Exploring how AI Tools Can Support Teacher Learning with Formative Feedback on Discussions

William R. Penuel

For more than two decades, research on academically productive talk has pointed to ways that well-structured discussions can “build the mind,” that is, help students develop understandings of disciplinary core ideas, whether in literature, mathematics, or science. Engaging in such talk also supports the development of repertoires for engaging in civic dialogue in ed de  deliberative spaces. Scholars Sarah Michaels and Cathy O’Connor developed the Talk Moves Primer to help science teachers learn to use “moves as tools” to foster this kind of academically productive talk in classrooms.

But as participants in our third AI session noted, it’s not easy to facilitate productive academic talk in science classrooms. These science educators have worked collectively with thousands of teachers seeking to engage their students in rich discussion. They note that many students aren’t used to making their thinking visible through talk and that teachers sometimes feel pressure to focus on eliciting correct answers rather than promoting argumentation due to time constraints. They pointed out that equitable and productive academic talk requires; a different culture in the classroom, high-quality instructional materials, and tools for supporting teachers and learners in this valuable endeavor. 

Addressing the need for more academically productive talk is important, because standards like the Next Generation Science Standards include science and engineering practices such as engaging in argumentation with evidence that require students to make their thinking visible and build on the ideas of their peers out loud.

In this session, we explored two applications of AI that are designed to help teachers become more comfortable facilitating academically productive talk. Both operate by providing feedback to users so they can improve their practice, like the CoBi application we were introduced to in Session 2. 

Talk Moves Application

Dr. Jennifer Jacobs of the Institute of Cognitive Science at CU Boulder shared her team’s development of an AI application designed to give mathematics teachers feedback related to the talk moves they use in class discussions. The team of learning scientists, mathematics educators, and computer scientists developed an application that draws on AI tools for automatic speech recognition and natural language processing, math education theory, and teachers’ self-recorded video to inform personalized feedback.

Most current methods for providing teachers with detailed feedback on their talk moves require highly trained observers. These observers use qualitative research methods to manually code transcripts to identify talk moves or they provide one-on-one expert coaching. The TalkMoves application was designed to automate and scale up the process of detecting talk moves and other classroom discourse practices, enabling teachers to receive immediate and accessible information about their mathematics lessons. The application consists of three interrelated components: a cloud-based big data infrastructure to manage and process recordings, deep learning models that reliably detect talk moves, and a user interface that delivers personalized feedback to teachers on their discussion strategies, both for individual teaching episodes and over a series of episodes.

The team investigated the use of the Talk Moves system with mathematics teachers and discovered that teachers like the system. They found that teachers who utilized the system started incorporating a broader range of talk moves in their discussions. Currently, the team is adapting the Talk Moves system to support volunteer tutors in a high-intensity tutoring program, aiming to enhance small group engagement and learning effectiveness in mathematics tutoring sessions.

TeachFX 

Jamie Poskin, CEO of TeachFX, shared a widely available commercial tool that provides teachers with feedback on their classroom discussions. The TeachFX program uses voice AI to analyze recorded classroom conversations. One of the key features of the TeachFX program is its ability to show the balance of talk between teachers and students. It also highlights silences in the discussion. These patterns can be shown within a particular “lesson snapshot" (see below) and  also tracked over time.

Lesson Snapshot from TeachFx

TeachFX uses four types of AI in its application. 1) Speaker diarization identifies who is speaking and when they are speaking. This is especially important with multiple speakers to identify the patterns of their discussion. 2) Automatic speech recognition (ASR) turns human speech into readable text. 3) Natural language processing (NLP) extracts information, categorizes it, and provides insights. 4) Generative AI creates original text, images, and other media based on prompts.

Poskin cited experimental research by other research teams with tools similar to TeachFX as strong evidence that automated feedback like that provided by  TeachFX can enhance teacher practice and student outcomes.

Discussion Prompts:

1. How might feedback on classroom discussion support teachers in becoming more effective orchestrators of equitable and productive academic discussions?

2. What conditions and professional learning opportunities need to be in place to support teachers in learning how to use such tools effectively to promote more equitable classroom discussions?