A few years ago, at a holiday party, I ran into the husband of one of my wife’s friends, and in the course of conversation, asked him how his job was going. “Great,” he said, “I’m training my replacement.” Turns out, the tech company he worked for was looking to outsource his job to another country, where labor costs were lower, and was having him show the ins and outs of his position to someone in the outsource facility. Merry Christmas!
In February of this year, a group of researchers reported that, for the first time ever, ChatGPT, a “language-based” artificial intelligence, performed at or near the passing threshold (60-percent) on the United States Medical Licensing Exam.1 It did this without any human assistance or a connection to the internet. The authors called this a “surprising and impressive result,” especially in light of the fact that, only months earlier, the best it could muster was 36.7 percent. Hearing this, I couldn’t help but think of my friend training his replacement.
I think artificial intelligence is a wonderful tool that’s already being used as a useful adjunct in spot duty, catching abnormalities on screening images to help physicians treat more people at a more efficient pace, for example. But that’s a very specific use, like a wrench. An AI that can analyze a variety of complex patient presentations from various specialties, develop a diagnosis and formulate the proper treatment plan is a different story entirely. That’s not just a wrench—it’s the whole toolbox. It’s what physicians do.
Now, I’m not paranoid about artificial intelligence replacing physicians, but I think it’s worth going into the future of AI with our eyes open to all the possibilities—both positive and negative. Why? Because a health-insurance company never met a cost-cutting measure it didn’t like. What if, at some point in the not-so-distant future, the insurance company discovered that a chatbot could perform clinical tasks as well as, or better than, a human physician? As a bean-counter bonus, the chatbot never sleeps, and doesn’t take lunches, vacations or sick days.
This probably won't be happening any time soon, though, because it appears that ChatGPT costs about $100,000 per day to run2. Also, when I queried ChatGPT about AIs replacing physicians, it acknowledged that, “Human doctors have important qualities such as empathy, creativity, and the ability to understand complex social and cultural factors that may affect patient health. These qualities are not easily replicated by machines and are an essential part of the health-care system.” It lacks the human touch.
I don’t know the answers to all these questions—and maybe it’ll all be OK in the end—but, as the researchers in the ChatGPT study wrote: The AIs are evolving at an ever-rapid rate. Maybe our ruminations about them should accelerate too.
— Walter Bethke
Editor in Chief
1. Kung TH, Cheatham M, Medenilla A, et al. Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. PLOS Digital Health. February 9, 2023 (online article).
2. Goodwin R. ChatGPT is VERY Expensive To Run – Here’s Why… KYM. January 24, 2023 (online article). https://www.knowyourmobile.com/news/chatgpt-is-very-expensive-to-run-heres-why/.
