Times are displayed in (UTC-05:00) Central Time (US & Canada) Change
About this paper symposium
| Panel information |
|---|
| Panel 24. Technology, Media & Child Development |
| Paper #1 | |||
|---|---|---|---|
| Preschoolers’ Mathematical Language Learning During Shared Book Reading with an AI Voice Agent | |||
| Author information | Role | ||
| Jisun Kim, Purdue University, United States | Presenting author | ||
| Caroline Byrd Hornburg, Virginia Tech, United States | Non-presenting author | ||
| Koeun Choi, Virginia Tech, United States | Non-presenting author | ||
| Abstract | |||
|
Digital media technologies have been extensively utilized in children’s daily lives and many researchers, educators, and developers have been interested in finding ways to utilize these technologies in educational settings to facilitate early cognitive development. Emerging research suggests that children as young as 3 years can learn from interaction with AI voice agents (Xu et al., 2022). However, there is a paucity of knowledge about how this advanced technology can be used to support young children’s important mathematical language learning, which is a critical domain-specific language highly associated with early numeracy skills and can be supported through repeated exposure to learning materials (Hornburg et al., 2018; Purpura et al., 2021). The purpose of this study was to systematically evaluate the impact of an AI voice agent on preschoolers’ mathematical language learning using a randomized pretest-posttest experimental design. First, we tested whether children’s mathematical language learning differed when reading a math storybook with the AI compared to when reading a non-math storybook with the AI. Second, we examined whether children’s mathematical language learning from math book reading with the AI differed with and without dialogic reading questions. Children (N = 66) aged 3 to 5 years (M = 53.20 months, SD = 8.18, 41% girls) completed five 20-minute one-on-one sessions with the experimenter, within 13 days on average. At the first and last sessions, children completed mathematical language, numeracy, and general vocabulary assessments, as pretest and posttest, respectively. During the intermediate sessions, children were randomly assigned to one of three conditions to read a storybook with an AI voice agent three times: 1) math storybook with dialogic questions, 2) math storybook without dialogic questions, and 3) non-math storybook with dialogic questions. The math storybook incorporated 12 mathematical words (e.g., few, least). The AI voice agent system, developed using the Google Dialogflow CX, reflected the storyline. All dialogic reading questions were added to the system along with the AI voice agent’s contingent feedback. This feedback was developed based on parent-child conversation data from prior storybook research (Purpura et al., 2021). A regression analysis adjusting for child age, general vocabulary, pretest numeracy, and pretest target mathematical language indicated a significant interaction between reading condition and pretest numeracy in predicting children’s posttest target mathematical language (b = 0.13, p = .019). Specifically, children with higher levels of pre-numeracy skills demonstrated greater benefits from simply listening to the math storybook (b = 1.89, p = .012), whereas children with lower levels of pre-numeracy skills showed a tendency to learn better when hearing questions and feedback from the AI voice agent, albeit marginal significance (b = -1.68, p = .066). The difference between math and non-math book conditions and its interaction with pretest numeracy skills were not significant in predicting children’s posttest target mathematical language (ps > .143). Together, we found that AI math storybook reading supports preschoolers’ target mathematical language learning differently based on children’s initial numeracy skills. Findings highlight the importance of content-specific prior knowledge in designing and using AI interaction for children’s learning from repeated reading. |
|||
| Paper #2 | |
|---|---|
| Math Talk in Childhood: Exploring Communication Patterns with AI and Human Partners | |
| Author information | Role |
| Echo Zexuan Pan, University of Michigan, United States | Presenting author |
| Xuechen Liu, University of Michigan, United States | Non-presenting author |
| Trisha Thomas, Harvard University, United States | Non-presenting author |
| Ying Xu, Harvard University, United States | Non-presenting author |
| Abstract | |
|
Introduction Early math learning extends beyond learning how to count or calculate; it also involves understanding and using math language, such as “more” and “equal”. Children often acquire math language through everyday conversations with adults or peers (Thippana et al., 2020). For example, when a caregiver says, “If you take one more apple, we will have an equal number of apples,” a child is exposed to the quantitative concepts of “more” and “equal.” However, the quantity and quality of math language children encounter in everyday conversations can vary widely, leading to an opportunity gap in early math learning (Dearing et al., 2022; Elliott & Bachman, 2018). Generative artificial intelligence (AI) holds the potential to address this gap by engaging children in consistent, targeted math-related conversations (Zhang et al., 2024). To this end, we developed a GPT-4-based conversational agent that can engage children with math-focused storytelling. We then conducted a study to compare the communication quantity and quality between child-AI and child-human conversations. Hypotheses The distinct embodiment and language of the AI partner might result in differences in communication quantity and quality, yet the advanced capacity of large language models to simulate human-like conversations might also yield consistent patterns across conditions. Study Population We recruited 119 children, aged 4 to 9, from