Using digital teaching simulations powered by generative artificial intelligence to propel teacher learning

JAMIE N. MIKESKA, MANAGING PRINCIPAL RESEARCH SCIENTIST, ETS, USA

AAKANKSHA BHATIA, LEAD SUBJECT MATTER EXPERT –SOCIAL EMOTIONAL LEARNING, ETS, INDIA

Introduction 

Ensuring that teachers have opportunities to develop and refine their instructional skills is imperative to their professional growth and their ability to facilitate student learning (See et al., 2020; Shuls and Flores, 2020). Research has suggested that one promising approach to developing teachers’ practice is the use of digital teaching simulations, where teachers can practise an aspect of the work of teaching, such as orchestrating productive discussions or eliciting student ideas, in settings where there is no risk to real students (Dieker et al., 2023; Kaufman and Ireland, 2016; Mikeska et al., 2021). When using digital teaching simulations, teachers can interact with animated student avatars, either via chat-based texts or verbally, as they practise their teaching skills within either mixed-reality or virtual reality settings (Bondie et al., 2021; Peterson-Ahmad and Landon-Hays, 2020). However, most digital teaching simulations require humans to be behind the scenes operating the student avatars, which limits the potential to scale the use of these tools efficiently and in a cost-effective manner (Dove et al., 2023). The recent technological advances in generative AI (GenAI) have led to new opportunities for developing and using GenAI digital teaching simulations to support teacher learning. In this research article, we describe the design of a newly created GenAI digital teaching simulation and share findings from a recent study where we examined teachers’ engagement in and perspectives about one GenAI simulation designed to support them in learning how to elicit student thinking – a critical teaching skill required across content areas and grade levels (Davis and Palincsar, 2023; Shaughnessy and Boerst, 2018). Understanding these aspects provides a way in which to ensure that these GenAI simulations provide useful learning opportunities to support teacher learning.

Designing and using a GenAI digital teaching simulation

To develop the GenAI digital teaching simulation, we adapted a previously designed performance task that was created for use in a human-led simulated classroom. In human-led simulations, a trained human interactor operates behind the scenes, portraying one or more student avatars in real time. These interactors undergo extensive training to learn how to use specialised equipment to move, sound and respond like the students and develop familiarity with student thinking patterns, common misconceptions and instructional strategies. This training allows them to respond dynamically to teacher prompts and questions, creating an authentic and responsive classroom experience. This earlier task was created to provide opportunities for elementary teachers to practise eliciting ideas from one first grade (age 6-7) student avatar about their strategy for solving an addition problem. The interactor training materials addressed the student’s problem-solving processes, conceptual understandings and misunderstandings about this topic. For example, this student avatar used the counting on strategy to solve simple addition problems, although they would always count on from the first addend, regardless of whether it was or was not larger than the second addend.

Our team used the task materials to create the teacher-facing interface for the GenAI simulation (Figure 1). This interface provides the teacher with information about their goal in the simulation, the classroom context, the GenAI student’s work and how they should prepare for the simulation chat. The task materials also informed our team’s creation of the AI prompt powering the GenAI simulation, which was developed for OpenAI’s ChatGPT-4 model. For this GenAI simulation, a first grade GenAI student simulation named ‘Cecilia’ was programmed to solve the problem by first drawing circles to represent Mike’s six crayons and then drawing two more circles for Ann’s crayons. Then, Cecilia counts on from the first number to determine how many crayons in all, by stating ‘6, 7, 8.’ Cecilia also has knowledge gaps, as she has been programmed in the GenAI simulation to not understand the commutative property and not be able to perform addition fluently using mental maths.

Figure 1: GenAI simulation interface for the eliciting student thinking task

For this study, we recruited 10 in-service teachers through the Prolific crowdsourcing platform to try out this GenAI simulation. Each teacher received a link to access the GenAI simulation platform, where they had up to 10 minutes to review the goals, the scenario background and the instructions, and to prepare for the simulation by reviewing the student work and generating questions and prompts that they could use during their chat. Each teacher then engaged in an interaction with Cecilia conducted via text-based chat, to determine the process that Cecilia used to solve this math problem and to uncover the GenAI student’s reasoning and conceptual understanding. Afterwards, each teacher completed a survey to reflect on their experience and share what they had learned about Cecilia’s thinking and their perceptions of the simulation. This feedback is crucial for evaluating the tool’s effectiveness, learning about the actions that teachers take to elicit student thinking and refining the tool in future work.

