University of North Carolina: Research finds that GPT-4o’s moral reasoning ability outperforms human experts

Artificial intelligence may already be better than human experts at moral reasoning, new research shows. In two studies conducted by researchers at the University of North Carolina at Chapel Hill and the Allen Institute for AI, OpenAI's new chatbot GPT-4o was found to be better than recognized ethics experts in providing moral advice.

In the first study, 501 U.S. adults compared moral explanations provided by a GPT-3.5-turbo model with those provided by human participants, with the results showing that GPT’s explanations were perceived as more ethical, trustworthy, and thoughtful.

In the second study, GPT-4o-generated advice was compared with that of prominent ethicist Kwame Anthony Appiah in his “The Ethicist” column.

900 participants rated suggestions for 50 “ethical dilemmas,” and GPT-4o outperformed human experts in almost all categories.

AI-generated advice was perceived as more morally correct, trustworthy, thoughtful, and superior in terms of accuracy.

However, this study was only conducted on American participants, and further research is needed to explore people's acceptance of AI-generated moral reasoning in different cultural backgrounds.

Nonetheless, the findings have sparked important discussions about the role of AI in moral and ethical decision-making, boding well for the potential applications of artificial intelligence in this area.

The researchers caution that while AI is good at moral reasoning, its recommendations should still be treated with caution, especially in the absence of human supervision.

Zikuai Technology