Pedagogical Agents Research for AI-Augmented Narrative-Centered Learning

• In Event: The Past, Present, and Future of Intelligent Pedagogical Agents: From Avatars to Human Digital Twins

Abstract

Objectives: Over the past two decades, pedagogical agents have emerged as a promising family of personified artificial intelligence (AI) technologies for providing support to students through expressive communication and rich interaction (Johnson & Lester, 2016). Early work in pedagogical agents focused on designing, developing, and evaluating agent functionalities that integrated verbal communication (speech, text) and non-verbal communication (facial expression, gesture, gaze) to engage in conversation with learners to provide guidance, feedback, scaffolding, and interactive demonstrations, serve as navigational and attentional guides, and act as virtual teammates (Johnson, Rickel, & Lester, 2000). Pedagogical agents have matured considerably. Notable developments include increasingly powerful capabilities for adaptive scaffolding (from advances in AI-empowered learning technologies), intelligent virtual agent behaviors (from advances in virtual human technologies), and affect-rich communication (from advances in affective computing) (Johnson & Lester, 2018). Recent years have seen dramatic advances in the underlying AI technologies, particularly in natural language processing and machine learning, which have introduced a broad range of new opportunities for re-envisioning pedagogical agents.



Background: Pedagogical agents hold significant promise for AI-driven narrative-centered learning environments, which leverage AI to create engaging interactive story-based problem scenarios. To support interest-driven learning explorations, narrative-centered learning environments dynamically generate interactive narratives linked to authentic problem-solving scenarios, agents’ behaviors and speech, curricular content, and support for learning. These are dynamically tailored to serve the needs and interests of learners in specific contexts (e.g., classrooms and museums). These capabilities are being created based on advances in generalizable and robust training of machine learning models with limited supervision so as to enable tailored generation of interactive narratives that foster engaged student learning. With an emphasis on supporting narrative scenarios that elicit rich communication, require coordination, and spark collaborative creativity, the narrative-centered learning technologies will be driven by advances in multimodal generative machine learning and reinforcement learning.



Scholarly Significance: In this presentation we will describe the research on pedagogical agents in AI-driven narrative-centered learning environments that is being conducted at the National Science Foundation AI Institute for Engaged Learning. The mission of the AI Institute is to produce transformative advances in STEM learning and teaching with AI-driven narrative-centered learning environments, and pedagogical agents are a central focus of the Institute’s work. The presentation will discuss the opportunities introduced by advances in natural language processing and machine learning, describe the AI Institute’s ongoing research on pedagogical agents and the multiple roles they play in narrative-centered learning environments (including the roles they play in collaborative narrative-centered learning environments), and explore the rapidly changing landscape of pedagogical agents, including their communication and scaffolding capabilities, their affordances for different learning populations and learning contexts, and the future of pedagogical agents as AI becomes increasingly powerful and ubiquitous.

Author

• James C. Lester, North Carolina State University