AI Inflates Student Grades by 13% in Unsupervised Tasks

A recent UC Berkeley study involving over 500,000 grades from a prominent Texas research university has revealed a significant uptick in student grades, particularly in courses heavy on writing and coding, since the introduction of ChatGPT in November 2022. This grade inflation, marked by a 13 percentage point increase in A grades, appears predominantly in homework assignments rather than proctored exams. The findings suggest that AI is largely replacing student effort on unsupervised tasks, raising concerns about the true value of college grades as an indicator of learned skills and potentially impacting future employment and graduate school selections.

Key Developments

• A UC Berkeley study analyzed over 500,000 grades from a large Texas university, spanning eight fall semesters from 2018 to 2025.
• Courses with a high proportion of writing and coding assignments experienced a sharp increase in grades after ChatGPT’s launch in November 2022.
• The share of A grades in these courses jumped by 13 percentage points, representing a 30 percent increase above the 2022 baseline.
• This grade inflation is primarily observed in homework, with little to no statistically significant effect on exam scores, suggesting AI is replacing student work.
• The average GPA rose by 0.12 points, and the grade distribution narrowed, indicating A-minus and B-plus grades are frequently being elevated to straight A’s.

What Happened

Researchers at UC Berkeley undertook an extensive analysis of over half a million student grades at a major public research university in Texas. The study tracked grade trends across 319 courses and 84 departments over eight fall semesters, from 2018 through 2025. A key aspect of the methodology involved assessing each course’s “AI exposure” based on its assignment mix from fall 2022 syllabi, prior to ChatGPT’s public release.

The core finding was a distinct pattern: in courses characterized by a high volume of writing and coding tasks—areas where AI tools excel—grades escalated notably following ChatGPT’s debut. The proportion of students receiving A grades in these specific courses surged by 13 percentage points, translating to an approximately 30 percent increase over the baseline established in 2022. This shift also resulted in an average GPA increase of 0.12 points across the board, with a noticeable narrowing of the grade distribution as lower A and higher B grades were bumped to straight A’s.

Crucially, the study differentiated between the impact on homework grades versus exam scores. The data strongly indicated that the grade spike was driven by performance on unsupervised homework assignments. Courses where homework contributed more than the median share to the final grade saw an additional 16 percentage point rise in A’s compared to courses with less homework weight. Conversely, in courses with lower homework reliance, the effect was negligible and not statistically significant, reinforcing the hypothesis that AI is being used to outsource work rather than facilitate deeper learning.

Why It Matters

The observed grade inflation, particularly concentrated in unsupervised assignments, poses a significant challenge to the integrity of academic evaluations. If grades increasingly reflect AI-generated output rather than genuine student skill development, the traditional signaling value of a college degree could be severely diminished. This erosion of grade credibility could lead to misinformed decisions by employers and graduate admissions committees, who rely on academic performance as a proxy for a candidate’s capabilities.

Furthermore, this trend could inadvertently create a feedback loop where students, relying on AI for skill-building tasks, graduate with weaker foundational abilities in critical areas like writing and coding. OpenAI CEO Sam Altman himself expressed concern that the education system has not adapted meaningfully to AI, warning of “significant atrophy” in critical thinking skills. Such a decline in essential competencies could exacerbate skill gaps in the job market, potentially accelerating automation in roles traditionally requiring these human aptitudes.

Analysis

The Berkeley study offers compelling evidence that AI is not merely a supplementary tool for learning but, in many instances, a direct replacement for student effort on take-home assignments. The clear divergence between inflated homework grades and stable exam scores indicates a systemic issue where the assessment structure is vulnerable to AI’s capabilities. This situation differs from historical drivers of grade inflation, which typically occurred at the grading stage; AI intervenes at the production stage of student work, fundamentally altering the input instructors receive.

This dynamic presents a complex challenge for educational institutions. While the allure of higher grades is undeniable for students, the long-term consequences of this AI-driven inflation could be detrimental to their intellectual development and future career prospects. If students are not genuinely engaging with the learning process in subjects critical for cognitive development, such as writing and coding, they risk entering the workforce with underdeveloped skills. The study’s findings resonate with concerns raised by industry leaders like Sam Altman, who advocate for the continued teaching of these foundational skills for their role in training the mind.

Addressing this issue will require more than simply banning AI tools. A more effective approach, as suggested by the study, involves a fundamental redesign of assignments. This could include tasks that inherently limit AI utility, such as oral presentations where grades remained unchanged, or assignments that deliberately integrate AI use while requiring documentation of the process or follow-up interactions to confirm understanding. The goal must be to preserve the integrity of learning and assessment in an AI-permeated educational environment.

Future Implications

Near-term (3-6 months): Educational institutions will likely face increased pressure to revise assignment structures and grading policies to mitigate AI’s influence on unsupervised work. Expect a rise in discussions and pilot programs exploring AI-resistant assessment methods.

Medium-term (1-2 years): Universities may adopt more sophisticated plagiarism detection tools capable of identifying AI-generated content, alongside a greater emphasis on in-class, proctored assessments. There could also be a push for curriculum development that teaches students how to ethically and effectively integrate AI as a learning aid, rather than a replacement.

Long-term (3-5 years): The value proposition of traditional college degrees might undergo re-evaluation by employers and graduate programs, leading to a greater focus on demonstrable skills and portfolios over GPA alone. This could also prompt a broader societal debate on the role of foundational skills like writing and coding in an AI-augmented world.

Actionable Insights

• Educators should redesign assignments to either restrict AI use or integrate it in a way that requires students to demonstrate genuine understanding and critical thinking.
• Institutions should explore methods for documenting the work process, such as version control for code or iterative drafts for writing, to verify student engagement.
• Students should prioritize developing core skills like critical thinking, problem-solving, writing, and coding, recognizing their long-term value beyond immediate grade outcomes.
• Employers and graduate programs may need to adapt their evaluation criteria to account for potential AI-driven grade inflation, possibly through skill-based assessments or interviews.
• Policy makers should consider guidelines for AI use in education, similar to Norway’s approach, to ensure AI supports learning rather than undermining it.

Key Takeaways

• A UC Berkeley study detected a 13 percentage point increase in A grades in writing and coding-heavy courses post-ChatGPT launch.
• This grade inflation is primarily driven by performance on homework, not proctored exams, indicating AI is replacing student work.
• The average GPA rose by 0.12 points, with a narrowing of the grade distribution as lower A/higher B grades became straight A’s.
• Concerns are rising that AI is eroding the signaling value of college grades and could lead to weaker critical thinking and foundational skills among graduates.
• Experts, including OpenAI CEO Sam Altman, call for a systemic overhaul in education to adapt to AI, focusing on redesigned assignments and preserving core skill development.