Key Takeaways
- Early use of large language models (LLMs) can create an “illusion of learning” where students produce polished work but cannot apply knowledge without AI assistance.
- Research shows AI‑assisted problem‑solving leads to weaker brain connectivity, poorer retention, and quicker giving up when the tool is removed.
- Surveys indicate a majority of students believe increased AI use harms critical‑thinking skills, and faculty report declining exam performance despite higher homework scores.
- Educators are observing students who can quote AI‑generated text but struggle to explain underlying concepts, signaling a erosion of deep understanding.
- In response, institutions are redesigning assessments (oral exams, in‑class writing, blue books), limiting take‑home AI‑friendly assignments, and emphasizing active, technology‑free engagement with texts.
- Faculty warn that without deliberate intervention, overreliance on AI may undermine the core purpose of a college education—teaching students how to think, feel, and express themselves as humans.
The Shift From Cheating Concerns to Deeper Learning Risks
When ChatGPT debuted four years ago, many professors warned that students might simply use the tool to cheat on assignments. Today, those fears have evolved into a more profound anxiety: AI may be weakening students’ capacity to learn and think independently. Eric Klopfer, an MIT professor who co‑chaired a committee on AI in teaching, captured this sentiment when he said, “It’s the illusion of learning… it feels like you’re learning, it feels like you’re understanding things.” Yet, when students must apply that knowledge without the model’s aid, they typically falter, suggesting that surface‑level performance masks a deficit in genuine comprehension.
Empirical Evidence of Cognitive Off‑Loading
A growing body of research points to measurable harms from unrestricted LLM use. Nataliya Kosmyna, a MIT research scientist and lead author on a recent brain‑imaging study, warned that “while longer‑term studies are needed, it’s very concerning that changes are already apparent in such a short time period.” Her work revealed that students who relied on AI exhibited the weakest brain connectivity during essay writing and struggled to recall or quote from their own essays after completion. Complementary findings from Grace Liu’s Carnegie Mellon study showed that after just a few minutes of AI‑assisted problem‑solving, participants who lost access to the AI performed worse and gave up more quickly than peers who had never used the tool. These converging metrics—brain scans, questionnaires, behavioral tests—signal that AI functions as a powerful cognitive off‑loading mechanism, akin to search engines or GPS, but with amplified effects on memory and reasoning.
Institutional Findings and Survey Data
MIT’s president highlighted the issue as a watershed moment for higher education, noting that a campus committee concluded overreliance on chatbots can cause “diminishing critical thinking, weakening memory, eroding confidence, and undermining mastery.” A December Rand think‑tank survey reinforced these worries: two‑thirds of students agreed that “the more students use AI for their schoolwork, the more it will harm their critical thinking skills.” At Brown University, a similar committee found that most students had experimented with LLMs in their studies and expressed anxiety about long‑term impacts on their thinking and learning processes.
Humanities Classroom: Polished Outlines, Shallow Understanding
Shawna Dolansky, a professor in Carleton University’s College of the Humanities, described a disconcerting pattern among her first‑year students. They arrived with essay outlines that read like dissertation proposals—highly polished, touching on undiscussed topics, and containing quotations not present in the cited sources. When Dolansky probed the ideas behind the outlines, “they couldn’t answer.” She lamented, “That’s a broken system,” emphasizing that the students, who had enrolled for a four‑year humanities degree centered on Great Books, were losing the ability to engage with texts on their own terms.
STEM Classes: AI Solving Problems, Students Missing the Proof
At the University of California, Berkeley, Gireeja Ranade, a professor of electrical engineering and computer sciences, observed a stark change in her optimization and linear algebra course. Problems that once stumped undergraduates were now routinely solved by LLMs, yet many students failed the final exam despite having completed similar homework all semester. Ranade noted that “sometimes they will confuse the idea of reading and understanding a proof with being able to do the proof themselves,” revealing a gap between procedural assistance and conceptual mastery. The inability to reconstruct or explain proofs without AI hints at a superficial grasp of mathematical reasoning.
Computer Science: Honor‑Code Violations and Oral‑Exam Failures
Dan Garcia, also in UC Berkeley’s EECS department, reported alarming trends in his introductory courses. In a non‑major CS class, more than half of the students apparently violated the school’s honor code and used AI for a take‑home exam, prompting coverage in the Daily Californian. In a CS‑major introductory class, students who relied on AI for take‑home projects scored poorly on in‑person exams because they had not internalized the required programming logic. Garcia added an oral‑exam component to combat overreliance, only to find that “they just don’t know their code because they didn’t write it,” with many receiving a zero or one out of four when asked to explain their work. He also noted a steep drop in attendance and office‑hour visits, signaling disengagement from the learning process.
A Global Epidemic Calls for Immediate Action
Garcia characterized the situation as a “significant epidemic,” urging educators worldwide to “scramble.” Similar concerns have surfaced faculty meetings from Europe to Asia, where instructors report that students can generate seemingly sophisticated outputs yet lack the foundational skills to critique, adapt, or extend those outputs independently. The consensus is that the current trajectory threatens not only individual achievement but also the collective knowledge base that higher education is meant to cultivate.
Pedagogical Shifts: Rethinking Assessment and Engagement
In response, universities are forming committees, revising curricula, and experimenting with alternative assessment models. Some professors are eliminating or reducing take‑home assignments that lend themselves to AI substitution, replacing them with in‑class essays, oral examinations, or traditional blue‑book tests. Others are “flipping” the classroom: delivering lectures online for students to review at home, then using class time for problem‑solving, discussion, and hands‑on activities that demand spontaneous thinking. A subset of educators is even designing AI‑powered tools that challenge students—such as systems that ask learners to predict the next step in a proof before revealing the answer—forcing active cognition rather than passive consumption. Erik Voss, an assistant professor at Columbia University’s Teachers College, summarized the imperative: “We as teachers need to think about how we structure our lessons for more in‑class engagement, and think about motivation and what’s relevant and important and exciting to students.”
Concrete Changes in the Humanities: Writing by Hand, Tech‑Free Zones
Dolansky turned her three‑decade‑long teaching approach on its head. She ceased assigning take‑home essays, noting that “it’s easy to see why a student would be tempted to not spend six hours on an assignment that their roommate finishes in seconds with AI.” Instead, she requires students to write during class, sometimes revising earlier drafts on the spot. Her classroom is now tech‑free: students bring physical copies of the Great Books, take notes by hand in notebooks, and discuss ideas face‑to‑face. She assigns less reading but insists that learners annotate the texts in the margins, a practice that compels them to wrestle with language, argument, and tone directly. Reflecting on the shift, Dolansky argued that “what she’s teaching students now is even more valuable in the wake of AI: ‘Here’s how humans think, and here’s how humans feel, and this is how humans express themselves.’”
Looking Forward: Urgency for Research and Balanced Integration
While the anecdotal and empirical evidence mounts, experts agree that more longitudinal studies are essential to parse AI’s long‑term influence on cognition. Kosmyna urged the academic community to act swiftly, stating, “There is a need and an urgency to do more of these studies.” The conversation is shifting from whether AI belongs in education to how it can be harnessed without eroding the very skills—critical thinking, memory, confidence, and mastery—that a college education promises to develop. As educators redesign their classrooms, the underlying goal remains clear: preserve the irreplaceable human capacity to think, feel, and articulate ideas in a world increasingly populated by intelligent machines.
https://www.detroitnews.com/story/news/nation/2026/09/22/how-some-educators-say-ai-hurts-students/91887523007/