The AI Double Standard: Humans Judge All AIs for the Actions of One

TL;DR

The paper addresses whether one AI or human agent’s immoral action spills over to attitudes toward other AIs or humans. It uses two preregistered online vignette experiments (Study 1 with a chatbot vs. human assistant and Study 2 with an individuated AI agent and generalized groups) and measures negative/positive moral agency and moral patiency. Key findings show moral spillover from the agent to the agent's group and, for AI, to AIs in general, with an asymmetry in Study 2 where AIs are judged harsher than humans after an AI transgression. This AI double standard has implications for HCI design, trust, and the widespread deployment of autonomous systems, underscoring the need to minimize cross-AI blame and to design transparent, accountable AI teamwork. Overall, the work extends moral spillover literature to AI contexts and highlights how perceptions of AIs can generalize beyond the specific system involved.”

Abstract

Robots and other artificial intelligence (AI) systems are widely perceived as moral agents responsible for their actions. As AI proliferates, these perceptions may become entangled via the moral spillover of attitudes towards one AI to attitudes towards other AIs. We tested how the seemingly harmful and immoral actions of an AI or human agent spill over to attitudes towards other AIs or humans in two preregistered experiments. In Study 1 (N = 720), we established the moral spillover effect in human-AI interaction by showing that immoral actions increased attributions of negative moral agency (i.e., acting immorally) and decreased attributions of positive moral agency (i.e., acting morally) and moral patiency (i.e., deserving moral concern) to both the agent (a chatbot or human assistant) and the group to which they belong (all chatbot or human assistants). There was no significant difference in the spillover effects between the AI and human contexts. In Study 2 (N = 684), we tested whether spillover persisted when the agent was individuated with a name and described as an AI or human, rather than specifically as a chatbot or personal assistant. We found that spillover persisted in the AI context but not in the human context, possibly because AIs were perceived as more homogeneous due to their outgroup status relative to humans. This asymmetry suggests a double standard whereby AIs are judged more harshly than humans when one agent morally transgresses. With the proliferation of diverse, autonomous AI systems, HCI research and design should account for the fact that experiences with one AI could easily generalize to perceptions of all AIs and negative HCI outcomes, such as reduced trust.

Figures (6)

• Figure 1: Summary of results. Across two studies, we found that attribution of negative moral agency, positive moral agency, and moral patiency spills over from an AI assistant to a group of similar AI assistants, and also all AIs. Contrarily, moral attribution to a human assistant only spills over to the group of human assistants, but not all humans. (Images adapted from flaticon.com).
• Figure 2: Flowchart of Study 1 and Study 2 procedure. In Study 1, participants were randomly assigned to read a vignette about a chatbot or human personal assistant who acted immorally or neutrally in their workplace, and rated the assistant and their specific group (chatbots or personal assistants) on negative moral agency, positive moral agency, and moral patiency. In Study 2, the similarity between agent and group was decreased by individuating the agent with a name ("Ezal") and measuring moral attribution of the agent's general group (humans or AIs, in general).
• Figure 3: Effect of agent type and action valence on the agent’s moral attributions. Bars represent the mean and error bars represent the standard error of the mean. Dots represent individual data points. A) Negative moral agency increased in the immoral condition compared to the morally neutral condition, more so for the chatbot than the human assistant. B) Positive moral agency decreased in the immoral condition compared to the morally neutral condition, more so for the human than the chatbot assistant. C) Moral patiency decreased in the immoral condition compared to the morally neutral condition, more so for the human than the chatbot assistant.
• Figure 4: Effect of agent type and action valence on the moral attributions to chatbot and human assistants in general. Bars represent the mean and error bars represent the standard error of the mean. Dots represent individual data points. A) Negative moral agency increased in the immoral condition compared to the morally neutral condition. B) Positive moral agency decreased in the immoral condition compared to the morally neutral condition. C) Moral patiency decreased in the immoral condition compared to the morally neutral condition.
• Figure 5: Effect of agent type and action valence on the agent’s moral attributions. Bars represent the mean and error bars represent the standard error of the mean. Dots represent individual data points. A) Negative moral agency increased in the immoral condition compared to the morally neutral condition, more so for the human than the AI agent. B) Positive moral agency decreased in the immoral condition compared to the morally neutral condition, more so for the human than the AI agent. C) Moral patiency decreased in the immoral condition compared to the morally neutral condition for both agent types, more so for the human than the AI agent.