By Gianna Melillo
Eric Topol, MD, MACP, struck an optimistic tone on the future of artificial intelligence (AI) in medicine during the Opening Ceremony of Internal Medicine Meeting 2024 on Thursday.
He kicked off the talk by quoting Geoffrey Hinton, a computer scientist regarded by many as the godfather of AI: “I always pivot to medicine as an example of all the good [AI] can do, because almost everything it's going to do there is going to be good.”
Dr. Topol laid out current and future benefits of the technology, from reducing diagnostic errors to cutting down on administrative tasks, while acknowledging its risks. He is founder and director of the Scripps Research Translational Institute and professor of molecular medicine and executive vice president of Scripps Research.
Early on, AI was primarily used for radiological image interpretation, with the most applications in mammography and gastrointestinal diagnoses. Not only can AI help detect cancer in this context, but it can also make the process much more streamlined, Dr. Topol explained, as the technology is able to review and interpret thousands of images in a short amount of time.
This marked a big step forward for diagnostic accuracy, which was largely expected, Dr. Topol noted. What was unexpected was the technology’s ability to determine a slew of additional data from a single image. Citing research that found AI could discern the sex of a patient based solely on a retinal image, Dr. Topol predicted, “Someday we’ll be taking a selfie of our retina to do a checkup.”
He also pointed to a recent study published by Annals of Internal Medicine that found a deep-learning model based on a routine chest radiograph was able to predict 10-year risk for major adverse cardiovascular events beyond the clinical standard risk score.
The next step beyond these applications is multimodal AI, an area that’s seen exponential growth in recent years, Dr. Topol explained. This next generation entails AI supervising itself. The most well-known application is the language processing tool GPT-4.
Dr. Topol highlighted the tool’s potential by sharing the case of a boy whose parents took him to 17 doctors over three years trying to determine the root cause of his symptoms, which included a gait problem and severe pain. After all failed, his mother entered the boy’s symptoms into ChatGPT, which ultimately delivered a diagnosis. The mother was able to take the boy to a neurosurgeon to receive treatment, and he is now doing well.
Other new research suggests multimodal AI may even have better empathy and communication skills than physicians in some cases.
“I never would have envisioned that you would have AI help you be more empathic and a better communicator,” Dr. Topol said in a discussion with ACP Executive Vice President and Chief Executive Officer Darilyn V. Moyer, MD, MACP, following his talk. AI could potentially coach physicians in these areas. “Even if you think you’re a really good communicator, you can always be better,” he said.
He noted that AI still has its shortcomings. For one, its clinical reasoning is not up to par with physicians’, Dr. Topol noted, citing a case where the tool yielded a diagnosis of an ectopic pregnancy in a 77-year-old woman.
This flawed clinical reasoning underscores one serious problem of AI in medicine: the lack of real-world proof of its safety and efficacy. “We don’t have the data, the evidence, to use these things at this point in time,” Dr. Topol said, dispelling the notion that AI has already been heavily integrated into the health care system. “The idea that there’s been a lot of implementation is crazy. There’s very little to date, and we’re only starting to see some on the administrative side.”
The tools are also only as good as the data they’re based on, meaning they could propagate cultural biases and worsen health inequities if left uncorrected.
But Dr. Topol is hopeful that AI's eventual integration into everyday clinical practice will help restore the doctor-patient relationship that he says has been eroded by administrative burdens and paperwork requirements.
“Today, medicine is fast and shallow,” Dr. Topol said. “Our ability to execute our role is impaired.”
As AI helps physicians cut down on administrative time, it can free up space for reflective thinking and reduce burnout, he predicted. Going forward, it’ll be crucial that physicians are able to use AI not to see more patients, but to increase the time with patients they do see.
“I don’t know any other alternative to get us out of the situation we’re in right now, outside of leaning on this technology,” Dr. Topol said. ■