People do not always know best: Preschoolers’ trust in social robots

Abstract

In this paper, we investigated whether preschoolers prefer to learn from a competent robot over an incompetent human using a selective word-learning task (Koenig & Harris, 2004). An adapted naive biology task was administered to assess children’s perception of robots (Gottfried & Gelman, 2005). In Study 1, 50 3-year-olds and 45 5-year-olds (51 males) were tested. The sample was half Caucasian and of mostly high socio-economic status. A social, humanoid robot (Nao) first labeled familiar objects correctly while a human informant labeled them incorrectly. Both informants then labeled unfamiliar objects with novel labels. It was found that 3-year-old children equally endorsed the labels provided by the robot and the human, but 5-year-old children learned significantly more from the competent robot. Interestingly, 5-year-olds endorsed Nao’s labels even though they accurately categorized the robot as having a mechanical inside. In contrast, 3-year-old children associated Nao with biological or mechanical insides equally. In Study 2, 43 3-year-olds (26 males) were tested to determine the impact of human-like appearance on 3-year-olds’ selective trust. This sample was primarily Caucasian and with high socioeconomic status. The procedure was identical, except that a non-humanoid robot, Cozmo, replaced Nao. It was found that 3-year-old children still trusted the robot and the human equally, suggesting that morphology does not play a key role in their selective learning strategies. It is concluded that by 5 years of age, preschoolers show a robust sensitivity to epistemic characteristics, but that younger children’s decisions are equally driven by the agency characteristics of the informant.