Improving Inclusiveness of AI Tutoring Platform’s User Experience For Marginalized Students

Generative artificial intelligence (AI), particularly large language models (LLMs), is becoming integral to intelligent educational systems, offering personalized learning and automated assessment. While promising, these systems raise significant concerns about their fairness, explainability, and trustworthiness, especially for learners from underrepresented groups and those who are neurodiverse. This research systematically investigates gender and neurodiversity bias within AI-driven educational interactions through a statistically grounded evaluation framework. Adapting the GenderMag methodology, we employed three personas, Abi (female, underrepresented), Tim (male, majority), and Pat (neurodiverse), to simulate learner-AI dialogues across distinct educational scenarios. A quantitative analysis of model outputs, focusing on affective engagement indicators such as sentiment, empathy, interaction depth, and stereotype cues, revealed consistent disparities. The findings demonstrate systematically reduced empathetic engagement and shorter responses directed at the Abi persona, alongside increased variability in interactions with Pat. These results expose inherent biases in current LLM-based educational tools and underscore the critical need for explainable and bias-aware AI design. Building on these diagnostic insights, a subsequent study designed an affect-aware AI educator informed by the same personas and evaluation metrics. This redesigned system demonstrated marked improvements, delivering balanced empathy scores, reduced stereotype signaling, and enhanced motivational scaffolding for marginalized learners. Collectively, this work provides an empirical foundation for auditing and mitigating bias in educational AI, advancing the theoretical underpinnings of explainable AI (XAI), ethical AI, and inclusive intelligent tutoring systems.

کلیدواژه ها:

Digital health ، Web and Society ، Web-Based AI Educational Platforms ، Bias Differential ، Neurodiversity ، generative AI