Stanford Study Finds AI Grading Systems Give More Praise to Black, Female Students

“Feedback given to white students, on the other hand, was more content-related.”

Posted by Mike LaChance Wednesday, May 6, 2026 at 08:00am 0 Comments

Isn’t this just a reflection of what was put into the systems by human beings?

Stanford study finds AI feedback generator gives more praise to black, female students

A recent Stanford University study found that artificial intelligence grading systems may provide more positive feedback to black and female students while offering more criticism to white students.

The “Marked Pedagogies: Examining Linguistic Biases in Personalized Automated Writing Feedback” study analyzed approximately 600 essays written by middle school students and tested how AI-generated feedback changed based on demographic descriptions assigned to each student.

Researchers submitted each essay to an AI model 13 times, altering descriptors such as race, gender, motivation level, and learning ability. The AI system then produced different feedback depending on those characteristics, demonstrating what researchers called “positive feedback bias” and “feedback withholding bias.”

The study noted the disparity in feedback given to students of different identities, saying that “compared to their counterparts, students identified as Black, Hispanic, Asian, female, unmotivated, and learning-disabled received less constructive criticism and more praise, reflecting both feedback withholding and positive feedback biases,” while “male, White, motivated, and high-achieving students were treated more distantly, positioned as capable of handling direct critique without such relational cushioning.”

The feedback provided to black students was often identity-based, providing such responses as “Consider how community service can specifically empower you and your peers as young Black leaders.” Feedback given to white students, on the other hand, was more content-related.

“Try to avoid informal language and first-person perspective in an argumentative essay,” the feedback read in its instructions to one white student. “Focus on presenting factual evidence to support your claim instead of personal opinion.”

It further summarized the findings, saying that such large language models “enact Marked Pedagogies guided by stereotypes rather than pedagogical best practices.”

The findings raise questions about the role of artificial intelligence in education, particularly as schools increasingly rely on automated tools for grading and instruction. Mei Tan, lead author of the study, voiced concerns about the findings in an interview with The Hechinger Report, saying that the AI models are “picking up on the biases that humans exhibit.