MetaSim: Developing and Assessing Metacognitive Skills using Simulated Learners during Complex Problem Solving in an Interactive Learning Environment

"Learning to learn" encompasses a set of essential metacognitive skills that help students become more effective and independent learners. This project aims to help high school students learn to learn through an Artificial Intelligence (AI)-powered interactive learning environment called MetaSim. At the heart of MetaSim are simulated learners--i.e., virtual students who model and explain how they plan, monitor, and reflect during algebra problem-solving. By watching and interacting with these virtual peers as they tackle challenging word problems in algebra, real students can develop the skills to manage their thinking and learning better. These simulated learners provide personalized feedback and guidance, helping real students apply their learning across different subjects. By combining expertise from learning sciences, computer science, and modeling and simulation, this project showcases the power of interdisciplinary collaboration in advancing how STEM problem-solving is taught and empowering students with lifelong learning strategies, making it a meaningful investment in the future of education and workforce readiness. This project addresses the challenge of effectively teaching high school students metacognitive skills--such as planning, monitoring, and reflecting--as they solve complex algebra problems. The interdisciplinary research team aims to design, develop, and test MetaSim, a new interactive learning environment with simulated learners--i.e., virtual agents powered by large language models (LLMs) and AI. These simulated learners model effective metacognitive and self-regulated learning (SRL) strategies in real-time, helping students see how expert learners approach and solve algebra word problems. The project is grounded in well-established metacognition and self-regulated learning theories and extensive research on multimodal data with interactive learning environments. The project includes classroom studies and controlled experiments to examine how students engage with the simulated learners, how their metacognitive skills develop, and how these skills transfer across problem-solving contexts. Data include learning outcomes, self-report surveys, and multimodal data sources such as interaction logs and think-aloud protocols. Analytic methods combine qualitative coding of metacognitive behaviors with advanced statistical techniques and machine learning to detect patterns in how students learn from and with the simulated learners. The goal is to understand whether this approach works and how and for whom it works best, with the exciting potential to inform future learning technologies, making them more adaptive, personalized, and effective in supporting student learning. This project is funded by the Research on Innovative Technologies for Enhanced Learning (RITEL) program that supports early-stage exploratory research in emerging technologies for teaching and learning. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.