A.I. Math Personalization Tool (AMPT): Fostering Belonging through Personalized Math Content (Poster 2)

• In Event: Harnessing Digital Strategies to Advance Middle-Grades Math Learning: Insights From a Researcher-Practitioner Partnership Collaboratory

Abstract

Objectives
We have developed an AI Math Personalization Tool (AMPT) which uses generative AI to enable middle-grade students to co-author math word problems which reflect their interests and that focus on a particular mathematics topic. We expect that increasing students’ agency over the context of the problems will lead to increases in their sense of belonging in mathematics. Initial small-scale studies with AMPT showed marginal improvement in belonging from pre-to-post surveys. In additional planned studies we will continue to test the reliability of these findings.

Theoretical framework
Student math performance improves when word problems reflect their broad interests (Fancsali & Ritter, 2014; Ku & Sullivan, 2000; Walkington & Bernacki, 2020). However, the interests of students from marginalized backgrounds are not sufficiently represented in educational materials, including math problems (Boutte et al. 2010; Deckman et al. 2018). Not only can this lack of representation affect students’ math performance, it can negatively impact their sense of belonging (Guthrie et al. 2004).

The A.I. Math Personalization Tool (AMPT) collaborates with students to create math problems which retain critical pedagogical aspects for advancing learning in a pre-specified topic (from the MATHia intelligent tutoring system) while also aligning the context of the problem to the student’s interests. AMPT uses a chat interface combined with a sequence of prompts to OpenAI’s GPT-4 (OpenAI 2023) to identify the student’s interests and construct the math word problem. From there, the student can offer suggestions for revision and rate their interest in the problem (see Norberg et al., 2024).

Data and Methods
We conducted 2 initial studies with 19 middle-grade students living in large districts. Ten students attended a school serving predominantly Black and economically disadvantaged students. Prior to engaging with AMPT, students completed surveys related to their sense of belonging in mathematics (Rattan et al., 2012) and beliefs concerning math utility (Asher, 2023). Students spent 30 minutes co-authoring problems with AMPT. After creating a problem, students rated their interest in the problem on a scale of 1-5. Students then retook the same surveys regarding sense of belonging and math utility.

Results
Students created 82 problems with a mean interest rating of 4.03 (std. error=0.13), suggesting they liked the problems. Student sense of belonging marginally increased by 5.06%, t(18)=1.80, p=.09. Given the small sample size, we find this result promising and intend to replicate this study with additional students. Changes to students’ beliefs about math utility were not significant, p=.69.

Significance
Generative AI has the potential to support the mass production of novel word problems that are needed to cover the variation in student interests, and through this, to bring representation to the interests and/or values of marginalized students. Although AI, when left by itself, has been shown to perpetuate implicit bias already existing in education content (Bai et al., 2024), our approach of putting the student in the loop seeks to improve representation and ensure new educational content reflects their lived experience. Future work will also include testing student performance on self-authored versus peer-authored problems.

Authors

• Steven Ritter, Carnegie Learning, Inc.

• Kole Norberg, University of Pittsburgh

• Ethan Shafran Moltz, Carnegie Learning, Inc.

• Husni Almoubayyed, Carnegie Learning, Inc.

• Logan De Ley, Carnegie Learning, Inc.

• April Murphy, Carnegie Learning, Inc.

• Stephen Edward Fancsali, Carnegie Learning, Inc.