Racial Inclusion in AI-Powered Learning Environments

Authors:
Publisher website
Google Scholar
Abstract
In this chapter, the authors explore the idea of a Generative Divide, wherein the commercialisation of advanced Large Language Models results in a pay-to-learn scenario. This situation gives an advantage to students who can pay subscription tools over those who rely on versions with fewer features. The technical analysis looks at the Impossibility Theorems about fairness metrics, particularly the tension between calibration and equalised odds. It also examines strategies to fix the problem, such as adversarial debiasing and causal inference. Moreover, the authors discuss the paradox of accessibility, highlighting how AI acts as a cognitive aid for neurodivergent learners, yet falls short for marginalised groups due to high error rates in speech recognition for African American Vernacular English (AAVE). The chapter concludes by suggesting operational governance frameworks, like the “AI Literacy Heptagon” and Algorithmic Impact Assessments, to shift from techno-solutionism towards a deliberately inclusive educational framework.

This publication has 24 references indexed in Scilit: