About this paper symposium

---

Panel information
Panel 24. Technology, Media & Child Development

Paper #1
Shall I trust the robot? Children’s selective trust in the human-robot comparisons
Author information	Role
Dr. Wei Deng, Xi’an Jiaotong-Liverpool University, China	Presenting author
Qin Hu, Beijing Normal University, China	Non-presenting author
Abstract
Learning selectively from the reliable and credible informational source enables children to benefit from the complex social environment. However, the sources to gather information in this modern era are more diverse. It raises a potential possibility of learning from new informational agents, such as robots. Compared to human informants, robots may indicate an entirely new form of informational source in terms of assessment to informants’ epistemic properties and the understanding of cognitive conceptualizations in children’s learning development. Nevertheless, most research in the past decades has focused on the trust in human informants or human-to-human interactions rather than other possible informants, such as digital informants or robots. It remains less clear how children’s selective trust in other form of informants develops. The current work investigated how children selectively chose to learn from robots compared to human informants. We conducted two experiments, each examining a specific epistemic property—competence and accuracy. We invited 3- to 6-year-old children (N = 92 in Experiment 1, N = 81 in Experiment 2) to participate in the experiments. Following a cover story and an introduction of each informant (i.e., a human and a robot), children were first presented with the manipulation tasks that had indicated the competence (Experiment 1) or accuracy (Experiment 2) of the informants. After the demonstration of the epistemic properties of informants, children were asked to choose which informant they would trust in a novel puzzle assembling task and a novel object labelling task. The results showed that children exhibited selective trust to the informants based on their epistemic properties: children were able to identify the informants who were more competent or more accurate, and more importantly, they were more likely to trust competent or accurate informants in the subsequent novel tasks. Furthermore, age differences were also found in selective trust to human and robot informants. Children at age 3 were less likely to trust the robot informant, even when the robot informant was as more competent or more accurate in the previous tasks. In contrast, older children chose to learn from the more accurate or more competent informants in the subsequent novel tasks regardless of the identity of the informant. These findings indicated that children were able to treat robots as their informational sources and were able to identify the trustful source based on the display of informant’s epistemic properties. Younger children had less trust in the robot informant regardless of the accuracy or competence of the informants, whereas older children did not discriminate informants based on identity. These findings are consistent with previous findings on children’s trust in digital informants. Current research contributed to the understanding of children’s selective social learning and trust in the human-robot comparison. Furthermore, this line of research would elucidate the state of robots or other artificial intelligence placed in the pedagogical context due to the popularity in usage of new teaching technology.

Paper #2
The contextual relevance hypothesis: How robot appearance influences children’s trust and learning across knowledge domains
Author information	Role
Xiaoqian Li, Ph.D., Singapore University of Technology and Design, Singapore	Presenting author
Jiawen Gou, Central China Normal University, China	Non-presenting author
Hui Li, Central China Normal University, China	Non-presenting author
Abstract
Social robots are emerging as an effective form of educational technology young children, due to their interactive capabilities, physical presence, and often humanoid or animal-like appearance. These features enable more natural and engaging human-computer interactions (Kanda et al., 2004). Research suggests that children aged 3 to 5 readily view robots as a source of information and demonstrate selective trust based on the perceived reliability of robot informants (Breazeal et al., 2016; Li & Yow, 2024). However, how the design of a robot’s appearance influences children’s trust and educational outcomes remains unclear. Based on the principle of Contextual Relevance in designing pedagogical agents (Veletsianos, 2007), it is suggested that the social robots’ appearance design should match the pedagogical context (e.g., the knowledge domain of learning) in order to maximize their teaching effect and improve educational outcomes in children. In this study, we explored this hypothesis by examining the effects of robots’ appearance type (i.e., humanoid vs. zoomorphic) and knowledge domain (i.e., learning human vs. animal knowledge) on preschoolers' learning gains in interactions with a robot tutor. One-hundred-twenty-eight 5-year-old children (M=5.56 years, SD=0.26; 61 girls) participated in this study. They were randomly assigned to one of four experimental conditions (n = 32 per condition), where they learned either human knowledge or animal knowledge from an interactive robot that was either human-like or animal-like (see Figure 1). Before the main learning task, all children completed an interview about their perceptions of humanoid and zoomorphic robots and a selective trust task assessing their preference between humanoid and zoomorphic robots to learn human or animal knowledge. During the learning task, children were taught facts about humans or dogs by the robot tutor in an interactive way (e.g., the robot sometimes asked questions before giving the information) for about 3.5 minutes. Task performance was measured by 11 questions related to the facts children had just learned from the robot. Results from the interview and selective trust task showed that children’s preference between the humanoid and zoomorphic robots varied based on both the knowledge domain (b = 2.90, p < .001) and the attribution of psychological properties (b = -0.60, p < .05). Specifically, children trusted the humanoid robot more for human knowledge and the zoomorphic robot more for animal knowledge. Furthermore, the more psychological properties children attributed to the humanoid robot, the more they trusted it over the zoomorphic one. We used ANOVA for the main analyses to compare children’s learning gains across four different conditions in the learning task. We found a significant interaction effect between robot appearance and knowledge domain on the learning outcomes (F = 7.24, p = .008). While children learned human knowledge more effectively from the humanoid robot than from the zoomorphic robot, they learned animal knowledge more effectively from the zoomorphic robot (see Figure 2). Overall, our findings suggest that 5-year-olds’ selective learning from social robots is influenced by the match between the robot’s appearance and the domain of knowledge. This may be due to children forming impressions and expectations of a robot’s expertise based on its appearance cues.

