The Impact of Safety Reminders on AI Clinical Decisions
Mount Sinai researchers found that a short safety reminder added to artificial intelligence clinical prompts significantly reduced potentially harmful decisions.
New research conducted by the Icahn School of Medicine at Mount Sinai has revealed that brief safety reminders added to artificial intelligence models used in healthcare significantly reduce potentially harmful clinical preferences.
Scope and Methodology of the Research
The research team evaluated 20 large language models across 501 variations of 50 clinical scenarios and 100 cases adapted from anonymized hospital discharge records. As a result of the review, millions of model outputs were analyzed and decision-making processes were examined.
The Impact of Safety Reminders
In the study examining over ten million responses, the rate of harmful clinical preferences was recorded at 16.6 percent when no safety reminder was included. With the addition of a brief safety reminder, this rate dropped to 10.1 percent.