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 | |
|---|---|
| AI-Powered Voice Assistants and Children’s Social-Emotional Learning: Usage Patterns Across Developmental Profiles | |
| Author information | Role |
| Dr. Zhiying Yue, Boston Children's Hospital, United States | Presenting author |
| Nicole Powell, Boston Children's Hospital, United States | Non-presenting author |
| David S. Bickham, Boston Children's Hospital, United States | Non-presenting author |
| Michael Rich, Boston Children's Hospital, United States | Non-presenting author |
| Abstract | |
|
Background Over the past two decades, diagnoses of neurodevelopmental conditions and mental health challenges in children have increased worldwide. As digital devices, especially AI-powered tools, become more prevalent in children’s lives, understanding how children with diverse developmental profiles use this technology is essential for developing interventions that support their unique social, emotional, and cognitive needs. This study uses survey data from parents of children aged 3-12 to examine how children with and without mental and behavioral health diagnoses engage with AI-powered voice assistants (VAs) like Siri, Alexa, and Google Assistant, and how these usage patterns relate to parents’ perceptions of VAs’ impact on children’s social-emotional learning (SEL). Methods Data were collected from a national online survey of 1,447 parents of children aged 3-12 in the US through Alchemer. Eligibility required at least one child who actively used VAs. The sample included 49.3% boys, 50% girls, 58.7% White, 17.3% Hispanic/Latino, and diverse racial/ethnic groups. Age distribution was 28.6% (3-5 years), 39.9% (6-9 years), and 31.4% (10-12 years). Parents reported whether their children had been diagnosed with ADHD, (Social) Anxiety, Autism, or Depression (ASAD; N = 459 with at least one reported diagnosis). They also assessed how frequently their children engaged in 17 VA activities (1 = not at all, 6 = more than once a day) and reported on their perceptions of VAs’ impact on children’s SEL, including abilities to connect with others, kindness, civility, emotional understanding, and independence (1 = worsens a lot; 5 = improves a lot). Results Exploratory factor analysis identified three distinct VA activity types (Figure 1): functional/task-oriented (e.g., checking time, messaging; α = .89); recreational/educational (e.g., listening to music, homework help; α = .81); and expressive/relational (e.g., expressing emotions, discussing personal experiences; α = .88). Repeated-measures analysis, controlling for diagnosis and demographics, showed that recreational/educational activities (M= 3.88) were reported more frequently than expressive/relational (M = 3.04) and functional/task-oriented activities (M= 3.06), p < .001. Children with ASAD diagnoses engaged significantly more in all activity types than those without, including functional/task-oriented (M = 3.25 vs. 2.98), recreational/educational (M = 4.07 vs. 3.79), and expressive/relational (M = 3.29 vs. 2.92) activities (Figure 2). Age was positively associated with functional/task-oriented (β = .13, p <.001) and recreational/educational activities (β = .08, p <.01) but negatively with expressive/relational activities (β = -.09, p<.01). Functional/task-oriented (β = .12, p <.001) and recreational/educational activities (β = .10, p< .01) correlated with parents’ positive SEL perceptions, but expressive/relational activities had the strongest association (β = .31, p<.001). Discussion Children with ASAD diagnoses engaged more frequently with VAs across all activity types. The structured, predictable, and non-judgmental nature of VAs may offer a comfortable way for them to engage. With age, children shifted towards task-oriented and recreational activities, while expressive and relational activities declined. However, since expressive and relational activities were most strongly linked to parents’ perceptions of SEL benefits, fostering these types of interactions may be especially beneficial for younger children with ASAD diagnoses to practice social skills and emotional expression. |
|
| Paper #2 | |
|---|---|
| Effects of AI-Enhanced Television Shows on Children’s Science Learning and Verbal Engagement | |
| Author information | Role |
| Dr. Ying Xu, Ph.D., Harvard University, United States | Presenting author |
| Kunlei He, University of California, Irvine, United States | Non-presenting author |
| Julian Levine, University of California, Irvine, United States | Non-presenting author |
| Daniel Ritchie, University of California, Irvine, United States | Non-presenting author |
| Echo Zexuan Pan, University of Michigan, United States | Non-presenting author |
| Andres Bustamante, University of California, Irvine, United States | Non-presenting author |
| Mark Warschauer, University of California, Irvine, United States | Non-presenting author |
| Abstract | |
|
Introduction Educational television programs are valuable learning tools for young children, and their impact can be enhanced when children interact meaningfully with media characters during viewing (Anderson & Hanson, 2017; Ewin et al., 2021). While television programs have traditionally presented content in a one-way manner, recent advances in artificial intelligence (AI) have introduced the possibility of adding conversational interactivity directly to video content. Partnering with PBS KIDS, we developed science-focused interactive videos where the AI-powered main character engaged children by asking questions and providing responsive feedback. To assess the impact of these videos on children’s science learning and verbal engagement, we conducted a randomized controlled trial in which children viewed the videos in one of three formats: 1) Interactive: the AI character asked questions and provided feedback tailored to each child’s responses; 2) Pseudo-interactive: the character asked the same questions but offered generic feedback; and 3) Non-interactive: traditional broadcast version without any interactive elements. Hypotheses 1. Children who watch the interactive videos would show better science learning outcomes compared to those in the pseudo-interactive and non-interactive groups. 