How do children trust STEM And Non-STEM information from robots? The role of children’s theory of artificial mind (ToAM)

Thesis

Given the integration of robots into educational contexts, understanding how children evaluate information from artificial agents is essential. This dissertation examines how children aged 4 to 6 selectively trust robots when receiving information across STEM (Science, Technology, Engineering, and Mathematics) and non-STEM domains. Building on the Theory of Artificial Mind (ToAM) which is an extension of the Theory of Mind (ToM), the association between ToM/ToAM and selective trust was investigated. Three research questions guided this study: (1) To what extent do children’s selective trust in robots and humans differ? (2) How does the domain of testimony (non-STEM versus STEM) influence children’s selective trust? (3) Do ToAM and ToM relate to children’s selective trust in robot and human informants? The study employed a conflicting informants paradigm wherein 107 Chinese children (M = 5.57 years, SD = 0.58, 44.86% girls) were randomly allocated into two between-subjects conditions: a Nao-accurate condition and a Human-accurate condition. Each child encountered four informant dyads (one human, one robot) who provided conflicting testimony across four domains: one non-STEM domain (object labelling) and three STEM domains (physical science, life science, mathematics). Standardised scales were used to assess children’s ToM and ToAM abilities, with a focus on core components including desire, belief, knowledge, and emotion. Statistical analyses indicated that children consistently preferred accurate informants, irrespective of informant type, particularly in non-STEM domain. However, domain-specific analyses revealed nuanced preferences. Interestingly, within the physical science domain, children demonstrated notable uncertainty; they were inclined to nominate robots despite their inaccuracies yet endorsed information provided by humans and deemed humans as reliable source. Lastly, in the Nao-accurate condition, children’s ToM and ToAM scores both negatively predicted selective trust, suggesting that children with stronger cognitive functions were more cautious about trusting robots. In contrast, no significant relationships emerged in the Human-accurate condition. Collectively, these findings deepen our understanding of how young children evaluate informational reliability in AI-integrated early STEM education and highlight their developing cognitive sophistication.

Authors: Mao, K

• DOI: 10.5287/ora-amer5m2x4
• Type of award: MSc
• Level of award: Masters
• Awarding institution: University of Oxford
• Language: English
• Keywords: selective trust; theory of mind; robot-child interaction; theory of artificial mind; artificial agent; human-computer interaction; CDE; CDE2024; trust; robot; preschooler
• Subjects: Human-computer interaction; Developmental psychobiology; Psychology; Educational psychology; Education (Preschool); Social robot; Human-robot interaction; Education; Child psychology; Children