Insights into Moral Reasoning Capabilities of AI: A Comparative Study between Humans and Large Language Models.

Background The primary purpose of this research is to investigate the moral reasoning capabilities of Large Language Models (LLMs), especially to find out possible biases in moral reasoning by ChatGPT and other LLM-based chatbots, and whether they engage in reasoning or simply reproduce information. The study employed an experiment to assess LLMs’ performance on their moral reasoning abilities. Methods Moral reasoning is assessed using the Moral Competence Test (MCT) and the Moral Foundation Questionnaire (MFQ). Both human participants and chatbots based on LLM technology undergo these exams to conduct a comparative analysis of their moral reasoning abilities. Results In the MCT, human participants consistently demonstrate superior performance compared to LLMs. Examining moral foundations through the use of the Moral Foundations Questionnaire (MFQ) uncovers subtle differences, which draw attention to possible prejudices in the moral decision-making procedures of LLMs. Conclusions Our study concludes that LLM-based chatbots have not yet achieved human-level moral reasoning. There is a clear bias whereby individualist values are prioritized over collective ones. This bias may affect how users judge when making moral decisions, thereby shaping public discourse and the decision-making process. It is important to be careful while incorporating LLMs into decision-making processes by focusing on regular evaluation and improvement to ensure the ethical development of AI.