Smart Human, Smarter Robot: How Cheating Affects Perceptions of Social Agency

Abstract

Smart Human, Smarter Robot: How Cheating Affects Perceptions of Social Agency Daniel Ullman 1 , Iolanda Leite 2 , Jonathan Phillips 1,3,4 , Julia Kim-Cohen 3 , and Brian Scassellati 1,2 Program in Cognitive Science | 2 Department of Computer Science | 3 Department of Psychology | 4 Department of Philosophy Yale University, New Haven, CT 06520 USA Abstract Human-robot interaction studies and human-human interaction studies often obtain similar findings. When manipulating high-level apparent cognitive cues in robots, however, this is not always the case. We investigated to what extent the type of agent (human or robot) and the type of behavior (honest or dishonest) affected perceived features of agency and trustworthiness in the context of a competitive game. We predicted that the human and robot in the dishonest manipulation would receive lower attributions of trustworthiness than the human and robot in the honest manipulation, and that the robot would be perceived as less intelligent and intentional than the human overall. The human and robot in the dishonest manipulation received lower attributions of trustworthiness as predicted, but, surprisingly, the robot was perceived to be more intelligent than the human. Figure 1. Snapshot of the human manipulation. Keywords: social robotics; trustworthiness; intelligence; intentionality; agency; human-robot interaction Introduction The importance of recognizing social agentic features is not confined to humans, but extends to other living beings and to nonliving social agents. Inferences about the behavior and cognitive capabilities of an entity greatly influence ascriptions of intelligence (Beer, 1990). Human-like properties related to intelligence can be attributed to animated shapes (Scholl & Tremoulet, 2000), virtual agents (Bickmore & Cassell, 2001), and social robots (Bainbridge, Hart, Kim, & Scassellati, 2011; Short, Hart, Vu, & Scassellati, 2010). While the concept of intelligence has been studied extensively with respect to humans, the properties that contribute to perceptions of other animated beings as intelligent, in particular social robots, are still unclear. A better understanding of how people make social attributions to robots will not only allow roboticists to design robots with better social interactive capabilities, but also will add to the knowledge base on features of social agency. Previous research by Short et al. (2010) showed that manipulating high-level behavioral cues, specifically cheating versus not cheating, causes attributions of different mental states to a robot. The researchers investigated attributions of mental state and intentionality to a cheating robot in a game of rock-paper-scissors, a high-level examination that explored how variations in robotic behavior affected perceptions of a robot’s agency. Participants in the two cheat conditions rated the interaction as less fair and honest than those in the third condition, the Figure 2. Snapshot of the robot manipulation. no cheat condition. Furthermore, the results pointed toward greater attributions of mental state to the robot in the cheat conditions than in the no cheat condition. The work by Short et al. (2010) directly motivates the present research. We seek to further this line of research by benchmarking an analysis of agentic cues in a human-robot interaction against a comparable analysis of agentic cues in a human-human interaction. Ultimately, we aim to examine perceptions of intelligence and intentionality in a context of cheating behavior. There are a number of factors that contribute to perceptions of entities as agentic. As stated by Bandura (2001), “To be an agent is to intentionally make things happen by one’s actions.” Researchers over the years have identified features important to ascriptions of agency, including intentionality (Bandura, 2001) and self-propelled, purposeful-looking movement (Premack, 1990; Scholl & Tremoulet, 2000). The concept of agency extends beyond humans; as