Preschoolers’ learning from social robots: from social to epistemic trust

• In Event: Children’s selective trust across domains, agents, and cultures

Abstract

A recent meta-analysis (Tong et al., 2020) has revealed that both epistemic and social characteristics of the informants are valuable to young children when they learn from testimony. With age, children place greater value on epistemic characteristics when deciding whether to endorse an informant’s testimony (Tong et al., 2020). Using robots as informants can shed additional light on this developmental shift. Most previous research using the selective learning paradigm has pitted two robots against each other (Breazeal et al., 2016; Brink & Wellman, 2020). The present work examined whether children aged 3 and 5 years learned from a competent humanoid robot, Nao, versus an incompetent human agent (Study 1, N=95) or from a competent non-humanoid robot, Cozmo, versus an incompetent human (Study 2, N=89).
These studies tested selective social learning (SSL) using the trust paradigm (Koenig et al., 2004). Testing occurred over Zoom. First, an experimenter introduced three familiar objects (toy car, ball, cup). Next, children watched a video where the robot agent (Nao or Cozmo) consistently labeled the objects correctly, and the human agent consistently labeled the objects incorrectly. The child was then asked what they believed the object was called. Then, the child was asked who (e.g., Nao/Cozmo or the human) said something right or wrong. Next, three unfamiliar objects (blue cylinder, white bulb, red funnel) were presented and labelled by each agent with a nonsense label (e.g., “dax,” “bosa”). After each video, children were asked what they believed the object was called.
The results indicated that a developmental shift occurs from the ages of 3 to 5. While 5-year-old children endorsed the accurate robots’ labels, 3-year-olds were split in their endorse decisions. This finding held regardless of the robot’s morphology, human-looking or non-human looking. This suggests that non-epistemic social cues, such as speech, might be more important when 3-year-old children are deciding whom to learn from.
An adapted version of a naïve biology task (Gottfried & Gelman, 2005) was also administered. Children were shown pictures of the robot (Nao or Cozmo), mechanical objects, and animals. Children were then asked whether something mechanical (e.g., gears, battery) or biological (e.g., organs, muscles) belonged inside the pictured items. This task was administered before the SSL paradigm. A similar developmental change was found here as well. While 5-year-old children correctly assigned both robots with mechanical insides, 3-year-old children assigned the robots mechanical or biological insides at chance. Interestingly, the 5-year-olds choose to trust the robots’ labels for novel objects, even while recognizing the robots as mechanical. This indicates that children may view robots as depictions of social agents by 5 years of age (Clark & Fischer, 2022).
The present study provides a conservative test of epistemic trust as a competent informant is preferred by 5 years of age, regardless of its animacy. This study also confirms the documented developmental shift from younger children’s reliance on social cues to older children’s reliance on epistemic cues.

Authors

• Anna-Elisabeth Baumann, Concordia University
  Presenting Author

• Elizabeth Jessica Goldman, Concordia University

• Alexandra Meltzer, Concordia University

• Aymée Bray Le Métayer

• Diane Poulin-Dubois, Concordia University