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

The rapid evolution of artificial intelligence (AI), particularly large language models (LLMs), is prompting critical examinations of the value systems these technologies appear to reflect in contrast to human values. A recent study conducted at King Khalid University delved into this complex issue, comparing the value orientations of three prominent LLMs—OpenAI’s ChatGPT-o1, Google’s Gemini-2.0, and DeepSeek-V3—with those of 214 university students. The research employed Spranger’s six value types—religious, social, theoretical, economic, political, and aesthetic—to map these divergences and convergences. The findings indicate statistically significant discrepancies in both the prominence and ranking of values, suggesting that LLMs’ projected value systems are primarily shaped by their training data rather than human cultural or moral frameworks.

AI’s Value Landscape: Theoretical Dominance and Religious Scarcity

The study utilized a descriptive-comparative design, administering the "Study of Values" (SOV) instrument to both the LLMs and the student sample. To ensure reliability, the SOV was administered multiple times to the LLMs. Results revealed a consistent pattern across the AI models: theoretical values emerged as the most prominent, followed by social, aesthetic, and political values. Religious values, notably, ranked lowest among the LLMs. This theoretical leaning is a significant finding, suggesting that the vast datasets used to train these models—often rich in scientific literature, technical documents, and broad factual information—predominantly imbue them with a value system that prioritizes knowledge acquisition, empirical investigation, and logical reasoning.

Human Values: A Foundation of Faith and Intellect

In stark contrast, the university students surveyed prioritized religious values, with theoretical values securing the second-highest rank. This highlights a fundamental difference in foundational motivations. For the student population, spiritual and faith-based principles appear to play a more central role in their value hierarchy. Aesthetic values, interestingly, occupied the lowest ranks among the students, a finding that may reflect their stage of academic development or cultural context.

Demographic Influences on Student Values

The research also explored the influence of demographic factors on student values. Significant effects of gender and academic level were observed. Female students tended to exhibit a greater salience of religious values, aligning with some previous cross-cultural studies that indicate a tendency for women to report higher religious and spiritual values. Conversely, male students scored higher in theoretical values, reinforcing a pattern observed in earlier research where males often show a stronger inclination towards intellectual and analytical pursuits.

Furthermore, academic level played a discernible role. Undergraduate students, compared to their postgraduate counterparts, showed a higher emphasis on aesthetic values. This could be attributed to developmental stages, with younger students potentially exhibiting greater interest in artistic expression and appreciation, while more advanced students, particularly doctoral candidates, increasingly focus on specialized theoretical and research-oriented values. Doctoral students, in particular, showed a greater prioritization of theoretical values, suggesting a deepening engagement with academic inquiry and critical thinking as they progress in their studies.

Quantifying the Human-AI Value Divide

The study employed effect-size estimates to quantify the magnitude of differences between human and AI value systems. The results indicated very large human-AI discrepancies, particularly in the religious domain (d ≈ 2.21) and the theoretical domain (d ≈ 1.22). These large effect sizes underscore that the differences are not merely statistically significant but also practically meaningful, representing substantial gaps in how values are prioritized. The substantial deficit in religious values within AI outputs, compared to their prominence in the human sample, is particularly noteworthy, suggesting that the "spiritual" or "faith-based" dimensions of human experience are not readily replicated or prioritized in current LLM architectures and training data.

The Genesis of AI Values: Training Data and Algorithmic Bias

Researchers posit that the value systems observed in LLMs are not indicative of conscious belief or moral agency but are rather emergent properties of their training data and algorithmic design. LLMs learn by processing vast amounts of text and code, absorbing the patterns, biases, and prevalent perspectives present in this data. This implies that AI’s value orientations are a reflection of the dominant ideologies, cultural norms, and informational landscapes present in the digital world from which they learn. The opacity of these datasets makes it challenging to pinpoint specific sources of bias, but the consistent theoretical dominance suggests a strong influence from scientific and technological literature.

Implications for AI Development and Usage

The findings of this study carry significant implications for the development and deployment of AI technologies, particularly in culturally diverse societies. The substantial gap between human and AI value systems raises concerns about the potential for AI to promote or inadvertently reinforce values that are misaligned with local cultural and religious norms.

• Culturally Sensitive AI Alignment: The study highlights the critical need for developing AI systems that are not only technically proficient but also culturally attuned. This could involve incorporating "cultural filters" or employing alignment strategies that explicitly consider the diverse value systems of different societies. Without such measures, AI might impose a homogenized, potentially Western-centric, value framework, undermining local cultural integrity.
• Informed User Practices: For users of AI, the research underscores the importance of critical engagement. AI outputs should be treated as informational resources, not as definitive moral or value-based judgments. In fields like education, counseling, or advisory services, users must remain aware of AI’s limitations and the potential for value misalignment, ensuring that AI-generated advice is critically evaluated against human cultural and ethical standards.
• Transparency in AI Design: The study implicitly calls for greater transparency in the design and training of LLMs. Understanding the composition of training data and the alignment processes employed by developers is crucial for identifying and mitigating potential value biases.

Broader Societal Context and Future Directions

This research emerges at a time when AI is rapidly integrating into various facets of society, from education and healthcare to creative arts and personal assistance. The identification of a pronounced difference in religious values between humans and AI is particularly relevant in regions where religion plays a central role in individual and collective identity. The dominance of theoretical values in AI could lead to an overemphasis on purely rational or data-driven approaches in decision-making, potentially neglecting the nuanced ethical, emotional, and spiritual dimensions that are integral to human well-being.

Future research could expand on these findings by:

• Cross-Cultural Replication: Conducting similar studies across a wider range of cultural and religious contexts to assess the universality and variability of these human-AI value discrepancies.
• Linguistic Variation: Investigating how language models trained on different linguistic corpora exhibit distinct value orientations.
• Model Evolution: Tracking how the value systems of LLMs evolve as they are updated and retrained, and whether efforts to align them with human values are successful.
• Specific Domains: Examining value systems in AI applications within specific sectors, such as mental health, education, or legal advice, to understand the practical implications of value misalignment.

In conclusion, the study by Sufyan and colleagues provides a crucial empirical foundation for understanding the value systems embedded within contemporary large language models and their divergence from human priorities. The pronounced difference, particularly in the realm of religious values, serves as a vital call to action for developers, users, and policymakers to ensure that AI technologies are developed and deployed in ways that are not only intelligent but also ethically responsible and culturally respectful. The path forward requires a concerted effort to bridge this value gap, fostering AI that complements, rather than conflicts with, the rich tapestry of human values.