TLDR: A study found that public moral judgments of AI applications are systematic and predictable, not random. Five core moral qualities—perceived risk, benefit, dishonesty, unnaturalness, and reduced accountability—collectively explain over 90% of the variance in AI acceptability. Greater personal experience with AI also leads to higher acceptance, largely by influencing perceptions of these moral qualities. This framework offers a tool for anticipating public resistance and guiding responsible AI development.
A groundbreaking study has shed light on the complex question of what makes artificial intelligence (AI) applications acceptable or unacceptable to the public. Moving beyond the idea that public reactions are random, researchers have developed a predictive moral framework that systematically explains how people evaluate AI technologies.
Unpacking Public Opinion on AI
As AI rapidly integrates into various aspects of society, from personal assistance to critical infrastructure, developers and policymakers often struggle to anticipate public moral resistance. This new research, detailed in the paper “What Makes AI Applications Acceptable or Unacceptable? A Predictive Moral Framework,” suggests that these judgments are not arbitrary but are instead driven by a predictable set of moral considerations.
The Five Core Moral Qualities
The study identified five core moral qualities that collectively explain a significant portion of how people judge AI applications: perceived risk, benefit, dishonesty, unnaturalness, and reduced accountability. These qualities act as fundamental building blocks for moral judgment, allowing individuals to systematically evaluate diverse AI uses.
- Risk: Concerns about potential harm to individuals or society.
- Benefit: Perceptions of positive value creation.
- Dishonesty: Related to deception, manipulation, or misrepresentation by AI.
- Unnaturalness: Concerns about violating natural or traditional ways of being.
- Reduced Accountability: Worries about the erosion of human responsibility and oversight.
For example, the application perceived as most risky was AI making decisions about launching conventional weapons, while deploying AI for tedious tasks was seen as creating the most benefit. Having a romantic relationship with an AI robot was deemed most unnatural, and deliberately training an AI with biased data was rated most dishonest. Justifying personal decisions by referencing AI advice was seen as most strongly reducing accountability.
A Systematic Approach to AI Evaluation
In a large, preregistered study involving 587 participants from a U.S. representative sample, researchers used a comprehensive taxonomy of 100 AI applications. These applications spanned personal contexts (like relationships and healthcare) and organizational contexts (such as employment and legal systems), including questions about the moral treatment of AI itself.
The findings were striking: the five core moral qualities collectively explained over 90% of the variance in acceptability ratings across all AI applications. This strong predictive power held true across different domains and even successfully predicted individual-level judgments for applications participants had not previously rated. This suggests that a structured moral psychology underlies how the public evaluates new technologies.
The Role of Experience
The research also explored the impact of personal experience with AI. It found a significant positive relationship between greater personal experience with AI and judging AI applications as more acceptable. This effect was not just a general preference; application-specific experience typically predicted higher acceptance, along with higher perceptions of benefit and lower perceptions of dishonesty, risk, and unnaturalness. Interestingly, the moral qualities were found to mediate a substantial portion of the effect of experience on acceptability, meaning that experience influences how we perceive these core moral qualities, which in turn shapes our acceptance.
Also Read:
- AI Models Outperform Individual Humans in Predicting Everyday Social Norms
- Making AI Predictions More Actionable for Humans
Guiding Responsible AI Innovation
These insights have significant implications for the future of AI development and governance. By understanding which moral qualities drive public acceptance or resistance, developers and policymakers can proactively design AI systems and implement regulations that align with societal values. Instead of broad “AI education,” efforts to build public support could focus on addressing specific moral concerns, such as demonstrating robust accountability mechanisms, ensuring transparency to reduce perceptions of dishonesty, or highlighting concrete benefits to offset perceived risks.
While the study was conducted in the United States, future research will explore cross-cultural variations and how these moral judgments evolve over time as societies gain more experience with AI. This framework provides a powerful tool for anticipating public reactions and fostering responsible innovation in the rapidly expanding field of artificial intelligence.
