§ Research

Studying the Role AI in classrooms

Our research braids learning sciences, learning analytics, AI, and human-centered design. We work with teachers and students, audit the systems that mediate them, and build evidence on what helps and what harms.

§ What drives the work

Problem Statement

Artificial intelligence (AI) exacerbates educational inequities by threatening heterogeneity and promoting cultural and linguistic hierarchies. When used in learners’ contexts that differ from the majority, AI tends to perform significantly worse — leading to biased assessments, perpetuated cultural stereotypes, increased hallucinations, and a failure to capture linguistic and cultural nuance.

  1. 01

    What does TRAIL Lab aim to solve through its research?

    TRAIL Lab’s research addresses persistent educational inequities that arise when AI is introduced into classrooms. These inequities often affect historically marginalized students and are rooted in how AI systems are designed, evaluated, and used in real-world educational settings.

  2. 02

    How does TRAIL Lab tackle these challenges?

    By working at the intersection of learning sciences, learning analytics, AI, and human-centered design, TRAIL Lab develops new methods to identify and mitigate AI biases. We center the lived experiences of historically marginalized students in our design processes to ensure AI systems reflect their realities.

  3. 03

    What role do teachers play in TRAIL Lab’s research?

    A key part of our work is enabling K–12 teachers to use AI in ways that recognize and amplify students’ linguistic and cultural assets. Our tools and methods are designed to support educators in creating more inclusive, responsive, and equitable learning environments.

§ Research areas

Active lines of inquiry.

01 / 04
Blurring the Language Boundaries: AI Support for Translanguaging in Classrooms

Funded by Spencer Foundation

02 / 04
Curating A High-Quality Spanglish Dataset to Support Collaborative Problem-Solving in Multilingual Classrooms

Funded by Institute for Diversity Science, UW-Madison

03 / 04
A Framework for Valid and Reliable Audits of Biases in Large Language Models

Funded by American Family Funding Initiative

04 / 04
Disillusioning AI for Teachers Fellowship

Funded by School of Education, UW-Madison