Teachers’ engagement with a GenAI digital teaching simulation 

Teachers employed various actions to elicit ideas from Cecilia. Notably, these actions are not limited to virtual environments but are also equally applicable in classrooms with real students, as they help teachers to guide, encourage and challenge students to deepen their understanding. Whether in a digital teaching simulation or a real classroom, these actions can foster student engagement, promote critical thinking and facilitate meaningful learning experiences. Here, we illustrate actions that the teachers used in the simulation to elicit Cecilia’s conceptual understanding and her procedural approach to solving addition problems, which teachers can adapt and use in their classrooms to elicit student thinking about various content topics.

Eliciting Cecilia’s conceptual understanding

We observed three kinds of actions (Table 1) that teachers used to explore Cecilia’s understanding of mathematical concepts related to this topic. First, teachers probed Cecilia’s understanding of mathematical symbols and operations. Second, to gauge her mathematical knowledge, teachers explored Cecilia’s understanding of mathematical concepts beyond addition. Lastly, teachers encouraged Cecilia to reflect on the problem-solving process that she used and explain why she chose specific methods.

Table 1: Teacher actions used in the GenAI simulation to elicit the student’s conceptual understanding

Eliciting Cecilia’s procedural approach

Teachers also used several actions (Table 2) to explore Cecilia’s approach to solving addition problems, focusing on the method that she employed. First, teachers used probes to learn about the visual representations that Cecilia employed to solve the addition problem. Second, teachers explored alternative problem-solving strategies that Cecilia might have used, aiming to assess her flexibility in problem-solving. Third, teachers prompted Cecilia to explain the steps that she used when problem-solving and the reasoning behind them.

Table 2: Teacher actions used in the GenAI simulation to elicit the student’s procedural approach 

Teachers’ perspectives on a GenAI digital teaching simulation 

Most teachers (90 percent) found the responses realistic, resembling those of a real student, and reported that it was easy to engage in conversation. Teachers highlighted several reasons for the simulation’s engaging nature, including the responsive and timely student responses, its authenticity and interactivity, and its usefulness for practising their teaching craft and understanding student thinking. For example, one teacher (P108) explained that ‘I really liked how timely the student’s answers were. I also liked that I could see her work on the screen in addition to the problem and the types of questions I should be asking her.’ Another teacher (P103) commented that they liked ‘seeing the method of how they [Cecilia] approached the problem. It seemed real life like a real student doing the work. They showed their work.’

Most teachers (80 percent) agreed or strongly agreed that the experience of eliciting student thinking closely resembled the work that teachers do in real classrooms, as they had opportunities to ask students questions about their thinking and receive age-appropriate student responses. For example, one teacher (P103) explained how they were ‘able to engage by asking questions’ and that ‘the student provided the way they worked’, which ‘seemed similar to a real interaction with step by step’ details about the student’s strategy. However, two teachers noted ways in which the GenAI simulation did not resemble the work that teachers do, because ‘there is no personal connection between the simulated student and the teacher doing the training’ (P101) and some students would not have the correct answer (P105).

Supporting teachers’ productive use of GenAI digital teaching simulations 

Findings suggest that teachers who took part in the study highly valued the GenAI simulation for practising questioning strategies and interpreting student responses. Teachers also perceived the GenAI simulation as an engaging and useful learning tool with which to practise a critical teaching skill. In closing, we offer several recommendations for how teachers can engage with GenAI simulations productively to support their learning. Teachers should:

• prepare adequately by understanding the simulation goals, reviewing student work, generating possible questions and prompts, and planning actions that they can try out during the simulation
• leverage their prior experiences with real students to determine effective ways in which to probe student thinking
• engage with the GenAI student as if they were a real student and try out various instructional approaches that they might use in a real classroom setting
• reflect on the student’s procedural approach to the problem-solving process and their conceptual understanding
• review the actions that they used in the simulation to identify their strengths and improvement areas.

Acknowledgements: We appreciate the team efforts of Beata Beigman Klebanov, Tom Florek, Tricia Maxwell and Michael Suhan in supporting the development and refinement of the generative AI simulation used in this study.

The examples of AI use and specific tools in this article are for context only. They do not imply endorsement or recommendation of any particular tool or approach by the Department for EducationThe ministerial department responsible for children’s services and education in England or the Chartered College of Teaching and any views stated are those of the individual. Any use of AI also needs to be carefully planned, and what is appropriate in one setting may not be elsewhere. You should always follow the DfE’s Generative AI In Education policy position and product safety expectations in addition to aligning any AI use with the DfE’s latest Keeping Children Safe in Education guidance. You can also find teacher and leader toolkits on gov.uk .