The Western Worldview Embedded in Multilingual AI Responses

Recent research highlights a significant concern regarding large language models (LLMs) like ChatGPT, Claude, and Gemini: despite their fluency in multiple languages, they often reflect a Western worldview that may mislead users from non-Western cultures. This phenomenon, termed "epistemological persistence," suggests that the underlying logic of these AI systems is predominantly shaped by American cultural values, even when they communicate in languages such as Indonesian or Arabic.

The Conversation

Experiments conducted by scholars revealed that LLMs translate user prompts into English for processing before generating responses in the user's preferred language. For instance, when asked about Indonesian concepts like "pendidikan" (education) or "malu" (shame), the models consistently framed their responses through an individualistic lens, emphasizing personal development and emotional regulation. This contrasts sharply with Indonesian educational traditions that prioritize ethical discipline and communal values, as documented by scholars Christopher Bjork and Robert Hefner.

The Conversation

The AI's interpretation of "malu" serves as a poignant example of this cultural misalignment. While the term encompasses a shared social awareness and relational dynamics, the models often reduce it to a mere emotional experience, neglecting its communal significance. This misrepresentation can lead to misunderstandings, particularly for users who rely on AI for culturally sensitive advice.

The Conversation

The predominance of English-language training data in AI systems is not merely a technical oversight but a structural issue rooted in the economic realities of knowledge production. As media scholar Safiya Umoja Noble points out, the profit-driven nature of AI development favors models trained on vast English-language datasets. While some alternatives, like Chinese models DeepSeek and Alibaba’s Qwen, offer different cultural perspectives, they too are shaped by their respective cultural contexts.

The Conversation

The implications of these findings are profound. As users increasingly turn to AI for emotional support and guidance, the embedded Western worldview may become normalized, shaping perceptions of family dynamics, education, and social responsibility. This trend raises concerns about the potential for culturally specific values to dominate global narratives, making it difficult for users to recognize and contest these perspectives.

The Conversation

Scholars emphasize the need for greater awareness among users regarding the cultural origins of AI-generated content. They argue that understanding the limitations of these systems is crucial for navigating the complexities of cross-cultural communication.

The Conversation

• “The relational dimensions of the concept disappeared entirely, replaced by the language of individual emotional regulation.” — Scholar of Indonesian Society

The Conversation

• “Certain ways of reasoning about family life, education and responsibility may come to feel natural and self-evident.” — Scholar of Indonesian Society

The Conversation

As the linguistic capabilities of AI systems expand, the challenge remains to address the cultural biases embedded within them. Without a concerted effort to diversify the cultural frameworks that inform AI training, users may unwittingly adopt a narrow worldview that does not reflect their own cultural realities.

The Conversation