Value Systems of AI and University Students: LLM vs. Human Comparison

A 2026 study comparing value systems between three large language models and 214 university students. The dataset, authored by Nabil Saleh Sufyan, contains results from the Study of Values assessment, measuring six value types: religious, social, theoretical, economic, political, and aesthetic. It includes test-retest reliability data for the LLMs and demographic effects (gender, academic level) for the student sample.

Use Cases

• Analyzing value system differences between AI models and human groups based on Spranger's six value types.
• Investigating demographic effects on value priorities based on gender and academic level variables mentioned.
• Benchmarking AI model outputs for value alignment research based on repeated test administrations.
• Studying the representation of cultural and moral frameworks in AI training data based on the described value rankings.

Strengths

• Includes data from 214 students and three LLMs (OpenAI-o1, Gemini-2.0, DeepSeek-V3).
• Reports test-retest reliability for the LLM assessments.
• Provides effect-size estimates for human-AI discrepancies (e.g., d=2.21 for religious values).

Limitations

• Dataset is very small (63.8 KB), indicating limited scope.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

figshare, author Nabil Saleh Sufyan.

Collection Method

Descriptive-comparative design using the Study of Values assessment.

Freshness

Last updated 2026-04-21 05:36:08; freshness should be verified.

Geography

Student sample from King Khalid University.