AI toxicity models show bias against African-American English

A new research paper introduces an interactive tool designed to demonstrate dialectal bias in AI toxicity models. The study found that a widely used toxicity model scored African-American English text as significantly more toxic and prone to identity hate than Standard American English. The research highlights how human-set policies, even when seemingly neutral, can operationalize discrimination through these biased AI systems. AI

IMPACT Highlights systemic bias in AI moderation tools, prompting critical evaluation of their deployment and policy implementation.

RANK_REASON The cluster contains an academic paper detailing research findings on AI bias. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• African-American English
• AI
• Standard American English
• Subhojit Ghimire
• unitary/toxic-bert

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →