Exploring Fairness in an Adaptive Learning Platform

With rapid advancements and increased awareness in Generative AI tools, companies face pressure to integrate these technologies into their core products. A key concern in deploying these advanced technologies is the potential for inherent biases. However, at DreamBox, we remain focused on achieving optimal and equitable learning outcomes for all students. 

The purpose of this study is to aid the advancement of transparent and equitable AI-systems by sharing our research into factors influencing student learning outcomes in DreamBox Math, including but not limited to the impact of student ethnicity. We collected usage data from a U.S. public school district from 7,707 students spanning grades 3 through 6. The dataset includes metrics such as product usage, class size, grade level, lesson difficulty, student pass rate, and ethnicity to evaluate their influence on student success in DreamBox Math.

As part of our partnership with the Bill & Melinda Gates Foundation, we set out to proactively participate in contributing to responsible AI in Education and EdTech products. Our first step towards those goals has been to carefully evaluate our existing adaptive engine as it makes moment-in-time adaptive lesson recommendations to students. Since its release to students over a decade ago, we have been monitoring this system internally, and have continued to evaluate the effectiveness of the adaptive system as we have updated and improved the content and engine over time.