The target article aims to address the puzzle of social robots: People interact with robots as if they were humans despite knowing that social robots are artifacts. This apparent cognitive dissonance is explained by the fact that people construe social robots not as social agents per se but as depictions of social agents. Such decoupling has been documented to account for adults' attribution of intentionality to inanimate agents ranging in abstractness from geometric figures to puppets to humanoid robots. Clark and Fischer (C&F) conclude that what children understand about social robots at each age remains an open question. We review the substantial body of work that addresses whether children are aware of the fictional nature of robots. We demonstrate that even infants interact with robots as if they were social agents but that the dual orientation toward robots, understanding them as artifacts, gradually develops during the preschool years.

It is well established that children display the same behaviors toward robots and people. For example, even infants follow the gaze of robots (Meltzoff, Brooks, Shon, & Rao, Reference Meltzoff, Brooks, Shon and Rao2010; Mwangi, Barakova, Diaz, Mallofre, & Rauterberg, Reference Mwangi, Barakova, Diaz, Mallofre and Rauterberg2018; O'Connell, Poulin-Dubois, Demke, & Guay, Reference O'Connell, Poulin-Dubois, Demke and Guay2009; Okumura, Kanakogi, Kanda, Ishiguro, & Itakura, Reference Okumura, Kanakogi, Kanda, Ishiguro and Itakura2013). Children also imitate social robots, but until age 6, less so than they do humans (Itakura, Okanda, & Moriguchi, Reference Itakura, Okanda, Moriguchi, Itakura and Fujita2008; Schleihauf et al., Reference Schleihauf, Hoehl, Tsvetkova, König, Mombaur and Pauen2021; Sommer et al., Reference Sommer, Davidson, Armitage, Slaughter, Wiles and Nielsen2020, Reference Sommer, Slaughter, Wiles, Owen, Chiba, Forster and Nielsen2021). Some studies have shown that young children can learn new information directly from social robots (Moriguchi, Kanda, Ishiguro, Shimada, & Itakura, Reference Moriguchi, Kanda, Ishiguro, Shimada and Itakura2011; Okumura et al., Reference Okumura, Kanakogi, Kanda, Ishiguro and Itakura2013) and appear to use similar mechanisms when learning from humans or robots (Stower, Calvo-Barajas, Castellano, & Kappas, Reference Stower, Calvo-Barajas, Castellano and Kappas2021). Children as young as 3 years can learn new words from robots but prefer to learn from a robot that has previously demonstrated accuracy (Brink & Wellman, Reference Brink and Wellman2020). When social (e.g., morphology, agency, animacy) and epistemic (e.g., expertise, competency) characteristics are pitted against one another, 5-year-olds prefer to learn new words from a competent robot over an incompetent human, whereas 3-year-olds are split about whom to trust (Baumann, Goldman, Meltzer, & Poulin-Dubois, Reference Baumann, Goldman, Meltzer and Poulin-Dubois2022).

C&F report that the more social cues robots display, the more competent they are judged to be by adults. There is also evidence to support this conclusion in children. For example, infants are more likely to follow a robot's gaze if the robot acts in a social and communicative manner (Itakura et al., Reference Itakura, Okanda, Moriguchi, Itakura and Fujita2008; Meltzoff et al., Reference Meltzoff, Brooks, Shon and Rao2010; Peca, Simut, Cao, & Vanderborght, Reference Peca, Simut, Cao and Vanderborght2016). Children also prefer to interact with and learn from a robot that displays contingent non-verbal social cues (i.e., gaze following) over a non-contingent robot (Breazeal et al., Reference Breazeal, Harris, DeSteno, Kory Westlund, Dickens and Jeong2016). Interestingly, cues such as goal-directedness and speech may be more important than morphology in determining how children affiliate. For instance, 3-year-old children learn equally well from a humanoid (Nao) and a non-humanoid-looking robot (Cozmo) (Baumann et al., Reference Baumann, Goldman, Meltzer and Poulin-Dubois2022).

Evidence that children can learn from and interact with robots as they do with humans is not sufficient to conclude whether children view robots as depictions of social agents. For this, how children conceptualize and categorize robots needs to be examined. Several tasks have been designed to answer this question, including interviews and a naïve biology task. In particular, interviews can assess children's perceptions of robots across many domains (Beran, Ramirez-Serrano, Kuzyk, Fior, & Nugent, Reference Beran, Ramirez-Serrano, Kuzyk, Fior and Nugent2011; Chernyak & Gary, Reference Chernyak and Gary2016; Goldman, Reference Goldman2021; Jipson & Gelman, Reference Jipson and Gelman2007; Manzi et al., Reference Manzi, Peretti, Di Dio, Cangelosi, Itakura, Kanda and Marchetti2020). The interview questions children are asked typically include: Mental (e.g., Can the robot think?), perceptual (e.g., Can the robot see?), social (e.g., Could you trust the robot with a secret?), emotional (e.g., Does the robot have feelings?), and biological (e.g., Is the robot alive?). A recent meta-analysis revealed that age is a factor in whether children attribute mental states to robots (Thellman, de Graaf, & Ziemke, Reference Thellman, de Graaf and Ziemke2022). Although the findings were mixed, most studies reported that people of all ages attribute mental states to robots. Some studies indicated a stronger attribution of mental states to robots in younger children, which lessened as children got older. The literature suggests that children tend to anthropomorphize social robots and that by age 5 children, like adults, recognize that robots are artifacts but still attribute mental states to them.

Another way to assess children's conceptualization of robots (e.g., whether children view robots as artifacts) is to ask them if robots are alive. For example, Kim, Yi, and Lee (Reference Kim, Yi and Lee2019) asked 3-, 4-, and 5-year-olds to make an animacy judgment about a humanoid robot (Vex) (e.g., Is it alive or not alive?). The older children were less likely to say the humanoid robot was alive than the younger children. As interviews can only be used with older children, other methods are required to assess whether children depict robots as social agents. There is evidence that non-verbal infants expect objects that act like animals (e.g., self-propulsion, vocalizations) to have an inside rather than be hollow (Setoh, Wu, Baillargeon, & Gelman, Reference Setoh, Wu, Baillargeon and Gelman2013). Future work could build upon these findings by testing infants in a related task with robots. In a similar vein, recent work has examined how children conceptualize robots with a naïve biology task (Goldman, Baumann, & Poulin-Dubois, Reference Goldman, Baumann and Poulin-Duboisin press). Using a modified version of Gottfried and Gelman's (Reference Gottfried and Gelman2005) task, children were shown images of robots, unfamiliar animals, and artifacts and asked to select whether something biological (e.g., heart) or mechanical (e.g., gears) belonged inside. A developmental shift in children's categorization of social robots was found: 5-year-olds believed that a humanoid robot (Nao) had mechanical insides, but 3-year-olds equally attributed mechanical and biological insides to the humanoid robot. This finding held when 3-year-olds were presented with a non-humanoid robot (Cozmo) (Goldman et al., Reference Goldman, Baumann and Poulin-Duboisin press).

Although more work is needed, existing research suggests that by the age of 5, children recognize that robots are not alive yet still attribute epistemic (Baumann et al., Reference Baumann, Goldman, Meltzer and Poulin-Dubois2022; Stower et al., Reference Stower, Calvo-Barajas, Castellano and Kappas2021) and social (Breazeal et al., Reference Breazeal, Harris, DeSteno, Kory Westlund, Dickens and Jeong2016) characteristics to them. Thus, preschool children treat robots as depictions of social agents as they interact with and learn from robots while still recognizing them as inanimate objects.