The Hidden Costs of Artificial Intelligence in the Classroom: How AI-Generated Instruction Can Undermine Student Motivation and Academic Performance

The rapid integration of generative artificial intelligence into the global education sector has been framed largely as a revolutionary breakthrough for pedagogical efficiency. Proponents of the technology argue that AI-driven tools can empower educators by automating the more tedious aspects of the profession—drafting lesson plans, generating classroom materials, and providing instantaneous feedback—thereby freeing up time for direct student engagement. However, new empirical evidence suggests that the uncritical adoption of these tools may carry significant hidden costs. According to one of the first randomized controlled trials testing AI in real-world classroom settings, the use of AI teaching assistants can paradoxically undermine student learning and diminish the intrinsic motivation of the youth they are intended to support.

The study, led by Alp Sungu, an assistant professor at the Wharton School of the University of Pennsylvania, reveals a troubling trend: when teachers rely on AI as a "crutch" rather than a supplement, the quality of instruction tends to deteriorate. This decline in quality is not merely a subjective observation but is reflected in measurable outcomes, including lower student engagement and, in certain demographics, a marked decrease in standardized test scores. The findings suggest that the "efficiency" gained through AI may come at the expense of the human element—the personal voice and unique teaching style—that is vital for fostering a productive learning environment.

The Experimental Framework: A 10-Week Longitudinal Study

The research, detailed in a draft paper titled “Generative AI Can Harm Teaching,” was conducted by a team from the University of Pennsylvania, which included renowned educational psychologist Angela Duckworth. The study took place during the spring of 2025 and involved a substantial sample size: 193 teachers and over 2,800 middle and high school students. The setting was a private school chain in Turkey, providing a controlled yet diverse environment to observe the effects of AI integration across various grade levels and subjects.

To ensure the scientific rigor of the findings, the researchers employed a randomized controlled trial (RCT) design. Teachers were divided into two distinct groups. The experimental group was granted access to a customized ChatGPT-based teaching assistant, specifically tailored to align with Turkey’s national curriculum. This tool was designed to assist with the creation of lecture notes, homework assignments, and exam questions. Conversely, the control group was instructed to continue their teaching practices as usual, without access to the specialized AI assistant.

Over the course of the 10-week intervention, the researchers monitored how teachers utilized the technology. The primary use cases identified were the generation of lecture materials and the creation of assessments. While the AI provided technically accurate content, the study aimed to measure how this shift in the production of educational materials affected the ultimate consumers: the students.

Diminishing Motivation and the Loss of "Teacher Voice"

One of the most immediate and significant findings of the study was the impact on student motivation. Students whose teachers had access to the AI tool reported a noticeable decline in their classroom experience. When surveyed, these students rated their classes as less enjoyable, less interesting, and less important compared to their peers in the control group.

The researchers noted that the decline in intrinsic motivation was particularly pronounced in classrooms where teachers had already been identified as frequent users of AI prior to the start of the experiment. This suggests a cumulative effect, where a heavy reliance on automated content progressively alienates students from the learning process.

Professor Sungu posits that the root cause of this decline lies in the loss of the teacher’s individual "voice." Every effective educator brings a unique style, a set of personal anecdotes, and a specific way of explaining complex concepts that resonates with their specific group of students. When AI generates a lesson plan or a set of lecture notes, the output is often uniform, technical, and devoid of the personality that makes human instruction engaging.

"When you start using AI-generated material, you’re losing your personal voice," Sungu observed. "It might be technically good enough, but it doesn’t really carry your own style. If everything is very uniform, it just becomes a bit more boring." This homogenization of instruction appears to signal to students that the material is a commodity rather than a curated experience, leading to a psychological detachment from the subject matter.

Academic Performance: The Performance Gap Widens

While the overall average academic achievement across the entire sample did not show a drastic shift, a deeper dive into the data revealed a concerning disparity. The negative impact of AI was disproportionately concentrated among students of "weaker" instructors—teachers whose students had historically lower marks before the experiment began.

In these specific classrooms, student achievement and self-confidence both suffered a decline. Because academic performance was measured through externally administered standardized exams, the researchers were able to rule out the possibility that the lower scores were the result of biased grading by the teachers themselves. Instead, the data points to a genuine erosion of learning quality.

