Exploring the Design of Pedagogical Agent Roles in Collaborative STEM+C Learning

This paper explores the design of two types of pedagogical agents-teaching and peer-in a collaborative STEM+C learning environment, C2STEM, where high school students learn physics (kinematics) and computing by building computational models that simulate the motion of objects.Through in-depth case study interviews with teachers and students, we identify role-based features for these agents to support collaborative learning in open-ended STEM+C learning environments.We propose twelve design principles-four for teaching agents, four for peer agents, and four shared by both-contributing to foundational guidelines for developing agents that enhance collaborative learning through computational modeling.tify student difficulties and provide adaptive support in collaborative settings (Andrini, 2023).Vygotsky's (1978) zone of proximal development emphasizes the importance of guided support in helping learners accomplish tasks they cannot yet complete independently.However, designing effective pedagogical support is complex, requiring consideration of students' prior knowledge, progress in their problem-solving tasks, conversations, affective states, and body language (Cohn, Snyder, et al., 2024).In classrooms, teachers lack the bandwidth to monitor the cognitive, metacognitive, and social dimensions of students' group conversations and problem-solving progress in computer environments (Wambsganss et al., 2021).Pedagogical agents can bridge this gap by managing interactions with multiple student groups simultaneously.Teaching agents "play the role of human teachers" (Kuhail et al., 2023), delivering explanations, examples, and immediate feedback; while peer agents serve as learning companions, fostering peer-to-peer interactions and guiding students along learning paths (Kuhail et al., 2023).Elshan & Ebel (2020) identified design principles for their collaborative peer agent, Timmy, suggesting agents in collaborative settings should enhance perceptions of "humanness and social presence."Nguyen (2023) focused on design considerations for teaching and peer agents while students worked in groups to build a conceptual map of a marine ecosystem.While both agents improved student learning, Nguyen (2023) found that the peer agent facilitated more transactive exchanges and was rated as more sociable compared to the teaching agent.Recent research has similarly focused on enhancing teaching agents with human-like qualities to improve their sociability, such as offering encouragement and celebrating learners' progress (Jurenka et al., 2024).Google's LearnLM-Tutor, a teaching agent spanning several domains (e.g., biology, math, history, computer science), incorporates insights from participatory and co-design activities with teachers and students (Jurenka et al., 2024).Their design principles suggest that teaching agents should "see what the student sees."Despite extensive research investigating the roles of agents as both teachers and peers (Kuhail et al., 2023;Zhang et al., 2024;Sikstrm, 2024;Dai et al., 2022), previous studies have not established feature sets for agents in collaborative settings (Nguyen, 2023).This study addresses this gap by conducting semi-structured interviews with teachers and students to identify role-based features for teaching and peer agents in open-ended, collaborative STEM+C environments for high school students.Using these features, we derive 12 design principles for these two types of agents: four for teaching agents, four for peer agents, and four shared by both.