Tradeoffs Should I Trust AI to Diagnose Me?
Oct 30, 2025
Join physician and New Yorker writer Dhruv Khullar as he navigates the promising yet perilous world of AI in healthcare. He explores how AI can expedite diagnostics but warns of potential dangers, such as cognitive de-skilling and reliance on faulty chatbot advice. Khullar emphasizes using AI as a tool for guidance rather than a definitive source, while highlighting its ability to enhance communication for non-English speakers. He discusses the balance of human judgment with technology's capabilities, urging caution and accountability in AI integration.
AI Snips
Chapters
Transcript
Episode notes
Fluency Makes AI Mistakes More Dangerous
- AI chatbots are far more fluent and persuasive than search engines, making their mistakes more dangerous.
- Fluent personalization increases risk when models suggest harmful or toxic recommendations.
Harm From A Toxic AI Suggestion
- A 60-year-old man asked ChatGPT for salt substitutes and was advised to use bromide, which can be toxic.
- He developed paranoia and hallucinations and required hospitalization.
Use AI For Wayfinding, Not Final Answers
- Use AI as a guide for wayfinding through diagnoses rather than a final authority.
- Ask it to suggest next tests, cite studies, or help prepare for clinical conversations with your doctor.