Integrating AI into Classroom Discussion Routines: Helping students express their science thinking

• In Event: AI as a Co-Pilot in Science Education: Reimagining Instructional Practice

Abstract

Objective
This study reports on an AI-enhanced Inquiry Sequence (AIS) that helps middle school science teachers guide students to collaboratively answer complex, real-world science problems. We explore how AI can augment teachers’ collaborative learning structures by using AI Dialogs to guide students to develop their ideas and then pairing them to test their ideas in dialogue with a peer.

Perspective
The array of student ideas within a classroom are relevant resources (Hammer, 2000) that enrich sense-making when taken up and built upon by teachers (Fulberg & Silseth, 2022). Building on this perspective, we leverage AI Dialogs that detect and respond to the ideas in students’ explanations (see Bradford et al., 2023 for AI model development and dialog guidance design)
AIS was collaboratively designed with two sixth grade science teachers. It embeds an AI Dialog into a 5-step instructional sequence informed by knowledge integration pedagogy (Linn & Eylon, 2011; Figure 3-1).

Data Sources and Methods
The teachers tested the AIS in a unit on climate change (student N=193). The AIS centered on a question asking students to compare the temperature of a car parked in the sun on a cold day to the outside temperature, an analogy for the greenhouse effect. We analysed the ideas present in students' initial explanations, their revised explanation after the AI Dialog, and their final revision after the think-pair-share (TPS).

Findings
Teachers noted that integrating AI Dialogs into their TPS routine supported students to deepen their thinking before peer discussion. Students added more ideas to their explanations after engaging in the AI Dialog (Figure 3-2). On average, student explanations included 1.46 ideas per student in their initial explanation, 2.26 in their explanation after the dialog, and 2.29 in their explanation after the TPS. The interaction with the AI Dialog supported students to add targeted mechanistic ideas with a 64% increase in total mentions of these ideas from initial explanation to explanation after the AI Dialog. Following the TPS, there was an increase in mentions of the targeted mechanistic idea that heat is trapped in the car (15% increase) and the claim that inside the car is warmer than the outside air (33% increase). After the TPS, students also incorporated misapplied mechanisms, like the idea that conduction of heat explains the temperature inside the car (25% increase) or the idea that metal attracts heat (35% increase).

Significance
Overall, the AIS supported students to engage in peer dialogue and write more elaborated explanations. The AI Dialog effectively elicits more of students’ ideas, preparing them to participate in the TPS. The AI Dialog was particularly effective at surfacing targeted mechanistic ideas, ensuring more students encounter and discuss those ideas during the TPS. The reemergence of misapplied mechanisms following the TPS suggests that students were still grappling with how to integrate ideas from past instruction or their experiences with the targeted mechanistic ideas raised in the AI Dialog. The TPS illuminates where students might need further support to distinguish when ideas hold explanatory power.

Authors

• Allison Bradford, University of California - Berkeley

• Libby F. Gerard, University of California - Berkeley

• Marcia Linn, University of California - Berkeley