Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise
Project Overview
The document discusses Tutor CoPilot, a generative AI tool aimed at enhancing real-time tutoring for K-12 students, especially in underserved communities. By utilizing advanced language models, Tutor CoPilot provides expert-like guidance to tutors, thereby improving student mastery in subjects like mathematics. A randomized controlled trial involving 900 tutors and 1,800 students revealed significant learning outcomes for those utilizing the tool, particularly benefiting less experienced tutors. The cost-effective pricing at $20 per tutor annually promotes the adoption of high-quality teaching strategies. Despite its successes, the implementation faced challenges regarding the appropriateness of suggestions based on grade levels. Overall, the findings highlight the potential of generative AI to transform educational practices and improve student learning outcomes in diverse settings.
Key Applications
Tutor CoPilot
Context: K-12 educational setting, particularly for students from underserved communities.
Implementation: Integrated into live tutoring sessions via a virtual platform, providing real-time suggestions to tutors based on student interactions.
Outcomes: Students whose tutors used Tutor CoPilot were 4 percentage points more likely to master topics, with lower-rated tutors seeing a 9 percentage point improvement in student mastery.
Challenges: Some suggestions generated were not grade-level appropriate, impacting their usability.
Implementation Barriers
Technical Barrier
Language models trained on web data may not perform well in specific educational contexts, leading to inappropriate or ineffective suggestions.
Proposed Solutions: Adapting language models using expert reasoning and context-specific data to better align with real-world K-12 interactions.
Implementation Barrier
Tutors may struggle with integrating AI suggestions into their teaching practices effectively. Additionally, the AI's suggestions may not always align with the tutors' needs or the students' comprehension levels.
Proposed Solutions: Providing targeted training for tutors to ensure they understand how to use AI-generated guidance in practical settings. Enabling customization options for tutors to adjust AI suggestions according to individual student needs.
Project Team
Rose E. Wang
Researcher
Ana T. Ribeiro
Researcher
Carly D. Robinson
Researcher
Susanna Loeb
Researcher
Dora Demszky
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Rose E. Wang, Ana T. Ribeiro, Carly D. Robinson, Susanna Loeb, Dora Demszky
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai