AI Model Value Surveying

AI model value surveying is the method in A hawk who flew on political winds: Lindsey Graham of asking AI models questions normally asked of humans, then comparing the answers with country, voter, or cultural clusters. Sondre Solstad and The Economist team use the World Values Survey, an American electorate survey, and contentious current questions to make model values measurable rather than merely impressionistic.

The source’s main finding is that many frontier models cluster near rich-country, secular, self-expression-oriented societies and lean left by American survey measures. That does not prove models have beliefs, but it does challenge easy claims that a model is neutral, placeless, or value-free.

Key Claims

• Human value surveys can make model-output tendencies easier to compare across models and against human populations.
• Survey results should be interpreted as model behavior under a test, not as evidence of consciousness or stable inner belief.
• Model values can reflect training data, post-training preferences, company policy, and the social environment of model builders.
• The method becomes more important as AI systems are used for advice, judgment support, and emotionally sensitive questions.

Connections

• World Values Survey and Sondre Solstad - survey framework and source participant.
• AI Model Censorship and Language-Dependent AI Bias - model-behavior problems surfaced by the tests.
• Taki AI - historical-corpus example.
• AI Advice Moral Outsourcing and Human Judgment Under AI - downstream risk when surveyed defaults influence advice.
• AI Governance And Compliance - broader governance context.