AI Partners to Support Group Collaboration in Classrooms

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

Abstract

Objectives: Research on how people learn has converged on a perspective of learning as both interactive and collaborative. Yet, the dominant approach to incorporating artificial intelligence (AI) in classrooms has focused primarily on a vision where students interact individually with technology that "optimizes" learning via individual-level adaptations (Grandbastien et al., 2016). There is therefore a disconnect between the ultra-personalization of AI research and the shifting landscape in the learning sciences pointing toward the essential role of collaboration to develop shared understanding and move thinking forward. There is a clear need to research and develop the AI technology of the future classroom; where the AI supports collaboration at the individual, group, and classroom level.



Methods: One AI partner developed within the National Science Foundation’s Institute for Student-Al Teaming (iSAT) aims to address this gap by developing JIA, an AI partner designed to support and encourage collaboration in groups during Jigsaw activities. Jigsaw activities provide a structure to classroom activities that supports cooperative learning. In Jigsaw activities, students first work in a small group to become ‘experts’ in a particular content area. Next they are grouped with other students who have different areas of expertise. These new groups must collaborate to solve a problem that requires knowledge from all content areas. Jigsaw activities provide a valuable opportunity for social learning and collaboration, but it can be challenging for teachers to monitor and support each group of students (and individual students within a group).

We are designing JIA as a peer-mentor that can provide language-based interventions to support students throughout the Jigsaw process. Specifically, the design of JIA includes (i) encouragements to promote flow of conversation and turn taking, (ii) content support to help students to promote knowledge-enhanced dialog, and (iii) content support to support brainstorming with a focus on enhancing curiosity-driven question asking and the promotion of divergent lines of thinking via group dialog. These techniques draw heavily from natural language processing and generative AI. For example, using Abstract Meaning Representations to look for semantic differences between a student’s dialogue about a given topic/expertise area while sharing out expertise during part one of the Jigsaw, or augmenting question generation with world knowledge and common sense reasoning during the second part of the Jigsaw.



Scholarly Significance: In this presentation we will describe the AI components underlying JIA and share findings from our iterative evaluations of JIA, which includes usability and empirical studies in lab and classroom settings. We will also detail our interdisciplinary design and innovation process, which includes a commitment to the Responsible Innovation (RI) Framework (Stilgoe, et al 2013). The RI framework is at the core of our AI innovations within iSAT, whose vision is to reframe the role of AI in education, expanding from a current emphasis on intelligent tools supporting personalized learning through unimodal, individualized, unidimensional instruction towards a future where AI is viewed as a social, collaborative AI Partner that collaborates with students and teachers to make learning more effective, engaging, and equitable.

Author

• Leanne Hirshfield, University of Colorado - Boulder