Paper #3
Children’s mental state attribution to humanoid and non-humanoid robots across theory of mind measures
Author information	Role
Dr. Elizabeth Jessica Goldman, Ph.D., Yeshiva University, USA	Presenting author
Anna-Elisabeth Baumann, University of Calgary, Canada	Non-presenting author
Laetitia Pare, University of Ottawa, Canada	Non-presenting author
Harry Meister, Yeshiva University, USA	Non-presenting author
Jenna Beaudoin, Concordia University, Canada	Non-presenting author
Diane Poulin-Dubois, Concordia University, Canada	Non-presenting author
Abstract
In two experiments, we examined whether young children attribute mental states to humanoid and non-humanoid robots at rates similar to human agents. In Study 1, 4-year-olds (N=102) were administered a subset of questions from the Attribution of Mental States Questionnaire (AMS-Q, Manzi et al., 2020), an interview measure. The children also completed the Wellman and Liu Theory of Mind Scale, a measure that provided context, with either human or robot figurines as the protagonists (see Figure 1). A parental report measure, the Children’s Social Understanding Scale (CSUS, Tahiroglu et al., 2014), measured children’s ToM skills. Results indicate that regardless of condition (Human or Humanoid Robot), children performed similarly on the AMS-Q and the Wellman and Liu Scale. However, some of the subsets of the AMS-Q yielded differences across the conditions. Specifically, the Epistemic subset of the AMS-Q (e.g., “Do you think this robot/person can learn?”) indicated that children assigned to the Human Condition scored higher than those in the Robot Condition. No correlations were found between either measure (Theory of Mind Scale, AMS-Q) and the CSUS. Overall, findings indicate that 4-year-old children attributed a wide range of mental states to a humanoid robot, similarly to how they attribute mental states to human agents. A follow-up study was conducted to examine whether the robot's appearance impacted children’s mental state attributions. In Study 2, 5-year-olds (N=165) were assigned to a Humanoid Robot, Non-humanoid Robot, or Human Condition for the Theory of Mind Scale, with an additional Location False Belief item added (see Figure 2). A Property Projection measure, an interview (Jipson et al., 2016), was also administered, which assessed animacy across various domains (Biological, Psychological, Sensory, and Artifact) of different items (Human, Rodent, Toy Car, Humanoid Robot, and Non-humanoid Robot). The findings of the Property Projection measure indicated that children knew that the Human and Rodent were animate. Children also correctly judged the Toy Car, Humanoid Robot, and Non-humanoid Robot as inanimate. However, looking at the domains individually revealed that children were unsure (at chance level) about both robots' (humanoid robot and non-humanoid robot) psychological and sensory capabilities. Regarding the Wellman and Liu Scale, children attributed mental states similarly regardless of whether they were assigned to the Humanoid Robot, Non-humanoid Robot, or Human Condition. As in Study 1, no significant correlations were observed between the measures and the CSUS. Overall, children across conditions performed similarly on both tasks (Theory of Mind Scale and Property Projection). These findings suggest that 4- and 5-year-olds attribute mental states equally to humans and robots. Additionally, the robot's appearance, specifically whether it was human-like or not, did not impact young children’s mental state attribution. These studies show similar mental state attribution to humans and robots when measured with a task that provides additional context but less mentalizing to robots than humans when measured with an interview (Study 1: AMS-Q; Study 2: Property Projection). This suggests that children’s mental state attribution might be driven by pretense when tested with vignettes, as in the Theory of Mind Scale.

⇦ Back to session