Thousands of doctors turn to ChatGPT in hospitals to respond to their patients

The artificial intelligence (AI) has begun to become part of medical care in hospitals in the United States, where thousands of doctors use tools like MyChart to answer patient questions, according to informa The New York Times. This technology, based on the GPT-4 language model, generates automatic responses that are monitored and corrected by health professionals before being sent. Although this innovation reduces burnout among healthcare workers, it also raises ethical concerns and risks of errors.

Developed by the Epic company, MyChart uses information provided by patients, such as their medical histories and medications, to generate responses that seek to improve efficiency in online queries. MyChart is presented as a solution for hospitals seeking to optimize their time and avoid trips and long waits. According to the company, the tool serves more than 190 million patients in the United States at health centers like UW Health in Wisconsin.

Hospitals using MyChart have opted to different strategies regarding communication with patients. While some, such as UC San Diego Health, include a notice at the end of each message indicating that it was “automatically generated” and reviewed by a doctor, other centers, such as Stanford Health Care and NYU Langone Health, have decided not to inform patients of AI intervention. The latter consider that revealing this information could generate mistrust or feeling of betrayal.

The system has been well received by patients, according to some specialists. There are those who have even expressed admiration for the use of this technology. However, doctors have also noted the possibility of significant errors in automatically generated responses. In one case, the AI ​​incorrectly claimed that a patient had received a vaccine. These types of failures show that, although technology can be useful, it is essential that doctors carefully review responses before submitting them.

The implementation of AI in medical consultations has also been supported by scientific studies indicating that it reduces the stress and workload of healthcare personnel. However, the time savings are not as significant as expected, since doctors must review and edit the messages generated by the AI. According to data from Duke Health, less than a third of draft responses are submitted without editing, reflecting professionals’ caution about potential errors.