Abstract

The main goal of my thesis was to investigate how 3- and 5-year-old children learn from robots versus humans using a selective trust paradigm. Children’s conceptualization of robots was also investigated. By using robots, which lack many of the social characteristics human informants possess by default, these studies sought to test young children’s reliance on epistemic characteristics conservatively.

In Study 1, a competent humanoid robot, Nao, and an incompetent human, Ina, were presented to children. Both informants labelled familiar objects, like a ball, with Nao labelling them correctly and Ina labelling them incorrectly. Next, both informants labelled novel items with nonsense labels. Children were then asked what the novel item was called. Children were also asked what should go inside robots, something biological or something mechanical. Study 2 followed the same paradigm as Study 1, with the only change being the robot used, now the non-humanoid Cozmo. Eliminating the human-like appearance of the robot made for an even more conservative test than in Study 1. Both studies 1 and 2 found that 3-year-old children learned novel words equally from the robot and the human, regardless of the robot’s morphology. The 3-year-old children were also confused about both robot’s internal properties, attributing mechanical and biological insides to the robots equally. In contrast, the 5-year-olds in both studies preferred to learn from the accurate robot over the inaccurate human. The 5-year-olds also learned from both robots despite understanding that the robot is different from themselves; they attributed mechanical insides to both Nao and Cozmo over biological insides.

Study 3 further investigated 3-year-olds ambivalence regarding their trust judgements, that is, who they choose to learn from. Instead of word learning, the robot demonstrated competence through pointing. The robot would accurately point at a toy inside a transparent box, and the human would point at an empty box. Next, both informants pointed at opaque boxes and the child was asked where the toy was located. Neither informant demonstrated the ability to speak, as speech is a salient social characteristic. 3-year-olds were still at chance, equally endorsing the robot and the human’s pointing. This suggests that goal-directedness and autonomous movement may be the most important characteristics used to signal agency for young children. The 3-year-olds were also still unsure about the robot’s biology, whereas they correctly identified the human as biological. This suggests that robots are confusing for children due to their dual nature as animate and yet not alive.

This thesis shows that by the age of 5, children are willing and able to learn from a robot. These studies further add to the selective trust literature and have implications for educational settings.