LLMs are Biased Teachers: Evaluating LLM Bias in Personalized Education

TL;DR

This paper evaluates biases in large language models used as educational teachers, revealing significant demographic biases that could harm student learning and proposing metrics to quantify these biases.

Contribution

It introduces two bias metrics, MAB and MDB, and applies them to analyze bias in 9 state-of-the-art LLMs across educational content.

Findings

Bias is similar across models, highest along income and disability axes.

Lowest bias observed for sex/gender and race/ethnicity.

Models can perpetuate or reverse harmful stereotypes in education.

Abstract

With the increasing adoption of large language models (LLMs) in education, concerns about inherent biases in these models have gained prominence. We evaluate LLMs for bias in the personalized educational setting, specifically focusing on the models' roles as "teachers." We reveal significant biases in how models generate and select educational content tailored to different demographic groups, including race, ethnicity, sex, gender, disability status, income, and national origin. We introduce and apply two bias score metrics--Mean Absolute Bias (MAB) and Maximum Difference Bias (MDB)--to analyze 9 open and closed state-of-the-art LLMs. Our experiments, which utilize over 17,000 educational explanations across multiple difficulty levels and topics, uncover that models potentially harm student learning by both perpetuating harmful stereotypes and reversing them. We find that bias is similar for all frontier models, with the highest MAB along income levels while MDB is highest relative to both income and disability status. For both metrics, we find the lowest bias exists for sex/gender and race/ethnicity.