Can AI Make Peer Learning Work at Scale?
Over the past twelve months, our research team at Columbia University has designed and piloted an AI-facilitated debate tool — exploring how large language models can serve as discussion facilitators rather than content delivery systems, and whether structured peer exchange produces deeper learning than solo AI tutoring.
The Limitations of One-to-One AI Instruction
The dominant paradigm for AI in education mirrors the model that defined MOOCs: a single learner interacting with a machine. Platforms like Coursera pioneered this approach at scale — and encountered attrition rates approaching 90%.
Research consistently demonstrates that learners struggle to sustain motivation in isolation. Without peers to challenge assumptions and surface gaps in understanding, comprehension remains shallow.
Our hypothesis is that AI's most significant contribution to education will not be as a tutor, but as a facilitator of structured peer interaction — enabling the kind of cooperative learning that decades of research has shown to be most effective.
“Students who engaged in cooperative learning had greater knowledge acquisition, retention of material, and higher-order problem solving and reasoning abilities than students working alone.”
— Johnson & Johnson, 2014 meta-analysis on cooperative learning
AI as Discussion Facilitator
Our central hypothesis: large language models can replicate the structured facilitation skills of an experienced teaching assistant — managing group dynamics, surfacing productive disagreements, and keeping discussions on track. We designed a tool to test this.
Write the Memo
Students prepare a policy memo or assignment. AI-assisted drafting is permitted — the writing is not the primary learning objective.
Get Paired
Students are matched with a peer. Each reads the other's memo and prepares to discuss and challenge it.
Debate with AI Facilitation
An AI facilitator moderates the discussion — posing probing questions, highlighting points of disagreement, and ensuring both participants engage substantively with the material.
Deeper Understanding
The learning occurs in the exchange. Students develop critical thinking, argumentation, and genuine understanding — competencies that cannot be automated.
Key insight: Rather than prohibiting AI use, we redirect the locus of learning — from content production (where AI excels) to structured peer discussion (where human interaction remains irreplaceable).
From Proposal to Classroom Pilots
Over the past twelve months, our team moved through iterative cycles of design, development, and testing — building multiple prototypes to refine our approach to AI-facilitated peer learning.
Research & Design
Deep dive into cooperative learning research, user persona development, and platform architecture. Defined three learning personas: group assignment, study group, and individual tutor.
First Prototypes
Built and tested early versions of AI facilitation models. Explored different LLM approaches for group moderation and peer matching.
The AI Debate Tool
Launched the debate tool MVP — pairing students for structured discussions around their written memos, with AI serving as facilitator and moderator.
Pilot & Iterate
Running classroom pilots, gathering data on student engagement and learning outcomes, and refining the tool based on educator feedback.
Iterative Prototyping
Multiple design iterations across different AI facilitation models informed by user feedback
Classroom-Tested
Piloted with students and educators in real higher-ed settings at Columbia
Research-Driven
Every design decision grounded in cooperative learning literature and user research
Principal Investigators & Research Staff
A multidisciplinary team spanning education, technology, and applied research — based at Columbia University's School of International and Public Affairs.
Fernando Fabre
Adjunct Professor at SIPA, Columbia University. CEO of Kauffman Fellows. Co-founder of Collective Academy and Matterscale Ventures. Former President of Endeavor Global.
Sarah Holloway
Faculty at SIPA, Columbia University. Senior Fellow for Social Entrepreneurship. Serial social entrepreneur and co-founder of MOUSE.org, CSNYC, and The Wellness Classroom.
Michel Mosse
Founder & CEO of Owners. MIT MBA. Former growth lead at Uber Argentina and OLX. Founded an online vocational education platform (acquired by Aprende Institute). Serial entrepreneur in EdTech and SMB services.
Patricio Mosse
20+ years in full-stack development and tech leadership. MSc in Computer Science (Universidad de Buenos Aires) with AI research at INRIA and University of Montpellier. CTO at BRX Finance. Co-founder of Wifers. Coauthor of academic publications on argumentation-based AI systems.
Advisors
Opportunities for Participation
For Educators
We are seeking faculty in higher education who are interested in piloting the AI Debate Tool with their students. Participating instructors will receive full access to the platform and support from our research team.
Request Pilot AccessFor Researchers
We welcome collaborations with researchers studying cooperative learning, AI in education, or peer instruction. We are particularly interested in partners who can contribute to our evaluation of learning outcomes and engagement metrics.
Propose a CollaborationFor Sponsors
This research is supported by philanthropic funding. If you are interested in learning about our progress or supporting the next phase of this work, we welcome your inquiry.
Contact UsGet in Touch
For inquiries about pilot participation, research collaboration, or general information about the project.