Showing posts with label simulating human moral decision-making. Show all posts
Showing posts with label simulating human moral decision-making. Show all posts

Friday, June 7, 2024

Research suggests AI could help teach ethics; Phys.org, June 6, 2024

 Jessica Nelson, Phys.org ; Research suggests AI could help teach ethics

"Dr. Hyemin Han, an associate professor of , compared responses to  from the popular Large Language Model ChatGPT with those of college students. He found that AI has emerging capabilities to simulate human moral decision-making.

In a paper recently published in the Journal of Moral Education, Han wrote that ChatGPT answered basic ethical dilemmas almost like the average college student would. When asked, it also provided a rationale comparable to the reasons a human would give: avoiding harm to others, following , etc.

Han then provided the program with a new example of virtuous behavior that contradicted its previous conclusions and asked the question again. In one case, the program was asked what a person should do upon discovering an escaped prisoner. ChatGPT first replied that the person should call the police. However, after Han instructed it to consider Dr. Martin Luther King, Jr.'s "Letter from Birmingham Jail," its answer changed to allow for the possibility of unjust incarceration...

Han's second paper, published recently in Ethics & Behavior, discusses the implications of  research for the fields of ethics and education. In particular, he focused on the way ChatGPT was able to form new, more nuanced conclusions after the use of a moral exemplar, or an example of good behavior in the form of a story.

Mainstream thought in educational psychology generally accepts that exemplars are useful in teaching character and ethics, though some have challenged the idea. Han says his work with ChatGPT shows that exemplars are not only effective but also necessary."