Clarifying Public Sector Ethics: Neutralization of Police Violence by Citizens and Generative AI

Abstract

The increased use of (generative) AI in public services puts pressure on public accountability of decision-making, i.e., responsibility for decisions and their consequences could be attributed to the AI-decision support systems. Hence, questions on the ethicality of the decisions arise. We first use an experimental design to analyze the extent that people neutralize police violence. Second, we compare these results with answers from generative AI (ChatGPT). We assess whether neutralization happens as a result of (1) an AI advised arrest of a passerby, (2) the actual background of the passerby (i.e., whether the passerby had a criminal background or not, as verified by a post-hoc evaluation), and (3) the severity of the injury caused by a violent arrest. We find that both humans and generative AI attribute high importance to the actual background of the passerby (actual criminal or not) to classify behaviors as ethical or unethical. We find that humans do not neutralize police violence as a result of AI-decision support, while in contrast, generative AI regularly and explicitly uses this as an argument to classify behavior as ethical or unethical.

Ähnliche Arbeiten