Last year, nearly 100 people tried a new form of medical care at Beth Israel Deaconess Medical Center in Boston. After scheduling a visit with the hospital’s urgent care facility, they discussed their ailments with a Google chatbot before meeting with their doctors.
The bot explored their symptoms, the possible causes of their maladies and the potential treatments, before generating a summary of each conversation. Doctors could then review the bot’s summary in preparation for a patient’s visit.
These patients were part of an eight-month clinical trial that explored whether this specialized chatbot could assist primary care physicians without providing misleading or otherwise harmful information.
In the first-of-its-kind study, published on Thursday in the medical journal The Lancet, the hospital’s primary care physicians said that the bot was helpful in about 75 percent of patient cases, and that the technology may have changed the way they handled more than half of those cases. “They compared it to a high-performing medical student or resident,” said Peter Brodeur, a resident physician at Beth Israel.
The study also found that the bot had “hallucinated” — made stuff up — only once during hundreds of conversations. Generally, patients said they trusted the bot, but only because it was operating alongside trained physicians.
“None of our patients trusted it until they realized that it was handing-off their doctor,” said Adam Rodman, a Beth Israel internist and medical historian who helped conduct the trial.
The study is another indication that artificial intelligence technologies are fundamentally changing health care. For more than a decade, Silicon Valley companies, academic labs and medical facilities have worked to build A.I. systems that can automatically diagnose common conditions by analyzing lab tests and medical records, identify signs of illness and disease in medical scans and accelerate the discovery of new medicines and vaccines.
But experts warn that these systems cannot yet replace the care of a trained doctor.
Researchers at Google have spent years developing the company’s specialized medical chatbot, called AMIE, short for Articulate Medical Intelligence Explorer. The bot is powered by Google’s flagship A.I. technology, Gemini, which also drives parts of its internet search engine.
A study by The New York Times has shown that the A.I.-generated answers provided by Google’s search engine were inaccurate about 10 percent of the time. But Google has trained AMIE specifically for “clinical reasoning” and medical diagnosis. Dr. Rodman says that the bot uses a more powerful version of Gemini than its search engine’s.
“Our model is reasoning from a relatively short conversation,” he explained. “This is a situation where hallucination is less likely.” A system that has to process a patient’s entire medical record, he added, is likely to hallucinate more often.
The single time the bot hallucinated during the trial, it misrepresented a date, discussing the date as if it were in the future, when it was in the past.
During the trial, trained physicians monitored the bot as it chatted with patients, ready to stop the conversation if it caused them emotional distress or otherwise led them into territory deemed to be unsafe. These physicians did not halt any of the hundreds of conversations conducted as part of the trial, but they did provide patients with additional context on five occasions.
In one instance, the bot listed cancer among the potential diagnoses, at which point a physician stepped in to reassure the patient that the bot was simply considering all the possibilities.
The clinical trial at Beth Israel was designed primarily to test the safety of the chatbot. Google has since begun a nationwide trial that aims to compare patient visits with and without chatbots.
“This is the technology of the future,” said Robert Califf, a cardiologist at Duke and former Food and Drug Administration commissioner who was not involved in the clinical trial. “We should not take it for granted that this will reduce the load on the doctors, but it could also unearth problems that never would have taken the doctor’s time.”
The post Clinical Trial Indicates A.I. Chatbots Could Aid Urgent Care appeared first on New York Times.
