HaeJin Lee's Dissertation Defense

Explainable artificial intelligence (XAI) techniques have advanced the development of personalized interventions in computer-based learning environments. Although previous studies have shown that XAI-driven interventions can improve students’ learning outcomes and foster the use of metacognitive strategies, it remains unclear whether these systems might inadvertently introduce biases against certain groups of students (e.g., individuals diagnosed with attention deficit hyperactivity disorder [ADHD]). Additionally, how and why these interventions affect learning have not been thoroughly examined—gaps that are critical for evaluations of XAI-driven interventions in computer-based learning environments. Investigating these factors is important because it allows us to employ unbiased and effective approaches for integrating XAI into interventions. Motivated by these research gaps, this dissertation investigates the causes of biases within the predictive model used in building XAI-driven interventions to obtain actionable insights into mitigating potential biases (Chapter 2). Specifically, we use XAI to understand how traditional approaches to engineering metacognitive strategies lead to a biased predictive model. To uncover how XAI-driven interventions impact students’ learning outcomes, Chapter 3, examines the mediating role of learning pattern interactions (e.g., the use of metacognitive strategies) in the effectiveness of the XAI-driven intervention. Leveraging insights from Chapters 2 and 3, we develop human-centered, XAI-driven interventions that supports students’ effective use of metacognitive strategies in online learning (Chapter 4). In these interventions, students receive personalized study guides designed to support their use of metacognitive strategies. One intervention provides explanations for the recommendations, whereas the other offers interactive recommendations based on the learning outcomes students want to achieve. Together, this work contributes to understanding how XAI methods can be used to identify biases in XAI-driven interventions, explain how these interventions affect learning, and design human-centered learning systems that better support students’ metacognitive strategies in computer-based learning environments. 

Questions? Contact HaeJin Lee.