Children's information search with social robots: A focus on Children's question-asking behavior

Abstract

As programmable virtual agents/robots are becoming increasingly common, it is essential to examine children's interactions with these technological tools. In this study, we examined how 35 5 to 9-year-old children interact with a social robot (NAO) during a Question Elicitation Task. Following receiving informed consent and demographics information from parents, we tested children individually in their schools. We invited children to ask questions to the robot about animals and objects. The robot provided scripted responses to children's questions across informative and noninformative answer conditions, presented in a fixed order. Children first received informative answers, and then non-informative answers. We simulated smooth conversations with the robot using a Wizard of Oz teleoperation interface using a Python GUI. This interface allowed us to respond to children's questions via a scripted response database generated by GPT-4 that we reviewed and edited for accuracy. If children asked questions beyond the database, this interface also allowed us to regenerate responses from GPT-4 online. Findings showed that children perceived robots as a friendly technological device rather than a living entity and asked the robot information-seeking questions about animals and objects. About 25–30 % of children's information-seeking questions were explanation-seeking. They asked fewer questions when they received non-informative answers than informative answers but this effect disappeared when frequency of questions were adjusted for the length of talk. Children mostly asked about label, function/behavior and physical properties of the items. We discuss these findings within the context of children's learning from social robots.