This research is grounded in a practice-based teacher education paradigm that emphasizes opportunities to learn in and from practice (Grossman et al., 2009). One way to do so is to provide teachers with opportunities to try out and refine specific teaching practices in digitally simulated classrooms. Digital simulations can provide a low-stakes environment for teachers to enact and learn how to engage in core teaching practices (Estapa & Davis, 2023; Wang et al., 2021), such as skillfully eliciting and responding to students’ mathematical thinking (NCTM, 2014; TeachingWorks, 2020). Recent advancements in generative AI (GenAI) have opened new possibilities for online simulations, where GenAI can be used to power these simulations, removing the complexity of human-powered simulations and increasing their potential scalability, access, and impact (Bywater et al., 2025; Mikeska et al., 2025).
Our cross-disciplinary team utilized GenAI to design and deploy a voice-activated digital teaching simulation, specifically designed for elementary teachers to practice making models explicit in mathematics instruction and to teach third-grade students how to compare fractions. The student Neil (GenAI student agent’s name) was powered by GenAI and could respond to the teacher’s prompts or questions in real-time through an automated speech recognition system. The learning process for Neil consists of two phases: pre-learning and post-learning. In the pre-learning phase, the teacher will elicit Neil's understanding and misconceptions about comparing fractions. Then, the teacher will help Neil learn how to use area models to clarify his understanding of the content. Following this part of the interaction, the post-learning phase will occur, during which the teacher will evaluate Neil’s learning to determine whether his misconceptions have been addressed. We explored two research questions: (1) How accurately did the GenAI student respond to the teacher’s prompt or questions and align with the student’s knowledge profile, as designed? and (2) How did teachers perceive the GenAI simulation’s authenticity and usability?
Twenty-eight participants, including elementary paraeducators, in-service teachers, and administrators, participated in this study. The demographic breakdown of the participants is as follows: 54% identified as White, 23% as Black or African American, 19% as Hispanic or Latino, 4% as Asian or Asian American, 4% as American Indian or Alaskan Native, and 4% as Native Hawaiian or Pacific Islander. Additionally, 81% of the participants identified as female, while 19% identified as male. Regarding teaching experience, 42% have 1-3 years, 23% have 4-10 years, 19% have more than 10 years, and 15% have less than 1 year. In addressing concerns about using AI-powered tools in teacher professional development, 42% of participants were not concerned at all, 35% were slightly concerned, 1% were moderately concerned, and 4% were very concerned. Participants completed the Ordering Fractions GenAI simulation, a background information questionnaire, and a post-simulation survey, which included questions regarding participants’ overall experience with and perceptions about the GenAI simulation. After completing the activities, participants received an automated feedback report that detailed the productive teaching moves the teacher had executed, and the teaching moves they had not addressed. Primary data sources include chat transcripts from the GenAI teaching simulation (one per participant) and online survey responses. We used an evaluation rubric (Mikeska et al., 2025) to analyze the chat transcripts, calculated descriptive frequencies for Likert scale survey items, and identified themes in open-ended survey responses.
In terms of accuracy, findings indicated that 94.4% of the GenAI student responses addressed the participants’ prompts and questions while 93.2% of the responses were accurately aligned with the GenAI student’s knowledge profile. Findings also showed that participants perceived that the GenAI student’s responses were realistic (39% agreement), timely (46%), used an appropriate emotional tone (27%), consistently on topic (73%), grade-level appropriate (70%), and coherent throughout the conversation (73%). Most participants thought the experience was engaging (53%) because the student was responsive to the teachers’ prompts (27% of participants), engaged in conversation (23%), and the student’s voice was realistic (15%). In contrast, participants found that there were aspects of the simulation that were disengaging including that the GenAI student’s responses could be delayed (35%), there was a lack of virtual manipulatives for interaction (19%), and they experienced difficulties engaging in the conversation to elicit the GenAI student’s thinking (15%). Participants’ agreement on usability components of the simulation were as follows: preparation materials provided detailed information about the simulation (61% agreement), the instructions were clear (53%), they understood what they had to do (53%), easy to navigate through the platform (77%), did not encounter any technical issues (69%), and well-designed interface (77%).
The study’s findings indicate that GenAI teaching simulations have the potential to be utilized productively in mathematics teaching, as the current simulation operated as designed in most cases. In the future, it would be beneficial to consider how to incorporate more variation in GenAI student profiles, which would enable teachers to practice and gain a more authentic and impactful learning experience. The improvement of authenticity is particularly required as it ensures that student interactions appear realistic, responses are timely, and emotions are appropriately expressed, ultimately supporting teachers’ dynamic teaching capabilities. Moreover, improving certain features of the simulation, such as providing clear instructions and more detailed information, can guide teachers to understand what they need to do in the simulation. Additionally, reducing technical issues is crucial for creating a seamless experience. Making the simulation more user-friendly for teachers may increase their interest in integrating this GenAI digital simulation into their teaching practice. The overall enhancement in the simulation can make it more beneficial for teachers to improve their teaching skills. Since the GenAI digital simulation tool has the potential to support teachers’ practice, the school administration may consider incorporating it into teachers’ professional development, and teacher educators may integrate it into their courses to support pre-service teacher preparation.
Bywater, J. P., Lilly, S., & Chiu, J. L. (2025, July). Labeling disagreements: Illuminating the classification of mathematics teacher questions. In International Conference on Artificial Intelligence in Education (pp. 49-59). Cham: Springer Nature Switzerland.
Estapa, A., & Davis, J. (2023). Prospective teachers’ instructional decisions and pedagogical moves when responding to student thinking in elementary mathematics and science lessons. International Journal of Science and Mathematics Education, 21(5), 1703-1724. https://doi.org/10.1007/s10763-022-10304-3
Grossman, P., Hammerness, K., & McDonald, M. (2009). Redefining teaching, re‐imagining teacher education. Teachers and Teaching: theory and practice, 15(2), 273-289.
Mikeska, J.N., Beigman Klebanov, B., Bhatia, A., Halder, S., & Suhan, M. (2025, July 22-26). Evaluating the use of generative artificial intelligence to support learning opportunities for teachers to practice engaging in key instructional skills. In A.I. Cristea, E. Walker, Y. Lu, O.C. Santos, & S. Isotani (Eds.) Artificial Intelligence in Education. Artificial Intelligence in Education, Proceedings, Part II (pp. 378-391). AIED 2025. Lecture Notes in Computer Science, vol 15878. Springer. https://doi.org/10.1007/978-3-031-98417-4_27
National Council for Teachers of Mathematics. (2014). Principles to actions: Ensuring mathematical success for all. Reston, VA
TeachingWorks Resource Library. (2020). Eliciting and interpreting. https://library.teachingworks.org/curri culum-resources/teaching-practices/eliciting-and-interpreting/
Wang, X., Thompson, M., Yang, K., Roy, D., Koedinger, K. R., Rose, C. P., & Reich, J. (2021). Practice-based teacher questioning strategy training with ELK: A role-playing simulation for eliciting learner knowledge. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), 1-27.
Other presentations in this group:
Back
Trips and Tours
