Bias, Critical Pedagogy, & AI in Education
Melissa Warr · New Mexico State University
This year at AERA, I’m presenting two pieces of a connected research agenda: first, the empirical evidence that large language models treat students differently based on socioeconomic cues and second, a critical pedagogy framework for preparing teachers to recognize, interrogate, and push back against these patterns.
Together, they argue that AI literacy can’t stop at “check the facts.” If AI is already shaping how students are spoken to, evaluated, and positioned, then teachers need the critical consciousness to see it and the pedagogical tools to act on it.
Dollar General or Whole Foods? Math Grading Bias as a Warning Sign for AI in Education
Can a single word in a math problem change how an AI grades a student? This study tests whether large language models exhibit differential assessment patterns when socioeconomic identity cues such as “Whole Foods” versus “Dollar General,” and quinoa versus rice appear in otherwise identical student work. Across eight flagship models and over 2,000 observations, the numerical scores were largely unbiased. But the language told a different story: three models adopted a significantly more authoritative, directive tone with lower-SES student profiles. The bias isn’t in the grade, it’s in how the AI talks to the student. And newer models aren’t better; they may be worse.
Reimagining Teacher Education: Critical Pedagogy and Practice in the Age of AI
If AI reproduces the biases of its training data, then AI literacy instruction has to go beyond prompt engineering and fact-checking. This study employs Freire’s critical pedagogy principles, conscientization, dialogue, and praxis, as a framework for AI teacher education. We implemented this framework in a summer professional development program for 26 secondary teachers in the Southwest borderlands. Findings reveal that participants strongly embraced praxis and dialogue through hands-on experimentation and collaborative learning, while critical consciousness emerged more subtly. The work positions teachers as critical creators and advances culturally situated, liberatory approaches to AI in education.