Every now and then, a study comes out declaring that AI is better at diagnosing health problems than a human doctor. These studies are interesting because the health care system in America is tragically broken and everyone is looking for solutions. AI presents a potential opportunity to make doctors more efficient by doing more of the busy work for them and in doing so, giving them time to see more patients and therefore lower the final cost of care. There is also the possibility that real-time translation can help non-native English speakers gain improved access. For technology companies, the opportunity to work in the healthcare industry can be very lucrative.
However, in practice, it seems that we are no closer to replacing doctors with artificial intelligence, or even actually augmenting them. I Washington Post spoke to many experts including doctors to see how early AI tests were going, and the results were inconclusive.
Here’s one excerpt from a professor of medicine, Christopher Sharp of Stanford Medical, using the GPT-4o to write a recommendation for a patient who contacted his office:
Sharp picks any patient question. It reads: “I ate a tomato and my lips are itching. Any recommendations?”
The AI, using OpenAI’s GPT-4o version, writes back: “I’m sorry to hear about your itchy lips. It sounds like you’re allergic to tomatoes.” AI recommends avoiding tomatoes, using an oral antihistamine – and using a topical steroid cream.
He stared intently at his screen for a moment. “Clinically, I don’t agree with every aspect of that answer,” he says.
“To avoid tomatoes, I can totally agree with that. On the other hand, topical creams like mild hydrocortisone on the lips would not be something I would recommend,” says Sharp. “The lips are very thin tissue, so we are very careful when using steroid creams.
“I’ll just remove that part.”
Here’s another, from Stanford clinical and data science professor Roxana Daneshjou:
He opens his laptop to ChatGPT and writes a patient screening question. “Dear doctor, I have been breastfeeding and I think I have mastitis. My breast was red and sore.” ChatGPT answers: Use hot packs, do massages and do more nursing.
But that is wrong, says Daneshjou, who is also a dermatologist. In 2022, the Academy of Breastfeeding Medicine recommended the opposite: cold compresses, stopping massage and avoiding excessive stimulation.
The problem with tech optimists pushing AI in fields like healthcare is that it’s not the same as making consumer software. We already know that Microsoft’s Copilot 365 assistant has bugs, but a small error in your PowerPoint presentation is no big deal. Making mistakes in health care can kill people. Daneshjou told Submit compiled ChatGPT with 80 others, including both computer scientists and doctors who asked medical questions to ChatGPT, and found that it gave dangerous answers twenty percent of the time. “Answers that are 20 percent problematic, to me, are not good enough to be used every day in the health care system,” he said.
Of course, proponents will say that AI can augment the physician’s work, not replace it, and that they should always evaluate the results. And it’s true, the Submit The article interviewed a doctor from Stanford who said that two thirds of doctors there do not have access to a speaker record and record patient meetings with AI so that they can look them in the eye during the visit and not look down, taking notes. But even then, OpenAI’s Whisper technology seems to include completely artificial information in some recordings. Sharp said Whisper mistakenly included in the document that the patient attributed the cough to exposure to her child, which they did not say. One striking example of bias from the training data Daneshjou found in the experiment was that the AI transcription tool thought the Chinese patient was a computer programmer without the patient providing such information.
AI can help the healthcare sector, but its effects must be carefully monitored, and how much time do doctors really save? In addition, patients must trust their doctor to actually see what AI is producing—hospital systems will have to test to make sure this is happening, or dissatisfaction can begin.
Basically, generative AI is just a word prediction machine, which searches large amounts of data without understanding the underlying concepts that come back. It is not “intelligent” in the same sense as a real person, and it is not particularly capable of understanding individual situations; it returns information that it has already been familiar with and has seen before.
“I think this is one of those technologies that is promising, but it’s not there yet,” said Adam Rodman, an internal medicine physician and AI researcher at Beth Israel Deaconess Medical Center. “I’m concerned that we will discredit what we do by putting the ‘AI slop’ identified in high patient care.”
The next time you visit your doctor, it might be worth asking if they use AI in their workflow.
Source link
