Researchers at MIT's Center for Constructive Communication evaluated GPT-4, Claude 3 Opus, and Llama 3 using datasets measuring truthfulness and factual accuracy with simulated user biographies varying education, English proficiency, and nationality. The study found systematic lower performance for users with lower formal education, limited English skills, or outside U.S. origins, including more frequent refusal to answer and instances of condescending language.
These findings caution against assuming AI chatbots democratize information access universally. The authors emphasize the necessity of mitigating biases and harmful tendencies to ensure equitable and safe AI assistance for all demographics, highlighting challenges in deploying LLMs fairly despite their potential.
Source & further reading
Study: AI chatbots provide less-accurate information to vulnerable users ↗Source published 2026-02-19. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.