The most recent upgrade of ChatGPT, known as GPT-4, has the potential to get a flawless score on medical licence tests. If it gets a medical question is incorrect, there may be a reasonable point of disagreement. It excels at activities previously assumed to need human sensitivity, such as selecting the correct phrases to use when breaking terrible news to patients.
The ability to analyse images is also improving in these systems. A human doctor still needs to feel a lump or evaluate a damaged ligament, but AI might interpret imaging tests like an MRI or CT scan and provide a prognosis. In a perfect world, AI wouldn’t have to replace human doctors but rather improve their job. But we still have a long way to go until we know when and when it’s appropriate to use AI’s suggestions.
The general public will undoubtedly use it for medical advice, just as many of us have relied on “Dr Google” for years. Our relatively low life expectancy may improve or worsen with GPT-4, but public health professionals blamed a plethora of disinformation for it this week.
A Harvard professor of biomedical informatics named Andrew Beam told me he could get GPT-4 to provide him dramatically different results by slightly altering the wording of his questions. He is astounded by the system as much as anybody else. For instance, you can’t assume it to get perfect scores on medical examinations just because you order it to pretend to be the smartest person alive.
He claimed the feature was only an autocomplete function, guessing at the next best word. And yet it resembles thought quite closely.
“The interesting thing, and the thing I believe few people foresaw, is that a lot of jobs that we assume need broad intelligence are autocompleted activities in disguise,” he added. Several kinds of medical logic are included in this category.
