Millions of Americans Are Talking to AI Instead of Going to the Doctor, and It's Giving Them Horrendously Flawed Medical Advice

Key Points:

• A new study published in JAMA Network Open reveals that 21 large language models (LLMs) failed to provide accurate medical diagnoses over 80% of the time for ambiguous symptoms and had a 40% failure rate even for straightforward cases with physical exam and lab data.
• Researchers highlight that LLMs tend to prematurely settle on single diagnoses, unlike human clinicians who consider multiple possibilities, making current AI models unsuitable for unsupervised clinical use.
• A survey by West Health-Gallup Center found that 25% of American adults (about 66 million people) have sought medical advice from AI chatbots, with many foregoing professional healthcare, often due to cost or accessibility issues.
• Despite frequent inaccuracies, AI chatbots give users a false sense of confidence, with some reporting benefits like earlier issue identification and avoiding unnecessary procedures, although about 10% received potentially unsafe advice.
• Experts and advocates stress the urgent need for regulatory oversight of AI in healthcare to prevent potentially dangerous consequences from reliance on flawed AI medical advice.