TutorMoments dataset reveals AI tutor behavior in real-time math sessions
The dataset includes 462 de-identified transcripts of U.S. math tutoring sessions. It captures AI tutor interactions and decision-making patterns. Researchers are analyzing when AI tutors intervene or step back.
TutorMoments is a dataset designed to explore the decision-making processes of AI tutors during real-time math tutoring sessions. It includes 462 de-identified, text-only transcripts of one-on-one tutoring interactions. The dataset is being made available through HuggingFace and is intended to help researchers understand how AI tutors determine when to assist and when to refrain from intervening.
The dataset was developed by a team at AllenAI and includes transcripts from real tutoring sessions conducted in the U.S. Each transcript captures a unique interaction between a student and an AI tutor. The transcripts are annotated to highlight moments where the AI tutor provided guidance or chose not to intervene. This allows researchers to study the nuances of AI tutor behavior in educational settings.
The TutorMoments dataset contains 462 transcripts, with each session averaging around 10.1 thousand words. The data includes interactions from 2 distinct tutoring platforms, providing a broad range of scenarios for analysis. Researchers can use this dataset to train and evaluate AI models that need to make decisions about when to offer help and when to let students work independently.
The release of TutorMoments has implications for the development of AI tutors in education. It provides a benchmark for evaluating how well AI systems can mimic human tutoring behaviors. The dataset may also influence the design of future AI tutors by highlighting the importance of timing and context in educational interventions. However, the use of such datasets raises questions about data privacy and the ethical implications of using real student interactions for AI training.
TutorMoments is still in development and has not yet been fully validated for all use cases. The dataset is being updated regularly, and researchers are encouraged to provide feedback on its utility. While the dataset offers valuable insights into AI tutor behavior, its long-term impact on AI education tools remains to be seen. The ongoing development of TutorMoments reflects the evolving nature of AI research in the field of education.