Which Humans? Mapping Cultural Clusters in a Data-Driven World

What does it really mean to say that AI reflects “human” values?

A 2023 study tackled this question by mapping cultural similarities across countries using a statistical method called multidimensional scaling. In simple terms, this technique takes thousands of survey responses and plots countries on a two-dimensional map based on how similar their values are. The closer two countries appear on the chart, the more alike their responses.

The data came from the World Values Survey (WVS), covering topics like happiness, trust, economic beliefs, political attitudes, and postmaterialist values. When researchers grouped countries into clusters using advanced statistical testing (including 5,000 Monte Carlo simulations to ensure reliability), distinct cultural groupings emerged.

One of the key findings? GPT’s responses tend to cluster closest to what researchers call “WEIRD” populations or Western, Educated, Industrialized, Rich, and Democratic societies.

To test this more rigorously, researchers compared GPT’s answers to average human responses in 65 countries. They then measured how similar GPT was to each national population and compared that similarity to each country’s “cultural distance” from the United States.

The result was striking.

There was a strong negative correlation (r = –0.70), meaning the farther a country is culturally from the U.S., the less GPT’s responses resemble people from that country. In other words, GPT aligns most closely with populations culturally similar to the United States.

The pattern held up under multiple robustness checks. It remained significant using non-parametric tests and multilevel modeling. Even more telling, GPT-human similarity was strongly correlated with measures of development: Human Development Index (0.85), GDP per capita (0.85), and internet penetration (0.69).

What does this mean in practical terms?

It suggests that GPT’s “default psychology” mirrors societies that are wealthier, more connected, and more institutionally developed. While this may reflect the distribution of training data, it also raises important questions about representation and cultural balance in global AI systems.

As AI becomes embedded in decision-making, education, and governance, understanding which humans it most closely resembles isn’t just academic. Instead, it’s foundational to how inclusive and globally representative these systems truly are.