Primary speaker:
Steve Joordens, Professor, Psychology
Description:
While most Large Language Models (LLMs) prioritize text-based interfaces, many pedagogical contexts demand the nuance of verbal dialogue. This session introduces "Parallels": voice-first conversational agents designed to move beyond the "helpful assistant" archetype. Unlike standard LLMs that often provide direct answers and bypass critical thinking, Parallels are "cognitively agentic." They are engineered with specific goals, mental states and pedagogical boundaries, allowing them to facilitate active learning and scaffolded inquiry without over-stepping into task completion.
Over the past two years, we have developed these agents to enhance student support and facilitate high-stakes assessments. We will share insights from a large-scale experiment where Parallels conducted oral examinations for over 1,000 students, focusing on how these agents balance rigorous assessment with student care and guidance.
We discuss the technical and ethical considerations of voice-based AI, share data on student learning outcomes and provide a roadmap for educators looking to implement these tools in their own courses. The session concludes with an invitation to join an ongoing research collaborative dedicated to developing the next generation of student-centric AI agents.
