Designing Xai Chatbots to Enhance Self-Efficacy in Youth in Pathfinding Problem Domains

Self-efficacy is a critical component of self-regulated learning (SRL), which empowers youth to overcome challenges and achieve learn- ing goals. In this study, we designed ALLURE, a multimodal artifi- cial intelligence (AI)-driven platform embodied with an XAI-driven chatbot to teach youth to solve a pathfinding problem, i.e., the Rubik's Cube, with the goal of improving their self-efficacy. Using ALLURE as a use case, we examine how various personas inter- act with an XAI-driven chatbot and how these interactions shape youths' self-efficacy. Personas offer a valuable design and evalua- tion strategy for modeling learner variability, enabling designers to represent and respond to differences in cognitive styles, motiva- tional profiles, and affective states, especially in early-stage systems like XAI-driven chatbots. Through interviews, think-alouds, and observational notes from usability testing, the results of this pre- liminary study indicate that while XAI-driven chatbots enhance self-efficacy, their effectiveness is not uniform across all youth. Findings underscore the need for educational chatbots to accom- modate diverse learning needs and preferences. Further, study findings offer implications for developing XAI-driven chatbots that provide transparent, tailored interactions to foster self-efficacy and competency in educational contexts.