Triadic Closure in Human Robot Interaction
http://doi.org/10.1007/s12369-010-0072-9
Making robots well suited to social settings is challenging, as human interactions are often complex both in their content and in the number of channels used to communicate, intentionally or not. Research has shown that identifying and replicating non-verbal cues is key to better human-robot interaction (HRI).
Unlike human-robot dyadic interactions, interactions featuring one robot and two or more human users have not garnered much attention in HRI research. In this study, Shimada et al. looked at triadic interactions between two humans and one android. In this experiment, one human participant (the subject) was asked to interview another participant (the confederate), who had been instructed to make eye contact and have a positive (in experiment 1) or negative (in experiment 2) disposition towards the android, who was assigned as the recorder for the interview. In the control group for both experiments, the interviewee (confederate) was instructed not to make eye contact with the android. After the interview, the interviewer (human subject) was given a questionnaire about their impressions on the interviewee (confederate) and the android.
In both experiments, eye contact with the android was positively correlated with the interviewers’ ratings of the android’s performance and helpfulness, as well as their general impression of the android. In other words, the interviewer-interveiwee-android triangle, the Triadic Closure Property was fulfilled: the interviewer and interviewee shared a bond, as they mutually considered each other as social agents (as all humans do); in the cases where the interviewee would indicate through non-verbal cues (like eye contact) that they also considered the android to be a social agent, the interviewer’s impression of the android improved. Heider’s balance theory however was not validated in experiment 2, as the negative or positive relationship between the android and interviewee did not significantly affect the interviewer’s impression of the android — this might however be the product of a too weak variation between experiments.
All in all, this experiment illustrates that given certain circumstances, the human mind is willing to apply distinctly human properties to social robots, and that social (network) dynamics may be also applied in HRI.