#257: From Talk to Evidence: Using Conversation-based Assessment to Elicit Scientific Reasoning

This study investigated the use of a conversation-based assessment (CBA) on storm formation and prediction to elicit high school students’ scientific reasoning. The task integrates AI-based agents to engage students in interactive dialogue across activities that involved science practices such as data analysis and constructing arguments. Results from a usability study revealed that scores from student-agent conversations, particularly those focused on causal explanation and data interpretation, were significantly associated with students’ final storm prediction performance while traditional item scores were not. Furthermore, analysis of student-agent dialogues showed that the AI agents supported students in revising and improving their responses. These findings highlight the potential of AI-enabled CBAs to surface evidence of scientific reasoning and support student learning and reasoning.