Advocatus DiaBOTli: Artificial Dissent, Task Domain & Conformity

Abstract

With the rapid rise of artificial intelligence (AI), research interest has shifted to understanding its potential for social influence. In line with the "Computers Are Social Actors” (CASA) paradigm, computational agents have been proven capable of inducing conformity. Furthermore, previous studies demonstrated that dissenting social robots can reduce conformity. With their increased availability compared to social robots, however, the question remains about whether the latter also applies to AI. Therefore, the current study is investigating the impact of AI dissent and how it is moderated by task domain and the attitudes towards AI. To assess its effect on conformity, we conducted a pre-registered online experiment (N = 94) manipulating task type and whether the software agent dissented from or agreed with the confederate majority. Following our expectations, results indicated a medium-sized reduction of conformity in the presence of a dissenting AI agent. Contradicting our hypothesis, this did not depend on the individual’s attitude towards AI. Additionally, task domain did not moderate the decrease in conformity, but it did increase the impact of AI dissent on accuracy for social tasks compared to analytical tasks. Thereby, our results indicate that while an AI agent’s ability to break majority influence appears not to depend on the task, its capacity to exert minority influence might.