The guilt machine: Behavioral confirmation in moral human-robot interactions

Abstract

The present thesis project concerns a two-phase study examining the hypothesized existence of behavioral confirmation during interactions between police interrogator robots (as perceivers) and human suspects (as targets) during a mock criminal interrogation. 20 participant-suspects were asked to read a mock theft scenario of which they were either innocent or guilty. For the interviews, interrogator robots were equipped with question sets that were either innocence-presumptive or guilt-presumptive in a 2 x 2 (suspect guilt status x interrogator expectation) design. The interviews were recorded and presented to independent observers who had no knowledge of the conditions or manipulations. Observers rated suspects as being more defensive and denying harder when interviewed by a guilt-presumptive robot. The robots were seen as more pressuring and trying harder to get a confession when the suspect was truly innocent rather than guilty. However, observers did not judge suspects as being more guilty, regardless of interrogator presumption or actual suspect guilt status. Implications of the results for the future of moral human-robot interaction are discussed.