Google Translate and other AI translation systems exhibit systematic gender bias when translating from gender-neutral languages, consistently associating male pronouns with prestigious occupations and female pronouns with traditional roles.
Multiple AI translation systems, particularly Google Translate, exhibit systematic gender bias when translating from gender-neutral languages like Turkish, Finnish, and Chinese into English. The systems consistently translate gender-neutral pronouns into gendered English pronouns based on occupational stereotypes - for example, translating 'o bir muhendis' (Turkish for 'they are an engineer') as 'he is an engineer' while 'o bir hemsire' ('they are a nurse') becomes 'she is a nurse'. This bias stems from word embedding algorithms that learn from large corpora of human-written text containing historical gender biases. Research published in Science journal demonstrated that these AI systems replicate human implicit biases, associating male names with career-related words and female names with family-related terms. The issue affects millions of users across Google's translation services, which support 103 languages and are integrated into Chrome browsers used by approximately 250 million Europeans. Studies found that 182 out of 440 translation pairs contained false gender assignments, with 68 of these marked as 'verified' by Google. Google has begun implementing fixes by showing both masculine and feminine translations for certain language pairs, but the underlying bias in word embeddings remains a widespread problem across the AI industry.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Unequal treatment of individuals or groups by AI, often based on race, gender, or other sensitive characteristics, resulting in unfair outcomes and unfair representation of those groups.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed