Developing an AI-based chatbot for practicing responsive teaching in mathematics

Responsive teaching promotes students' mathematical reasoning and positive attitudes toward mathematics. Due to the complexity of the work of teaching, preservice teachers (PSTs) have been provided with approximated opportunities to practice responsive teaching skills in teacher education programs. Although increasing demand for adaptive learning reinforces the need for research on artificial intelligence (AI) in education, there have been few approaches that engaged learners in meaningful interactions. Our goal was to develop an AI-based chatbot that engaged PSTs in an authentic, meaningful, and open-ended teaching situation to enhance PSTs' responsive teaching skills, specifically questioning skills through approximations of practice. The chatbot was designed to act as a virtual student who displayed misconceptions on the topic of fractions. By employing design-based research, we examined 1) design features and structure of the chatbot, 2) coverage of users' input, 3) PSTs' questioning patterns, and 4) users' experiences. Two iterations of design, implementation and evaluation took place in an elementary mathematics education methods course. To build the chatbot we qualitatively analyzed the training data, categorized them into the smallest meaningful intents of users, and prepared corresponding responses to each intent. At the final iteration, the refined chatbot adequately covered PSTs’ questions and provided realistic responses. We found a pattern of PSTs asking similar questions repeatedly in the conversation data. Through multiple iterations, certain design features could lead to improved questioning patterns and user perceptions, including sequential responses, informing responses, and personification. Implications, design features, and limitations are discussed.