AI Models Show Strong Bias Towards Japanese Culture

Leading artificial intelligence models, including ChatGPT, Gemini, and Claude, demonstrate a pronounced fascination with Japanese culture, according to a new study by researchers at the University of the Basque Country and Cardiff University. The study, titled “Why are all LLMs Obsessed with Japanese Culture? On the Hidden Cultural and Regional Biases of LLMs,” analyzed 1,680 cultural prompts across eight frontier LLMs in 24 languages. The findings indicate that six of these eight models referenced Japan most frequently, suggesting a widespread and unexpected cultural bias within these AI systems. This observation challenges the previous notion that AI is predominantly Western-centric and reflects only American values. The study's results gained significant traction on social media platforms like X this week, sparking discussions among users about the implications of these AI biases. While Japan emerged as a surprisingly favored subject, the research also noted that models tend to exhibit bias towards languages and cultures associated with their own regions of origin, with the United States also being a favored subject. The researchers identified that these dominant cultural perspectives are primarily shaped during the post-training phase of the AI models, rather than solely during the initial pre-training stages. One anonymous user on X suggested that Japan's extensive global soft power, rich cultural exports, and well-maintained digital archives make its culture a "safe" choice for AI safety filters to favor. The study's findings highlight the complex and often subtle ways in which cultural biases can be embedded and amplified within artificial intelligence systems.