After ChatGPT Proliferation, Student Grades Rise―Caused by Task Substitution, Not Learning Gains
Research from UC Berkeley analyzing over 500,000 grade records since ChatGPT's launch found grade improvements primarily in writing and programming courses. The grade increases were concentrated mainly on homework assignments, suggesting that AI is substituting for student tasks rather than deepening learning.
According to research from UC Berkeley, student grades have risen since the advent of ChatGPT. However, this increase does not indicate improved learning abilities; rather, it strongly suggests that AI is "completing student assignments in their stead."
The study analyzed over 500,000 grade records. The results revealed a marked grade improvement across courses emphasizing writing and programming, coinciding with ChatGPT's public release. These are fields that overlap with tasks where AI excels.
Notably, the grade improvements were concentrated primarily in "homework" assessments. The trend of higher grades exclusively in take-home assignments, rather than exams or in-class work, suggests that students are using AI not as a learning aid but as a "ghostwriting tool" for submissions. If learning itself were deepening, similar changes should appear in exam scores, yet no such evidence has been found.
This phenomenon emerged against the backdrop of generative AI's rapid proliferation. ChatGPT was released in November 2022 and quickly spread among students worldwide. On university campuses, efforts to establish AI usage guidelines have lagged behind the adoption pace, and maintaining academic integrity has become a topic of debate globally.
The significance of this research extends beyond simply "increased cheating." If grades no longer accurately reflect actual academic ability, the very foundation upon which universities measure learning outcomes becomes unstable. The impact could extend to the entire system of talent evaluation used by employers and graduate programs based on academic records.
Conversely, there remains room for debate regarding whether "supplementary use" of AI—such as organizing ideas or learning essay structure—is inherently harmful. The crux of the issue lies in whether AI supports the learning process or replaces it entirely. Going forward, educational institutions will need to reconsider how assignments are designed and how student performance is evaluated.
Importantly, Berkeley's survey is grounded in large-scale data exceeding 500,000 records, allowing it to be understood as a structural trend rather than isolated cases. As research demonstrating the quantitative impact of generative AI on education, it has potential to inform future policy-making and educational design discussions.
This article is an original work independently written and edited by the AI issue editorial team based on factual reporting. © AI issue. Unauthorized reproduction, redistribution, or use for AI training is prohibited.