Learners who engage in inquiry-based learning often struggle when they do not have adequate domain knowledge or an understanding of where to find it. Domain knowledge often needs to be scaffolded to assist learners with inquiry-based learning. VERA, the Virtual Experimental Research Assistant, is an inquiry-based modeling-focused interactive learning environment where learners create ecological models, and then evaluate their results through simulation. The system is designed to serve as an introduction to scientific modeling and walk learners through the scientific method. With little domain knowledge in ecology, the models that learners build, while representing ecological concepts at a high level, often do not depict real world interactions that occur between species. To assist with this common pitfall, we have developed an ecology domain knowledge scaffolding AI which suggests realistic species interactions to learners. In a think-aloud with five participants, we found that learners, with the assistance of the scaffolding, were more likely to build ecological models that more accurately represent the real world. Conversely, learners who used the scaffolding did not always take its suggestions, and when using the AI, learners spent a similar amount of time on average looking for real world species interactions, despite the AI offering them an answer. In this paper, we investigate, with quantitative and qualitative methods, whether the AI can adequately scaffold domain accurate species interactions, whether the learner takes the feedback into account, and whether the scaffolding leads to gains in ecological knowledge.

Keywords: Agent-based modeling, Inquiry-based modeling, Ecology, College-level education, Science education, Domain Knowledge Scaffolding.