Co-Design at the Human-AI Frontier: Secondary Science Educators’ Practices and Perspectives on Generative Molecular Design

• In Event: AI as a Co-Pilot in Science Education: Reimagining Instructional Practice

Abstract

Objectives
This study examines how secondary science teachers engage in a GenAI-supported molecular design challenge during a co-design professional learning experience. Within an initiative to integrate data and molecular science education, the study addresses: (1) How do secondary science teachers use GenAI for molecular design goals and evaluate AI-generated structures? (2) How do teachers perceive the value of GenAI for science education?, and (3) How does participation in a co-design process influence the development of a GenAI tool? This work informs the responsible use of AI in science education and supports the co-creation of context-responsive learning environments.

Theoretical Framework
We draw on a co-design framework that positions teachers as partners in shaping pedagogies and tools for science education (Roschelle et al. 2006). Influenced by participatory design and situated expertise, co-design is a research method and learning experience. This initial design cycle involved a foundational version of the GenAI tool, utilizing teacher work and feedback. Their work aligns with key features of model-based reasoning (Halloun, 2007) and engineering design (Arık and Topçu, 2020), pillars of 3-dimensional learning, as they explored structure-function relationships, evaluated properties, and used feedback to refine their outputs. Prompt engineering and human-AI collaboration serve as an analytical lens for understanding interactions and learning (Cukurova, 2025).

Methods
Participants included a convenience sample of eleven U.S. high school science teachers with varied backgrounds and expertise. They used a ChatGPT-enabled web app for molecular design to complete a structured design task. They articulated a design goal (e.g., a novel sweetener based on AH-B-X binding theory), experimented with prompts, evaluated outputs, and reflected on the process. Data sources included a design worksheet capturing prompts, molecular structures, rationales, reflections, and a group discussion. A semi-structured interview with the software developer provided insight into the design responses. Thematic analysis examined strategies, reasoning, affordances, and the role of teachers in refining the tool. Trustworthiness was enhanced through triangulation and member checking.

Preliminary Results
Teachers drew on experience, demonstrating increasing sophistication in using GenAI, employing iterative strategies that balanced scientific accuracy, creativity, and goal alignment. The process was described as engaging and novel, highlighting their growing awareness of prompt specificity influencing outputs. Reflections underscored excitement about GenAI's scaffolding of students’ understanding of structure–function relationships, supporting inquiry-based learning, and visualizing abstract phenomena. Concerns included a lack of transparency in the AI’s reasoning, the risk of overreliance, and the need for teacher facilitation. Many framed their experience not as learning to use a tool, but as contributing to its development, underscoring the value of co-design.

Significance
This study contributes to emerging scholarship on AI in science education by demonstrating how co-design highlights meaningful use cases, exposes pedagogical challenges, and inspires innovation. Rather than positioning AI as a replacement for teachers, this work highlights genAI as a partner in creating student-centered, inquiry-driven learning. Co-design offers a replicable model for integrating GenAI in ways that are ethically grounded, pedagogically meaningful, and responsive to local context.

Authors

• Kent J. Crippen, University of Florida

• Minji Yun, University of Florida

• Charles Xie, Institute for Future Intelligence

• Annie Levendusky, University of Florida

• Dylan Bulseco, Institute for Future Intelligence