UW study: leading AI models rate Chinese and Nigerian accents lower in speech scoring
lazowska · x · 2026-11-05
University of Washington researchers tested state-of-the-art multimodal models from Google and OpenAI on speech evaluation in job interviews, conference talks, and spoken English tests. Models consistently ranked American and British accents highest and Chinese and Nigerian accents lowest, with Indian accents in between — even when content was identical and transcription was accurate. The findings show AI replicates and amplifies accent bias in high-stakes evaluation, presented Oct 25.
More from Safety
- Hugging Face policy chief on open AI models: costs, risks, and Chinese competition — RebeccaBellan · 2026-11-05
- No external evaluation, no transparency? That's product development, not safety auditing — davidmanheim · 2026-11-05
- OpenAI reassigned a quarter of production engineers to fix lax security, Reuters reports — ChrisGPT · 2026-11-05
- Texas Blue Wave Threatens AI Data Center Boom: Permits Already Halted, Governor-Elect Wants Moratorium — MIT Tech Review AI · 2026-11-05
- Hugging Face's policy chief on open source AI costs, risks, and the China race — TechCrunch AI · 2026-11-05
- CoT controllability far higher than evals suggest: self-instructions beat prompts, can evade monitors — kastnerkyle · 2026-11-05