Perceptions of AI mind and moral patiency

Engineering notes

Key topics: autonomous driving, large language model, perception. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

There is growing philosophical, scientific, and media interest in whether artificial intelligence (AI) systems have, or will develop, genuine mental capacities, particularly experiential mental capacities (e.g., feelings of pleasure and pain), and in turn, whether they could be moral patients (i.e., deserving of moral concern). This thesis explores how laypeople think about these questions. In Chapter 2, we reviewed the existing literature and found that people think AI systems have minimal capacity for experience and generally grant them little moral patiency. This contrasts with the moderate degree of agentic mind (e.g., capacity to plan and act) and moral agency (i.e., capacity to carry out morally right or wrong actions) people attribute to AI systems. However, much of this research was conducted with earlier, less advanced AI systems, before the proliferation of chatbots based on large language models. In Chapter 3, we therefore conducted an online experiment (N = 975) in which participants evaluated 14 present-day AI systems (e.g., ChatGPT, Replika, Siri) and 12 non-AI entities for comparison (e.g., inanimate objects, humans, non-human animals). While people attributed a moderate degree of agentic mind and moral agency to the AI systems, attributions of experiential mind and moral patiency remained extremely low, comparable to, or only slightly above, inanimate objects, and far below non-human animals, including an ant. In Chapter 4, we conducted a series of online experiments (total N = 3,455) testing people’s perceptions of experience and moral patiency in increasingly advanced, hypothetical AI systems that might eventually be created. We found that people remained extremely skeptical: even when AI systems were described as being cognitively, behaviorally, and (externally) physically indistinguishable from humans, they were typically attributed only as much experience and moral patiency as an ant. We also found that people’s skepticism was due to AI systems’ non-biological substrate—that they are made of artificial materials such as silicon and wires rather than biological tissue. Chapter 5 introduces “substratism”—the moral devaluation of AI systems based on their non-biological substrate—as a psychological construct, and details five studies (total N = 2,129) which develop and validate an eight-item scale to measure it. We found that substratism is a unique, measurable construct that varies widely across individuals. It is correlated with other AI-related beliefs and behaviors (e.g., perceived threat from AI) and predicts relevant outcomes such as whether people prioritize biological beings over AI systems in moral dilemmas. However, it is weakly correlated or uncorrelated with other prejudices (e.g., racism, sexism, speciesism) and their underlying causes (e.g., social dominance orientation), suggesting that it has some unique psychological explanations. Overall, this thesis highlights people’s persistent skepticism about AI experiential mind and moral patiency, and provides a concise, validated scale to measure the moral devaluation of AI systems due to their non-biological substrate.

Links and sources

• PDF from original source

Comments