Value systems of artificial intelligence and university students: theoretical dominance in large language models and religious priority in humans

Abstract

The rapid advancement of artificial intelligence (AI), particularly large language models (LLMs), raises critical questions about the value system these systems appear to reflect in comparison with human values. This study aimed to examine Spranger’s six value types (religious, social, theoretical, economic, political, and aesthetic) as manifested in three LLMs (OpenAI-o1, Gemini-2.0, and DeepSeek-V3), and to compare them with the value system of a sample of students at King Khalid University. A descriptive–comparative design was employed, administering the Study of Values to both groups: 214 students (male and female across academic levels) and the three LLMs, with repeated administrations to the latter to ensure test–retest reliability. Results indicated statistically significant differences in both the prominence and ranking of values across groups. Theoretical values consistently dominated in the LLMs, followed by social, aesthetic, and political values, with religious values ranking lowest. In contrast, students prioritized religious values, followed by theoretical values, while aesthetic values occupied the lowest ranks. Further, significant effects of gender and academic level were observed among students: religious values were more salient among females, theoretical values among males, and aesthetic values among undergraduates. These findings suggest that LLMs project value system shaped by their training data, rather than by human cultural or moral frameworks. The study highlights the importance of integrating culturally diverse value dimensions into AI development and calls for raising students’ awareness of using AI tools in ways aligned with human values. Effect-size estimates further indicated very large human–AI discrepancies, particularly in the religious (d = 2.21) and theoretical domains (d = 1.22).

Ähnliche Arbeiten