Item-level implicit affective measures reveal the uncanny valley of robot faces

Abstract

• The recent explosions of AI technologies have dramatically increased opportunities to interact with artificial agents, such as AI-powered chatbots, avatars, and robots. • Socio-emotional impact of such interactions is a key consideration for acceptance of these technologies in the future. • The uncanny valley effect is a psychological phenomenon that arises in human-robot interaction, which receives mixed support from recent studies. • The present study is the first study to use behavioural affective paradigms to measure implicit reactions to robots’ appearances, and the results demonstrated the uncanny valley effect emerging in the implicit affective reactions. • By finding the uncanny valley effect, the results indicate that these behavioural measures are robust and effective in revealing affective reactions to individual items rather than categories of items. As the opportunity to interact with humanoid robots and virtual avatars increases, the emotional impact of the interaction with these artificial agents becomes an important consideration. The uncanny valley effect is a psychological phenomenon relevant to such a consideration. Although the uncanny valley remained untested for several decades, recent empirical studies confirmed the uncanny valley effect when human observers rated their liking of robots’ faces. To uncover the uncanny valley in behavioral measures of affective response, the present study used two implicit affective tasks, affective priming and single-category IAT. Positivity scores for each of the images of robot faces were derived and were plotted against the humanness rating of the robot faces. The results demonstrated the uncanny valley effect in these implicit behavioral measures. The finding indicates the effectiveness of using these implicit measures to assess affective responses to individual items rather than to groups of items, and it suggests the potential of these behavioral paradigms for wider application outside laboratory research.