Eye Gaze Behavior as a Guilty Knowledge Test: Initial Exploration for Use in Automated, Kiosk-based Screening

This paper presents findings from an eye tracker experiment based on the guilty knowledge test (GKT) paradigm. We conducted a two-group, posttest-only, randomized initial experiment in which the guilty group constructed an improvised explosive device (IED) and then was screened by an embodied conversational agent (ECA). Participants in the control condition went straight to screening. During the screening interview, participants were shown three images. The second image was a modified picture of an IED similar to the one constructed by the guilty participants. Eye gaze behavior was dramatically different between the groups. A classification algorithm correctly classified 100% of the participants as either part of the guilty or control group.