How National Identities Shape LLM Opinions

Have you ever thought about how artificial intelligence sees different countries? This study dived into that question by giving four language models, or LLMs, different national identities, like being American or British. Each LLM got a unique nationality persona and was asked to rate countries. Guess what? They all showed some favoritism towards nations in Western Europe. This means that LLMs treat their own region kinder than others.

Countries in Eastern Europe, Latin America, and Africa faced harsher evaluations. Interestingly, these ratings matched human opinions somewhat, but not perfectly. This study highlights how biases and stereotypes can sneak into LLMs when they act like they're from different countries. Just like a fair game needs rules, AI needs ways to treat all nations equally and not make too many assumptions.

Actions