The study suggests that there is a stark difference in how "strong" versus "weak" teachers interact with AI. Stronger instructors likely treat AI-generated content as a "first draft," using it as a foundation that they then revise, adapt, and infuse with their own expertise and classroom-specific context. In contrast, weaker instructors may be more inclined to use the AI output "as is," delegating the core task of pedagogical creation to the machine. This "delegation" rather than "integration" results in a lower quality of instruction that fails to address the nuances of student needs.

The "Crutch" Phenomenon and Previous Research

The findings in "Generative AI Can Harm Teaching" echo Sungu’s previous research from 2024, which investigated how students use AI. In that earlier study, Sungu found that students who used ChatGPT as an "answer machine" rather than a learning tool saw their math skills erode. The common thread between the two studies is the concept of AI as a cognitive crutch.

Just as students may use AI to bypass the mental effort required to solve a problem, teachers may use AI to bypass the intellectual labor of lesson preparation. In both cases, the removal of the "actual work" leads to a degradation of the final outcome. For teachers, the "actual work" involves thinking through the logical progression of a lesson, anticipating student questions, and tailoring examples to the students’ interests. When AI performs these tasks, the teacher becomes a mere delivery mechanism for a third-party script, losing the deep familiarity with the material that is necessary for dynamic, responsive teaching.

Contextualizing the AI Efficiency Paradox

The current educational landscape is under immense pressure to do "more with less." Teacher shortages, administrative burdens, and the need for differentiated instruction have made the promise of AI-driven efficiency incredibly seductive. However, the Wharton study highlights an "efficiency paradox": tools designed to save time may actually necessitate more time if they are to be used effectively.

Professor Sungu, who uses AI in his own university-level teaching to create interactive games and polls, notes that the technology is rarely a true time-saver. "When I first get the output, it just looks great," he said. "And then, if I don’t immerse myself in it, the examples, the numbers don’t make sense. I end up spending an equal amount of time to improve the output or calibrate it to my class."

The danger lies in the temptation to accept the "good enough" output of the AI to meet a deadline. For an overworked teacher, the 10 minutes saved by not revising an AI-generated syllabus might result in 10 weeks of disengaged students.

Implications for Teacher Training and Policy

The study’s conclusions are not an indictment of AI technology itself, but rather a warning about its organic, unguided implementation. As Sungu cautions, it would be a mistake to conclude that "AI is terrible and will ruin education." Instead, the research serves as a call to action for school districts and policymakers to move beyond providing "access" and toward providing "instruction" on AI usage.

To prevent AI from becoming a detrimental crutch, several interventions are likely necessary:

1. Pedagogical Integration Training: Professional development should focus not on how to use the software, but on how to maintain "human-in-the-loop" standards. Teachers need to be trained to view AI as a brainstorming partner rather than an author.
2. Guardrails and Quality Standards: Schools may need to implement standards for AI-generated materials, ensuring that they are vetted for "personal voice" and local relevance before being used in the classroom.
3. Interface Design: AI developers in the EdTech space may need to create interfaces that discourage "copy-pasting" and instead prompt the user to make modifications or provide specific context before the final output is generated.
4. Support for Underperforming Educators: Since the study found that weaker instructors are most at risk of over-relying on AI, targeted support and mentorship for these teachers are crucial to ensure that technology does not widen existing achievement gaps.

Looking Ahead: The Future of the AI-Enhanced Classroom

As the 2020s progress, the presence of AI in schools will only increase. The challenge for the educational community is to ensure that this technological shift supports, rather than replaces, the essential human relationship at the heart of teaching. The study by Sungu and his colleagues provides a vital reality check for the "AI-first" narrative, reminding educators that the goal of technology should be to enhance human capacity, not to automate it out of existence.

The ultimate lesson of this randomized trial is that the value of a teacher is not found in the ability to generate a list of facts or a set of lecture notes—tasks a machine can do in seconds—but in the ability to inspire, to adapt, and to bring a unique perspective to the classroom. If AI is used to strip away those human elements, the result is a sterile learning environment where both motivation and achievement are the casualties of a false efficiency. Strategies for the future must prioritize the preservation of human judgment and creativity, ensuring that AI remains a tool in the teacher’s hand, rather than a replacement for the teacher’s mind.