The ChatGPT artificial intelligence system, developed by a California-based company, was able to respond with a score equal to or greater than 60% to the license exam to practice medicine in the United States, reveals a study published this Thursday.
The authors of the exercise, Published in PLOS Digital Health magazineexplains that the United States Medical Licensing Examination (USMLE) is a “notoriously difficult and specialized” test and that the fact that an artificial intelligence (AI) was able to pass it “without any human reinforcement” is a “milestone remarkable”.
To obtain these results, researchers at California-based AnsibleHealth put ChatGPT through a three-part exam that is typically given to medical students and physicians-in-training in the United States.
The exercise focuses on topics related to various medical topics, from basic sciences to biochemistry, diagnostic reasoning and bioethics, and the ChatGPT was subjected to 350 of the 376 questions that were made public since the June exam. of 2022.
Without any prior specialized training, the AI system not only answered image-based questions, which were removed, but also faced open-ended or multiple-choice questions.
The answers were evaluated by two evaluating physicians, who did not contact each other, and assigned a final score to the ChatGPT between 52.4% and 75% in the three parts of the exam.
The first focuses on basic sciences and pharmacology and is presented to students with fewer than 400 hours of study, the second is an assessment tool for fourth-year students and focuses on clinical reasoning, medical management, and bioethics, and the third and The latter serves to evaluate physicians with six months to a year of postgraduate training.
ChatGPT has grown in popularity and versatility, but so have its ethical limits for educational purposes in schools and universities.
Earlier this year, Microsoft announced “multi-billion-dollar, multi-year investments” in start up OpenAI artificial intelligence, creator of ChatGPT and other instruments that can write human-readable text and computer code and generate images.
Source: TSF
