AI's fluency in other languages hides a Western worldview that can mislead users
5 hours ago
- #Epistemological Persistence
- #AI Cultural Bias
- #Multilingual AI
- AI models like ChatGPT respond fluently in many languages but often reflect Western, particularly American, cultural values, not local ones, a pattern termed 'epistemological persistence'.
- Training data for major AI systems is predominantly from English-language sources, leading them to reason primarily in English and translate final outputs, embedding a Western worldview.
- Experiments show that even when asked in Indonesian, AI responses about concepts like education or 'malu' (a social-relational concept) default to individualistic frameworks, ignoring local cultural nuances.
- Chinese models offer an alternative cultural lens, but other regional efforts often build on U.S. foundations, keeping core Western logic intact.
- The use of AI for emotional support and advice can make these embedded cultural perspectives harder to detect and challenge, potentially normalizing certain worldviews globally.