2. Children who watch the interactive videos would show higher verbal engagement compared to those in the pseudo-interactive group. 3. Children’s verbal engagement would be positively correlated with their science learning outcomes. Study Population This study involved 246 children, aged 4 to 7 years, recruited from a predominantly Latine charter school in the Western United States. Methods Children viewed two science-focused episodes, each lasting approximately 10 minutes. The first episode demonstrates the principles of aerodynamics, and the second episode introduces the phenomenon of reptile molting. Children’s science learning was measured using a posttest administered after each episode. All questions were read aloud to children, and children responded verbally. Children’s verbal engagement was measured by analyzing their responses during video viewing, focusing on the quantity (rate and length) and quality (relevance and accuracy) of their answers. Results Children in the interactive group outperformed those in the pseudo-interactive and non-interactive groups on science learning (Figure 1), F(2, 243) = 3.44, p = .03. Post-hoc comparisons indicated that the interactive group showed significantly better learning outcomes than the pseudo-interactive group (p = .046) and the non-interactive group (p < .001). Children in the interactive group also showed higher verbal engagement. In terms of response quantity, children in the interactive group responded to 80.80% of follow-up questions compared to 68.80% in the pseudo-interactive group, t(157) = 10.17, p = .002. Although response length did not differ significantly between groups, children in the interactive group gave slightly longer responses during follow-up questions, t(157) = 1.73, p = .19. In terms of response quality, children in the interactive group showed higher response accuracy (t(157) = 4.12, p < .001) and relevance (t(157) = 3.21, p = .002). Additionally, children’s verbal engagement was positively correlated with their science learning outcomes (Figure 2). In particular, stronger correlations were observed for response rate (r = .53), length (r = .27), relevance (r = .62), and accuracy (r = .65) in the interactive group. |
|
| Paper #3 | |||
|---|---|---|---|
| Young Children’s Trust in Novel and Familiar Search Engines | |||
| Author information | Role | ||
| Lauren Girouard, University of Michigan, United States | Presenting author | ||
| Judith H. Danovitch, University of Louisville, United States | Non-presenting author | ||
| Abstract | |||
|
Child development research has long explored children’s intuitions about when various kinds of human and digital sources are reliable sources of information. In the age of generative AI chatbots like Google’s Gemini and OpenAI’s ChatGPT, examining children’s intuitions about search engines provides a foundational understanding of children’s trust in AI chatbots. Search engines increasingly use these chatbots to not only rank order results, but also to summarize them into a single cohesive “best” answer. Therefore, this study asked children about the capability of search engines and the internet to answer questions about different kinds of information. Participants were 120 4- to 8-year-old children (Mage = 6.53, range = 4.10-8.93; 59 girls, 61 boys; 80% White, 10% mixed-race, 6% Black or African-American, 4% Asian-American; 6% Hispanic/Latino). The majority of children in the sample were familiar with Google (84%) and the internet (81%). Children viewed target questions about stable information (e.g., where a lake is located) and changing information (e.g., what sport is being played in the local park this afternoon). They were then asked if one of three digital sources could answer the question. These sources were Google (a familiar search engine), Anu (an unfamiliar search engine), and the internet. Children were also asked if they thought a teacher could answer the question. To explore children’s intuitions about each of the technological informants (Anu, Google, and the Internet), and how they compared to their intuitions about a person, we ran three multilevel models. All three models produced a main effect of age, a main effect of informant, and a main effect of question type. For the novel search engine, there were two significant two-way interactions: an interaction between age and informant and an interaction between age and question type. For Google, there was a single significant two-way interaction between age and informant. For the Internet, there were two significant two-way interactions between age and informant and age and question type (see Table 1 for full list of main effects and interactions). Children began to display a clear preference for Anu over a person between ages 6 and 7, yet preferences for Google and the Internet over a person emerged between ages 5 and 6. Additionally, children under 5 more frequently endorsed the person than Anu, but they endorsed Google and the Internet at roughly the same rates as they do a person (see Figure 1). These findings suggest that children believe that search engines and the internet can provide correct answers to questions. Younger children are wary of search engines that they have not heard of before, but older children tend to apply their intuitions about familiar sources to new and unfamiliar sources. As search engine browsers increasingly rely on generative AI to produce search results, older children’s tendency to readily trust novel search engines in the same way they do more familiar technological tools may mean that they will readily use and trust AI chatbots without thinking critically about the accuracy of the information that these kinds of sources provide. |
|||
⇦ Back to session
AI-Powered Tools for Children: Understanding Engagement, Learning, and Trust in Emerging Technologies
Submission Type
Paper Symposium
Description
| Session Title | AI-Powered Tools for Children: Understanding Engagement, Learning, and Trust in Emerging Technologies |