Study Reveals ChatGPT Overprescribes Treatments in Emergency Departments
UCSF
|
Contributed by: Kate Gamble
Summary
A study from UC San Francisco highlights concerns regarding the use of ChatGPT in Emergency Departments, finding that the AI often overprescribes unnecessary treatments, including x-rays and antibiotics, and may inappropriately suggest hospital admissions. Comparing ChatGPT's recommendations to those of resident physicians across 1,000 emergency visits, researchers found that ChatGPT-4 was 8% less accurate and ChatGPT-3.5 was 24% less accurate than human doctors. Lead author Chris Williams warns that while AI can assist with certain tasks, it lacks the critical clinical judgment needed in complex emergency situations, raising concerns about the potential harm from overprescribing and the need for improvements in AI frameworks for better evaluation of clinical information.
