EAGER AI-DCL Collaborative Research: Understanding and Overcoming Biases in STEM Education Using Machine Learning

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

Diversity is the cornerstone of innovation and essential for the progress of science. However, the number of female students in engineering, computing, and physical sciences in the United States remains strikingly low. The lack of diversity in science, technology, engineering, and mathematics (STEM) education is, to a significant extent, due to biases at different stages of schooling (e.g., different perceptions of math achievements by male and female students, lack of encouragement for female student enrollment in advance placement classes, stereotypes influencing college course selection). These biases appear as early as middle school: a critical period when student's educational experience can significantly influence their academic choices in high school and, ultimately, in deciding whether or not to enroll in STEM majors in college. In order to broaden the participation of women in STEM, it is critical to identify factors and practices in middle school learning environments that may attract (or repel) students into science. This award will use machine learning (ML) to develop new, automated, and data-driven methods for discovering and monitoring biases in STEM classrooms, focusing on middle school and early adolescence science and mathematics education. The project combines methods from social psychology, machine learning, and information theory to create algorithmic tools that monitor middle school student, teacher, and school-level data for factors that impact students' engagement in STEM. These tools will (i) help identify pedagogical or socio-economic factors that have a disparate impact on the decisions made by female students, (ii) predict which students are most vulnerable to being discouraged from pursuing STEM fields, and (iii) inform effective interventions that help close the gender gap. Despite its potential, the use of ML in education is a double-edged sword: while ML algorithms may be able to flag discriminatory patterns, they can also propagate biases and have an unwarranted disparate impact if left unchecked. Thus, in parallel, this project also aims to characterize the fairness challenges involved in deploying ML in education settings. The proposed approach will be validated on a dataset collected during a five year period from middle school students from across the United States. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.