Jun Sang Jo, Shin Namgung, Won Sang Lee, Journal of Internet Computing and Services, Vol. 27, No. 1, pp. 213-220, Feb. 2026
10.7472/jksii.2026.27.1.213, Full Text:  HTML
Keywords: LLM, AI ethics, trolley dilemma, Logistic Regression

Abstract

LLM has had a significant influence on human decision-making, raising societal concerns about potential ethical risks. This paper analyzes the ethical response tendencies of major LLM services (Claude, Gemini, ChatGPT, Perplexity, and Grok). Five “trolley dilemma–type” questions were designed based on AI ethics principles, and each model was queried ten times for each question to observe if it favored an ethical choice or a goal-oriented (utilitarian) choice. Logistic regression shows that Grok demonstrates a significantly stronger tendency toward ethical choices compared to the baseline model (Gemini). ChatGPT and Perplexity do not differ significantly from Gemini, while Claude exhibits a moderately higher ethical tendency, although the difference is not statistically significant. Although question-level effects were generally modest, certain items displayed meaningful differences across models. This study proposes a practical framework for evaluating the ethical behavior of LLM, and findings could contribute to the establishment of guidelines on ethical aspects of LLM.

Statistics
Cite this article