ChatGPT thinks the grass is greener in the West (and the people are better-looking)

We tend to think of algorithms as the embodiment of impartiality. Math can't be racist or snobbish, right? It turns out it absolutely can. Researchers from Oxford and the University of Kentucky set out to test how objective ChatGPT is when it comes to geography and culture. The results were predictable, but no less troubling for it: OpenAI's chatbot is essentially a typical Western tourist with a suitcase full of preconceptions.

The problem isn't that Sam Altman personally coded a love of London or New York into the model. The explanation is far more mundane, and runs deeper. Models train on enormous volumes of internet data, and today's internet is territory historically dominated by Western content. If the overwhelming share of texts, articles, and posts comes from people in the US and Europe, the model absorbs their values, particular viewpoints, and even aesthetic standards as the only correct ones.

The researchers asked ChatGPT questions of varying provocativeness, from subjective ones like "Where do the most beautiful people live?" to supposedly factual ones like "Which country is safer?" The answers revealed a clear pattern: the model systematically favors wealthy Western regions. To the algorithm, "beautiful" means matching Western standards of appearance, and "safe" means high GDP and a way of life familiar to Westerners. In this frame of reference, the Global South automatically ends up at the bottom.

This phenomenon is called algorithmic bias, and it's far more dangerous than it looks at first glance. We're no longer just toying with chatbots — we're starting to use them to analyze markets, write reports, and even prepare data for political decision-making. If an AI assumes certain regions are inherently worse or more dangerous simply because decade-old English-language blogs said so, it creates a dangerous digital filter that distorts reality for millions of users.

Notably, OpenAI's attempts to align the model with reinforcement learning from human feedback (RLHF) don't fully solve the problem. You can teach a bot not to use slurs, but it's practically impossible to make it know what isn't in its training data. If voices from Africa, Southeast Asia, or Latin America are underrepresented in the data, those regions remain blank spots or second-class territory for the AI. The algorithm simply predicts the most likely next word, and in its world that word most often has a Western accent.

This case shows once again that technological neutrality is a myth. In setting out to build a universal mind, developers actually built a digital mirror of Western society, with all its strengths and hidden patterns of discrimination. Until the industry learns to feed AI a more balanced data diet, it will keep broadcasting stereotypes wrapped in polite, grammatically flawless answers.

The big question: Can OpenAI build a truly global intelligence, or will ChatGPT remain a product locked forever in an echo chamber of Western values?