Robot Gaze Behavior Affects Honesty in Human-Robot Interaction

Abstract

As the use of humanoid robots proliferates, an increasing amount of people may find themselves face-to-"face" with a robot in everyday life. Although there is a plethora of information available on facial social cues and how we interpret them in the field of human-human social interaction, we cannot assume that these findings flawlessly transfer to human-robot interaction. Therefore, more research on facial cues in human-robot interaction is required. This study investigated deception in human-robot interaction context, focusing on the effect that eye contact with a robot has on honesty toward this robot. In an iterative task, participants could assist a humanoid robot by providing it with correct information, or potentially secure a reward for themselves by providing it with incorrect information. Results show that participants are increasingly honest after the robot establishes eye contact with them, but only if this is in response to deceptive behavior. Behavior is not influenced by the establishment of eye contact if the participant is actively engaging in honest behavior. These findings support the notion that humanoid robots can be perceived as, and treated like, social agents, since the herein described effect mirrors one present in human-human social interaction.

Keywords: deception; eye contact; gaze; honesty; human-robot interaction.

Figure 1

Experimental procedure. At the start of each block, a sequence of colored circles would light up in a randomized order, lighting up for one second each. After the sequence presentation finished, the on-screen robot would attempt to reproduce this sequence by pointing at the appropriate circles in the correct order. The robot would point at the correct color 50% of the time, and at a random incorrect answer 50% of the time. After each attempt, the participant would be asked to provide feedback on the robot's performance, pressing the “T” key to convey that it had picked the correct answer, and the “F” key to convey that it picked the wrong answer. Participants were not required to remember the order of the sequence, as the correct answer was displayed to them on the screen's interface.

Figure 2

Proportion of truthful trials per looking condition, after telling the truth (top) or after telling a lie (bottom). Errors bars represent standard errors. *p < 0.05.

Figure 3

Grand average change in phasic heartrate per trial. t = 0 is the time of responding. The red line denotes deceitful trials, the green line represents truthful trials.

Figure 4

Grand average change in Galvanic skin response. t = 0 is the time of responding. The red line signifies deceitful trials, the green line signifies truthful trials.