Humanoid Robots as Interviewers for Automated Credibility Assessment

摘要

Humans are poor at detecting deception under the best conditions. The need for having a decision support system that can be a baseline for data-driven decision making is obvious. Such a system is not biased like humans are, and these often subconscious human biases can impair people's judgment. A system for helping people at border security (CBP) is the AVATAR. The AVATAR, an Embodied Conversational agent (ECA), is implemented as a self-service kiosk. Our research uses this AVATAR as the baseline and we plan to augment the automated credibility assessment task that the AVATAR performs using a Humanoid robot. We will be taking advantage of humanoid robots' capability of realistic dialogue and nonverbal gesturing. We are also capturing data from various sensors like microphones, cameras and an eye tracker that will help in model building and testing for the task of deception detection. We plan to carry out an experiment where we compare the results of an interview with the AVATAR and an interview with a humanoid robot. Such a comparative analysis has never been done before, hence we are very eager to conduct such a social experiment. This research paper deals with the design and implementation plan for such an experiment. We also want to highlight what the considerations are while designing such a social experiment. It will help us understand how people perceive robot agent interactions in contrast to the more traditional ECA agents on screen. For example, does the physical presence of a robot encourage greater perceptions of likability, expertise, or dominance? Moreover, this research will address the question on which interaction model (ECA or robot) elicits the most diagnostic cues to detecting deception. This study may also prove very useful to researchers and organizations that want to use robots in increasing social roles and need to understand its societal and personal implications.