| For Chinese who do almost everything on their smartphones, seeing a doctor online doesn't sound like a novel idea. But the way technology is headed in the country, they'll soon wonder if they're talking to a real doctor — or one that's powered by so-called generative AI tools similar to ChatGPT. Chinese online healthcare provider Medlinker in May unveiled an AI doctor dubbed MedGPT, claiming it's capable of diagnosing some of the most common diseases with the same degree of accuracy and consistency of a human doctor. In fact, AI has already been widely applied in online consultations offered by some of China's leading Internet healthcare services — but it largely serves an auxiliary role. Patients wanting to chat with a doctor online sometimes find themselves talking to chatbots first, which collect basic information from patients and redirect them to the suitable human specialist. What sets MedGPT apart, according to Medlinker, is that it could handle the entire process on its own, from diagnosis to prescribing tests and medication — replacing human doctors. Medlinker crammed tens of billions of medical records and academic journals into MedGPT, with the chatbot powered by ChatGPT-style services developed by domestic and foreign firms. A team of over 100 human doctors then trained the chatbot to assess a patient and take appropriate actions, such as ordering medical tests, prescribing medicines or offering dietary guidance. MedGPT is a very inquisitive doctor: The chatbot engages patients in multiple rounds of questioning to get as much information as possible before arriving at a conclusion. But how does it stack up against human counterparts?  Attendants photograph the stage as Sam Altman, chief executive officer of OpenAI, speaks during an event in Seoul last month. Photographer: SeongJoon Cho/Bloomberg A month after MedGPT's unveiling, Medlinker ran a trial pitting it against 10 senior human specialists, consulting more than 100 patients with issues from cardiovascular problems to kidney disease. MedGPT and the physicians got similar scores, and a judging panel of seven doctors credited the chatbot for being comprehensive in its questioning and refraining from giving diagnoses too early. Still, they also noted shortcomings: Some of the tests MedGPT ordered were repetitive, and recommended treatments could be excessive. "MedGPT still has lots of problems, but I think the strides it has made is a milestone," said Ren Jingyi, a cardiologist at a top Beijing hospital, one of the seven experts — and only one of two to rate the bot higher than its human counterparts. MedGPT isn't yet in commercial use, pending government approval — and guidelines for the real-world application of such generative AI tools for healthcare. The company said it is looking for partnership with companies and medical institutions to improve the AI doctor's accuracy, train it to diagnose more diseases and run a bigger trial later this year. It also hopes to work with medical experts to establish standards for AI-based healthcare services. It could be a matter of when, rather than if, AI doctors will be able to independently treat humans. But whether they'll replace some human doctors or be support tools remains to be seen. | 
No comments:
Post a Comment