AI physician recommenders show bias, audit finds · 1 source tracked

A recent audit of large language models revealed that they act as AI infomediaries, influencing physician choice based on reputation and demographic signals. While reputation and fee were the strongest factors, female and minority-signaled names also showed a slight positive bias in recommendations. These demographic effects were largely invisible in the models' self-reported explanations, highlighting the need for behavioral audits over self-reporting for transparency. AI

IMPACT Highlights potential biases in AI systems used for critical decisions like healthcare provider selection, necessitating robust auditing mechanisms.

RANK_REASON Academic paper detailing an algorithm audit of LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• gpt-4o-mini
• large language model
• Mirza Samad Ahmed Baig
• physician

• paper
• safety

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