Mental Models of Autonomy and Sentience Shape Reactions to AI

TL;DR

This study disentangles two core mental models of AI—autonomy and sentience—and demonstrates that each shapes human reactions along mind perception, moral consideration, and perceived threat. Through three pilot studies and four preregistered experiments using a science-fiction-inspired vignette (Corion), the authors show that sentience robustly enhances mind perception and moral concern, while autonomy more strongly drives perceived threat and certain governance motivations. A meta-analysis confirms that sentience generally exerts larger effects than autonomy, though both capacities meaningfully alter responses to AI. The findings offer design and policy guidance for creating nuanced, domain-specific, and ethically informed human-AI interactions, moving beyond monolithic conceptions of AI minds.

Abstract

Narratives about artificial intelligence (AI) entangle autonomy, the capacity to self-govern, with sentience, the capacity to sense and feel. AI agents that perform tasks autonomously and companions that recognize and express emotions may activate mental models of autonomy and sentience, respectively, provoking distinct reactions. To examine this possibility, we conducted three pilot studies (N = 374) and four preregistered vignette experiments describing an AI as autonomous, sentient, both, or neither (N = 2,702). Activating a mental model of sentience increased general mind perception (cognition and emotion) and moral consideration more than autonomy, but autonomy increased perceived threat more than sentience. Sentience also increased perceived autonomy more than vice versa. Based on a within-paper meta-analysis, sentience changed reactions more than autonomy on average. By disentangling different mental models of AI, we can study human-AI interaction with more precision to better navigate the detailed design of anthropomorphized AI and prompting interfaces.

Figures (6)

• Figure 1: Experiment procedure. Dashed lines show elements that were not explicitly labeled during the experiment. Bold text is quoted from the experiments.
• Figure 2: Autonomy effects (Experiment 1). Note. ns = non-significant, *$p$ < .05, **$p$ < .01, ***$p$ < .001. The means are labeled over each box and displayed as a white point. The black midpoint line is the median. The edges of the box are the lower and upper quartile values. The whiskers are 1.5*IQR (inter-quartile range), the default in R’s ggplot2 package. Dots are outliers.
• Figure 3: Sentience effects (Experiment 2). Note. ns = non-significant, *$p$ < .05, **$p$ < .01, ***$p$ < .001. The means are labeled over each box and displayed as a white point. The black midpoint line is the median. The edges of the box are the lower and upper quartile values. The whiskers are 1.5*IQR (inter-quartile range), the default in R’s ggplot2 package. Dots are outliers.
• Figure 4: Autonomy and sentience effects (Experiment 3). Note. ns = non-significant, + < .10, *$p$ < .05, **$p$ < .01, ***$p$ < .001. Red shows the sentience main effect (dark = sentience, light = no sentience). Blue backgrounds the autonomy main effect (dark = autonomy, light = no autonomy). The H1 and H2 significance tests are shown for Perceived Autonomy and Perceived Sentience. The H3 and H4 pairwise comparisons are shown for Mind Perception, Perceived Harm, Moral Treatment, and Scope of Justice. The significant sentience effect is shown for AI Caution. The means are labeled over each box and displayed as a white point. The black midpoint line is the median. The edges of the box are the lower and upper quartile values. The whiskers are 1.5*IQR, the default in R’s ggplot2 package. Dots are outliers.
• Figure 5: Robust autonomy and sentience effects (Experiment 4). Note. ns = non-significant, + < .10, *$p$ < .05, **$p$ < .01, ***$p$ < .001. Red shows the sentience main effect (dark = sentience, light = no sentience). Blue backgrounds the autonomy main effect (dark = autonomy, light = no autonomy). The H1 and H2 significance tests are shown for Perceived Autonomy and Perceived Sentience. The H3 and H4 pairwise comparisons are shown for Mind Perception, Moral Treatment, and Scope of Justice. The significant autonomy effect is shown for Perceived Harm. There were no significant effects on AI Caution. The means are labeled over each box and displayed as a white point. The black midpoint line is the median. The edges of the box are the lower and upper quartile values. The whiskers are 1.5*IQR, the default in R’s ggplot2 package. Dots are outliers.