Integrating generative AI with multimodal interaction to inquire into scientific phenomena

This study developed and evaluated two customised generative artificial intelligence (GenAI) chatbots designed to provide stepwise guidance for secondary students’ inquiry into scientific phenomena. Biology and physics chatbots were developed using the Predict–Observe–Explain (POE) framework and Retrieval-Augmented Generation (RAG) to align with Singapore science syllabuses and incorporated segmented video demonstrations. 14 evaluators quantitatively and qualitatively assessed the chatbots in terms of Accuracy, Clarity, Dialogue flow, Ease of use, and Efficiency. Both chatbots were rated positively overall, particularly for Ease of use, Clarity, Efficiency, and Accuracy, while Dialogue flow received lower ratings, especially for the physics chatbot. Qualitative feedback identified needs for stronger scaffolding, better handling of follow-up questions, and additional multimodal support. The findings highlight key design considerations for GenAI-supported scientific inquiry.