tailor them. Training programs will help faculty develop sophisticated AI literacy. The hope is that informed teachers can guide students toward productive use rather than dependency. Yet the committee itself admits consensus remains elusive. Different disciplines face different realities. What works in computer science may fail in philosophy.
One theme repeats across reports and surveys. Teachers matter more than ever. Their role shifts from sole knowledge provider to designer of learning experiences, evaluator of authentic understanding and coach in critical engagement with powerful tools. AI will not replace them. It may, however, expose those who fail to adapt.
Serrano’s blunt message lingers. Society cannot choose to become idiots. His data suggests a portion of one elite university’s students edged in that direction during a single semester. The question now is whether institutions, instructors and students will treat the episode as an anomaly or a warning. Brown’s committee leans toward the latter. Its recommendations aim to preserve the hard work of thinking while harnessing new capabilities. Success depends on execution. And on whether faculty receive the support and time required to redesign courses for an AI-saturated world.
Other recent analyses add nuance. A Medium essay by an instructor described grading AI-generated text instead of student writing. The experience highlighted how reliance on the technology outsources not just production but comprehension. Students struggled to defend or even understand arguments their tools produced. Cognitive offloading, the piece argued, carries measurable costs to original thought and organization of ideas.
Commoncog, in a 2025 piece still circulating, proposed rules for using AI without eroding personal capability. The core idea involves maintaining control over the thinking process. Treat the tool as an aid after initial effort, not a starting point. Educators searching for practical strategies have shared similar heuristics. Force students to generate their own outlines before consulting AI. Require reflection on how the tool’s suggestions differ from their initial instincts. These tactics demand effort. They also build habits that counteract passive acceptance.
As more campuses confront similar incidents, the debate sharpens. Some see inevitable progress toward AI tutors and personalized pathways. Others warn of a generation whose intellectual muscles atrophy from disuse. Brown’s experience suggests the outcome is not predetermined. It will be shaped by policy, pedagogy and the daily choices of teachers and students. The tools have arrived. The question is whether humans will direct them or surrender to them.
WebProNews is an iEntry Publication