AI Grading Tools Systematically Over-Score Student Essays, Study Finds
Research indicates generative AI fails to reliably reproduce human judgment, often inflating marks for lower-quality writing.
Generative AI systems tend to assign higher marks to student essays than human graders, according to new research. The findings suggest a significant discrepancy between how AI evaluates writing quality and the established academic standards used by educators.
A study published in the journal 'Assessment & Evaluation in Higher Education' found that generative AI tools, such as ChatGPT, do not reliably reproduce human judgment when marking extended written work. The research specifically observed a pattern of inflation: essays that were lower-scoring by human standards tended to receive higher marks from AI systems. Conversely, some higher-scoring essays actually received lower marks from the AI compared to human assessments.
The Shift Toward Automated Feedback
This research arrives as educational institutions increasingly integrate AI into their grading and feedback workflows. The push for automation is driven by the desire to reduce the administrative burden on faculty and provide students with more immediate feedback. However, the integration of these tools has sparked growing concerns regarding the reliability and consistency of AI compared to the nuanced evaluation provided by human educators.
Risks of Grade Inflation
The systematic over-grading of students presents a risk of widespread grade inflation. If AI tools consistently award higher marks than earned, it creates a misalignment between a student's perceived performance and their actual academic proficiency. This gap could potentially mask learning deficits and undermine the credibility of academic credentials if automated systems become the primary arbiter of student success.
The Path Forward
As the industry grapples with these discrepancies, the focus remains on whether AI can be calibrated to match human academic rigor. While AI offers speed, the current evidence suggests it lacks the critical judgment necessary for high-stakes assessment. Educators and policymakers must now determine if AI is best suited as a supplementary tool for initial feedback rather than a replacement for final human grading.