Co-Designing Human-AI Collaborative Tools for Learning to Model Complex Systems

Many important real-world phenomena arise from complex interactions among local-level elements (e.g., traffic congestion from driver behaviors, or stock market bubbles from investor behaviors). Agent-based modeling (ABM) has been shown to be able to help students explore how local-level interactions produce larger patterns through simple computational rules. Interacting with pre-built models is beneficial, but modifying and creating them has further benefits of engaging students in higher level scientific practices and empowering them to create models of new systems they want to understand. However, creating agent-based models still requires programming and modeling skills that many high-school students have not yet developed. Generative artificial intelligence (AI) could lower this barrier by translating everyday language into computer code, but it can also introduce errors and hidden assumptions that learners may struggle to recognize. Instead of asking AI to create an entire computer program at once, learners will work step by step from an idea to model rules, and then to code and a working simulation. The tools will make these connections visible so that learners can question the AI's choices, revise their models, and remain in control of the process. By lowering access barriers to computational modeling, the project will contribute to science and computing education, help prepare the next generation of students for scientific and technical careers, and enable young people to investigate complex systems that affect their lives and communities. The project will investigate how design of human AI tools can support the learning of agent-based modeling. First, the research team will develop and evaluate a structured human-AI method that connects a model's goals, measurements, rules, code, and simulated world, so that learners and AI can identify inconsistencies and revise the model together. Second, the team will conduct formative studies, co- design workshops, and focus groups with students, teachers, modeling experts, parents, and school administrators from two public high schools to create a web-based human-AI learning interface. Third, the team will compare the co-designed interface with a general-purpose AI programming interface in four after-school programs. Surveys, interaction logs, students' model rules and code, and interviews will be analyzed using qualitative and quantitative methods to examine learning processes, outcomes, usability, and the influence of students' prior experience. The project will produce open-source software distributed through the NetLogo open-source ecosystem, the most widely used agent-based modeling platform. It will also produce evidence-based guidance for educators, and research findings about reliable AI support for computational modeling and complex-systems 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.