How much of a pluralist is ChatGPT? A comparative study of value pluralism in generative AI chatbots
Bibliographic Data
| ID | 20396454 |
|---|---|
| Authors | Nurit Novis-Deutsch (0000-0002-8683-3008, University of Haifa, corresponding author), Tal Elyoseph (University of Haifa), Zohar Elyoseph (0000-0002-5717-4074, Imperial College London) |
| Year | 2026 |
| Volume | 41 |
| Issue | 2 |
| Pages | 1561-1575 |
| Publication date | 2026-02-01 |
| Open Access | Yes |
| Type | ARTICLE |
| Venue | AI & Society (JOURNAL) |
| Journal identifiers | ISSN: 0951-5666 • E-ISSN: 1435-5655 |
| Publisher | Springer Science and Business Media LLC (PUBLISHER) |
| DOI | 10.1007/s00146-025-02450-3 |
| OpenAlex | W4412403718 |
| Language | EN |
| References cited | 31 |
This study introduces pluralism as a benchmark for evaluating generative AI chatbots’ ability to navigate and potentially endorse diverse values and perspectives. Pluralism—the recognition and valuing of multiple, distinct perspectives—is essential for contemporary democratic societies, making systematic assessment of how AI systems handle value pluralism crucial. Using the Magic Wand Survey (MWS), which presents diverse value conflicts, we empirically measured 2 dimensions of pluralism—“Both/and Reasoning” (cognitive pluralism) and “Willingness to Preserve Difference” (behavioral pluralism)—across 120 iterations of 4 prominent AI chatbots, comparing them to a human sample ( N = 335). AI chatbots exhibited higher cognitive than behavioral pluralism. Furthermore, while chatbots’ cognitive pluralism significantly exceeded human levels, their behavioral pluralism was comparable overall yet notably lower than humans on some morally charged issues (e.g., capital punishment). Clear and systematic variations also emerged between models: while general patterns held, the degree of pluralism varied substantially, with Gemini 1.5 Pro being the most pluralistic and ChatGPT 4o the least among those tested. These findings underscore the importance of pluralism as a key benchmark in AI evaluation
Art · Epistemology · Generative grammar · Linguistics · Performing arts · Sociology · Visual arts · AI in Service Interactions · Computer Science · Ethics and Social Impacts of AI · Philosophy · Psychology · Psychology of Moral and Emotional Judgment
The Hedgehog and the Fox
Justice as Fairness
The New Synthesis in Moral Psychology
Most people are not Weird
Does the chimpanzee have a theory of mind?
An fMRI Investigation of Emotional Engagement in Moral Judgment
Pluralism as an antidote to epistemic violence in psychological research
The one and the many
The plasticity of ChatGPT’s mentalizing abilities
Pluralism and Liberalism
The Dynamics of Democratic Exclusion
The moral foundations hypothesis does not replicate well in Black samples
Liberals and conservatives rely on different sets of moral foundations
Our Compelling Interests
| Highly cited | No |
|---|