a mid-sized city in the Midwest region of the United States. Methods We conducted a randomized controlled trial where children were assigned to engage in math-related conversation with one of three partners: an AI agent, a present human face-to-face, or a human hidden from their view. We coded communication quantity by word count and response rate, and quality by the relevance and coherence of children’s responses to the questions. Results The communication patterns of children and partners are displayed in Figures 1 and 2, respectively. Communication Quantity In terms of children’s total word count, the ANCOVA and post-hoc test revealed that children spoke more words when interacting with a present human compared to both AI (mean difference = 232, p < .001) and hidden human partners (mean difference = 164, p < .001). Children’s response rates also varied depending on the nature of the partner they interacted with, F(2, 107) = 7.83, p < .001. Children interacting with present human partners responded to 95% of the questions posed by their partner, whereas those interacting with hidden human and AI partners responded to 80% of the questions. In terms of partners’ total word count, no condition effect was observed, F(2, 107) = 2.86, p = .062, implying that AI, present human, and hidden human partners contributed comparable amounts of verbal output during conversations. In terms of partners’ math word count, the ANCOVA and post-hoc test revealed that AI partners used 35 more math words than present human partners (p < .001) and 32 more than hidden human partners (p < .001). Communication Quality In terms of children’s response relevancy, the ANCOVA and post-hoc test showed that children interacting with hidden human partners demonstrated higher relevancy in their responses compared to those interacting with AI partners (p = .015). |
|
| Paper #3 | |||
|---|---|---|---|
| Parents’ Perceptions and Attitudes Towards AI Adaptive Learning Systems in 3rd-Grade Math Education | |||
| Author information | Role | ||
| Sofia Aparício, Michigan State University, United States | Presenting author | ||
| Fashina Aladé, Michigan State University, United States | Non-presenting author | ||
| Abstract | |||
|
Introduction The integration of Artificial Intelligence (AI) in education, offers the potential to revolutionize foundational skill development through hyper-personalized learning (U.S. Department of Education, 2023). AI adaptive learning systems tailor content to students’ unique needs, which can enhance learning outcomes, particularly for younger learners. However, the successful implementation of these systems requires acceptance from parents, who play a critical role in shaping their children’s interactions with new technologies (Venkatesh et al., 2012). This study aims to explore parents’ attitudes, concerns, and expectations regarding the integration of AI adaptive learning technologies in Mid-Michigan’s 3rd-grade classrooms, with a specific focus on improving early math education. Theoretical Framework This study is rooted in Piaget's theory of cognitive development, which emphasizes the stages of children's learning, and Bandura's social learning theory, which underscores the role of social influences, such as parental attitudes, in the learning process (Bandura & Walters, 1977; Piaget, 1976). Piaget suggests that 3rd-grade children are at a critical point of transitioning to abstract thinking, while Bandura’s theory emphasizes how parental attitudes shape children’s learning, making it essential to explore parental perceptions of AI integration. Methods This research is being conducted in two phases. First, we are holding qualitative focus groups with 10-12 parents of 3rd-grade children in Michigan. Focus groups will be held between November 3rd and November 16th, 2024, and will include participants from a variety of racial and socioeconomic backgrounds to mirror the demographics of the area. The aim is to gather detailed insights into parents' perceptions, concerns, and expectations regarding the use of AI in math education. For example, parents will be asked questions such as: “How do you perceive the use of AI in helping your child learn math?” and “What concerns, if any, do you have about data privacy or the ethical use of AI in schools?” The data from the focus groups will be analyzed using thematic analysis (Creswell, 2009). Based on the findings from the focus groups, we will then develop a quantitative survey for phase two of the study. The survey will be distributed to a national sample of approximately 600 parents to capture broader trends and correlations in attitudes towards the adoption of AI in schools. The survey will include questions that probe for concerns around privacy, data security, and the perceived effectiveness of AI in enhancing learning outcomes. The survey data will be analyzed using descriptive and correlational statistics. Survey data will be collected in early December, and analysis for both phases will be completed by January 15th, 2025. Results Insights from this multi-method study will contribute to an in-depth understanding of the obstacles and advantages related to incorporating AI into elementary math education. This research will reveal how parental participation and acceptance/hesitancy toward this new technology might impact the effectiveness of AI adaptive learning systems in classrooms. These findings will help shape educational policies and strategies that encourage productive partnerships between schools and parents to improve AI technologies' capacity to promote fair math learning results. |
|||
⇦ Back to session
Understanding Individual and Contextual Factors Related to Children’s Math Learning Using Artificial Intelligence (AI) Technology
Submission Type
Paper Symposium
Description
| Session Title | Understanding Individual and Contextual Factors Related to Children’s Math Learning Using Artificial Intelligence (AI) Technology |