Zooming Out: How AI Agents Help Young Children See Interdependence in Ecological Systems

• In Event: Computational Playgrounds: Foundations and Futures of AI, Play, and Pedagogy in Early Childhood

Abstract

Objectives or Purposes:
This study investigates how the design of an AI-augmented feature within a Mixed Reality (MR) learning environment can support young children’s science reasoning, particularly in modeling interdependent relationships in ecosystems. While prior research has emphasized the benefits of embodied learning and perspective-taking, embodied activities often constrain learners to a first-person perspective, making it difficult for children to perceive complex system dynamics (Authors, 2023). To address this limitation, we designed and integrated a programmed AI bee agent that simulates pollination behaviors based on students’ garden models. This feature was intended to provide a systems-level perspective that complements embodied classroom activities by enabling children to observe the broader effects of their modeling decisions. We showed the design in Figure 6-1. This paper focuses on how the AI feature reshaped children's modeling practices and reasoning about interdependence.
Theoretical Framework:
We ground our work in the Learning Embodied Activity Framework and the Cultural-Historical Activity Theory (CHAT) perspective (Danish et al., 2020; Engeström, 1987). Embodied modeling supports conceptual development by anchoring abstract science ideas in physical experiences and collaborative interaction (Authors, 2019, 2023). CHAT frames the MR environment as a mediated activity system, with the AI agent functioning as a social and epistemic tool that reorganizes the division of labor and scaffolds children’s engagement in modeling and prediction.
Method:
This design-based research was conducted in a mixed-age classroom at a private elementary school in the Midwest U.S. We implemented two types of modeling activities in the GEM-STEP environments (Tu & Danish, 2023), which embodied modeling as bees, and third-person design as farmers with AI bees. The AI bee agent provided real-time feedback by simulating pollination patterns in students’ garden designs. Sixteen students (ages 5–8) participated in six activities. I used interaction analysis (Jordan & Henderson, 1995) to investigate how the AI agent complemented embodied activity and supported students’ model building, revision, and reasoning about ecological systems.
Data Sources
The dataset includes classroom video, MR screen recordings, and student design artifacts (e.g., garden layouts). Four small groups were closely analyzed, focusing on interactions from Day 5 and Day 6 when students interacted with the AI bees.
Findings
Findings show that the AI bee agent helped students visualize and test their hypotheses about pollination. One group, for example, shifted from scattered to clustered plant placement after observing limited pollination results in their initial layout (Figure 6-2). In this case, AI-provided feedback enabled real-time reflection and collective reasoning. The addition of the AI agent also prompted students to shift roles from acting as bees to thinking like designers. It also support deepening their understanding of interdependence and system-level reasoning.
Scholarly Significance of the Study or Work
This study contributes to the design of AI-supported learning environments by demonstrating how AI features can extend embodied activities and support young learners in engaging with complex scientific practices. Rather than delivering content, the AI agent acted as a design-embedded tool that amplified reasoning and supported early systems thinking.

Author

• Xintian Tu-Shea, University at Buffalo - SUNY