Robots, Chatbots, Self-Driving Cars: Perceptions of Mind and Morality Across Artificial Intelligences

Research Background and Issues

• What problems or challenges did the authors identify?
  Artificial Intelligence (AI) systems have rapidly developed and become widespread in recent years, but our understanding of how people perceive AI's psychological traits (e.g., perceived agency) and moral characteristics (e.g., responsibility attribution) remains incomplete. These perceptions are crucial for assessing public trust in AI and defining accountability, especially when AI causes negative impacts.

• Why is this issue important?
  Incorrect attribution of psychological and moral properties may lead to a lack of public trust, shifting responsibility to AI itself rather than its developers, and could influence the use of AI in high-risk scenarios. Understanding these perceptions can help design more responsible and reliable AI systems.

• Research motivation and related work
  A significant body of prior research has shown that people attribute certain psychological and moral traits to AI. However, these studies often focus on individual AI systems or abstract concepts of "AI." This paper provides a more comprehensive picture of psychological and moral perceptions across diverse AI systems by comparing 14 specific AI entities and 12 non-AI entities.

Solutions

• What methods or solutions did the authors propose?
  The authors conducted a pre-registered online experiment where participants rated 26 entities on psychological attributes (agency, experience) and moral attributes (moral agency, moral patiency), covering a wide range of AI and non-AI entities.

• What is innovative about this solution?
  Unlike previous studies, this research systematically compares psychological and moral perceptions across a broader range of entities, exploring differences both among AI systems and between AI and non-AI entities.

• What are the implementation steps and key techniques used?

1. Participant recruitment and sample distribution: 975 participants from the United States were recruited via the Prolific platform, with the sample being representative of demographic attributes.
2. Entity introduction and evaluation: Participants were randomly assigned 13 entities (including AI and non-AI entities), each accompanied by an image and textual description.
3. Measurement method: Psychological and moral attributes of each target entity were quantified using two independent questions, adapted from existing questionnaires.
4. Statistical analysis: Mixed-effects regression models were used to analyze the data and test differences across multiple entities, controlling for confounding variables such as age, gender, and familiarity with AI.

Research Findings

• What specific findings were obtained?

  • AI entities were perceived to have low to moderate levels of agency, while their perceived experience was extremely low (e.g., ChatGPT was considered to have almost no experiential capacity, similar to a rock).
  • In terms of moral attributes, AI's moral agency scores were relatively high, with some AI entities, such as Tesla's autonomous vehicles, being attributed responsibility levels comparable to non-human animals (e.g., chimpanzees).
  • Significant associations were found between psychological attributes (agency and experience) and moral attributes (moral agency and moral patiency), but these associations differed between AI and non-AI entities.

• What advantages does it have compared to existing solutions?
  Compared to prior studies focusing on individual AI systems or abstract "AI" concepts, this paper provides a broad comparison and detailed analysis of attribution patterns between AI and non-AI entities, offering more actionable insights for AI designers.

• What were the experimental or evaluation results?

  • Cross-entity ratings revealed significant differences in agency and experience between AI and non-AI entities.
  • While AI is generally perceived as having low experiential capacity, certain design choices (e.g., more anthropomorphic appearances or emotional expressions) can enhance its moral patiency.
  • Familiarity with AI significantly influenced attributions of moral patiency.

• Limitations and future directions

  • Cultural context: The study was limited to a U.S. sample, and perceptions of AI may vary across cultures.
  • Entity interaction: The experiment relied on images and textual descriptions rather than real-world interactions.
  • Measurement dimensions: The study used two simple questions to measure psychological and moral attributes; future research could expand these dimensions to capture more nuanced characteristics.

In summary, this paper highlights significant differences in psychological and moral perceptions of AI systems, emphasizing the need for designers to focus on normative design to guide public understanding of AI appropriately. This provides valuable guidance for AI developers, particularly in managing moral responsibility and trust in high-risk scenarios. Future research could explore these findings further in cross-cultural samples and real-time interaction settings to validate and extend the results.