A major study tracking grade distributions at a large selective university has found a striking pattern: since ChatGPT arrived two years ago, excellent grades have surged by 30 percent in subjects where artificial intelligence excels, while remaining flat in disciplines where it cannot help.
The research comes as colleges grapple with a deepening question about what their degrees now signify. Igor Chirikov, a UC Berkeley professor who conducted the analysis, framed the problem starkly: "We have a C student who is now an A student." The shift is not marginal grade creep but a wholesale elevation of performance in courses like English composition and programming.
Chirikov examined grades from 2018 through 2025 and deliberately chose not to identify the Texas research university in his sample, arguing the phenomenon extends far beyond one campus. "It's something that's happening across the higher ed sector," he said. The school he studied enrolls over 50,000 students and maintains publicly available grade records.
The pattern becomes clearer when looking at course structure. Classes that emphasize homework assignments over supervised exams show the steepest inflation. Lab-based courses and hands-on disciplines like sculpture saw no comparable shift. That correlation points squarely at unsupervised work as the entry point for AI assistance.
Grade inflation itself is not new. Excellent marks have drifted upward since the early 2000s. But Chirikov identified another mechanism accelerating the trend: faculty incentives. Professors often face pressure to grade leniently because student evaluations frequently influence promotion decisions. That creates a structural incentive to inflate marks, independent of AI entirely.
Some universities have begun fighting back. Professors are requiring handwritten exams and oral defenses, forcing students to demonstrate knowledge in real time rather than through polished written assignments. Yet Chirikov argues there is no single solution.
Instead, he advocates for what he calls AI-integrated assignments, where students use language models but must document their use transparently. "That's not an easy process, but we definitely should invest in that more than we do right now," he said. The goal would be teaching students to leverage AI as a tool while proving they understand the underlying material.
Chirikov cautioned against overstating the threat. Students have always found ways to pad their GPAs by cherry-picking easier courses. AI, he said, simply amplifies an existing dynamic. "I think AI just exacerbates the existing trends." The difference is scope and speed. What once required strategic course selection now happens within a single assignment.
Author James Rodriguez: "Universities can't grade-police their way out of this,they need to redesign how they test knowledge before an entire generation of graduates walks out the door untethered from what they claim to know."
Comments