Crafting a Responsible Dialog System for Collaborative Learning Environments

E. Margaret Perkoff, Emily Doherty, Jeffrey Bush, Leanne Hirshfield

Track: Innovations in AI for Education (Day 1)

Paper Length: short-paper (2 pages + references)

Keywords: controllable response generation, LLM, collaborative learning, dialog system

TL;DR: We propose a design for a dialog policy- driven conversational system to improve collaboration in a small group work setting using policy-design sessions with experts and an ongoing pilot WoZ study..

Abstract: This work-in-progress explores the use of large language models (LLMs) within a Computer Supported Collaborative Learning (CSCL) environment to enhance collaborative learning outcomes. To design a dialog policy-driven conversational system for small group work settings, we gather stakeholder feedback through a Wizard-of-Oz (WoZ) study and dialog analysis sessions with educational experts. Grounded in pedagogical frameworks of CSCL, our system aims to align with educational goals while addressing the limitations of current conversational AI evaluations that overlook education-specific objectives. Our preliminary laboratory study involves real-time interventions by a simulated agent (JIA), steering the ongoing dialog policy design. We discuss future work that involves transitioning to classroom implementations to ensure real-world validation while adapting the system to foster enhanced collaboration and knowledge building within authentic educational settings.

Submission Number: 20