🐑 Basic Lesson
Foundational complex systems concepts around the sheep flock: recognizing emergent flocking patterns and matching agents in the environment to their roles.
Project
SMILE 🐑
Science Misconception-Focused Immersive Learning Environment
SMILE is a research project focused on designing and studying an immersive science learning environment that helps students understand emergent phenomena and address common misconceptions in complex systems. The environment combines interactive simulation, LLM-based agents, and opportunities for generative learning activities (e.g., making predictions) as well as metacognitive monitoring (e.g., confidence-check reflections).
Research Questions
- How does learner engagement within an open-world, self-paced, LLM-supported learning environment relate to learning outcomes?
- How do parameter control and questioning agency affect learners' understanding of emergent phenomena in complex systems?
- How can conceptual frameworks such as PAIR-C (Patterns, Agents, Interactions, Relations, and Causality) be integrated into the design of interactive simulation-based learning environments?
The Flocking Simulator
An immersive open-world environment where learners control a virtual dog to explore animal groups exhibiting collective behaviors: sheep flocks, duck flocks, and bat swarms as analogies for emergent processes in complex systems.
🦆 Intermediate Lesson
The three flocking rules from the Boids model (separation, cohesion, alignment) explored with the duck flock, covering Interactions and Relations.
🦇 Complex Lesson
Advanced ideas about Causality with the bat swarm: net effects vs. chain effects, common misconceptions, and transfer to phenomena like Earth's climate.
The Butterfly Tutor
Learners can call a dynamic LLM-based Butterfly Tutor (powered by GPT-4o) at any time for adaptive, context-aware support. Its design follows three principles:
- Theoretical alignment: responses are grounded in the PAIR-C framework to ensure conceptual consistency.
- Environmental awareness: the tutor uses regional metadata to contextualize its responses, intervening with certainty when it detects inactivity or off-task behaviors.
- Non-intrusive representation: a blinking butterfly icon gently reminds learners of the feature without demanding interaction, and softly redirects off-topic conversations back to complex systems concepts.
Inside the Environment
Animated snippets from the flocking simulator showing key features of the learning experience.
Lesson 1: Emergent Flocking (Sheep)
Learners answer prediction questions and toggle "Show Interactions" to reveal the invisible network of local interactions behind the flock's pattern.
The Butterfly Tutor
Learners chat with the LLM-based tutor, which responds with context-aware, PAIR-C-grounded guidance about the emergent behavior around them.
Lesson 2: Predict & Manipulate (Ducks)
Built-in prompts ask learners to predict flock behavior, then test their predictions by tweaking the alignment, cohesion, and separation sliders.
Lesson 3: Causality & Transfer (Bats)
Learners explore advanced causality concepts with the bat swarm, testing whether emergent patterns hold as parameters such as speed change.
AI-Supported Corrective Feedback
For open-ended questions, the system generates real-time elaborated feedback on student responses via the OpenAI API.
Non-Intrusive Tutor Access
A gentle "Call Butterfly" prompt reminds learners that the LLM tutor is available without demanding interaction.
Key Findings
In a 2*2 experiment with 69 postsecondary students varying control agency (Yes/No parameter control) and questioning agency (Yes/No LLM-based tutor), There are no significant differences on learning gains detected across conditions. Greater engagement with the LLM-based tutor, particularly through content-focused interactions, was associated with higher learning gains. Time spent in exploratory regions (not formal lessons) also significantly predicted posttest performance, highlighting the importance of fostering meaningful engagement with both conversational tutors and interactive features in LLM-supported environments.
Scientific Outputs
- ISLS 2026 Short Paper. Nair, A., Su, M., & Nagashima, T. (2026). Exploring Learner Engagement in an LLM-Supported Simulation Environment for Complex Systems Understanding. International Society of the Learning Sciences (ISLS) 2026.
- Journal Article (Under Review). Su, M., Nair, A., & Nagashima, T. Exploring Learner Agency in an AI-supported Simulation Environment for Complex Systems Education.
- Conference Presentation. EARLI SIG 6&7 (2026). Accepted contribution on the flocking simulator and complex systems learning.
Project Team
Developers
Amit Nair (Flocking Simulator)
Researchers
Dr. Man Echo Su (PI, Project Lead)
Dr. Lidia Altamura (Postdoc collaborator)
Amit Nair (Research Assistant)
Prof. Tomohiro Nagashima (Advisor)
Design Partners
Developed through participatory design with a high school biology teacher and a domain expert in complexity science.