AUSTIN (KXAN) — Cooper Myers is a member of the AI generation. After the 24-year-old moved to Austin to pursue a job in tech, he discovered that not only could artificial intelligence help him at work but it could save his life.
“I’ve been using (AI in) my day to day. Every time I have a question or something doesn’t quite make sense, I asked ChatGPT,” Myers said. This time the thing that didn’t make sense was his health.
In 2020, Cooper was diagnosed with Type-1 diabetes. Around two million Americans have been diagnosed with Type-1 diabetes, which causes the pancreas not to produce insulin. Type-1 diabetes is thought to be caused, according to the Centers for Disease Control, by the immune system attacking the cells that produce insulin.
“When I was first diagnosed, I never had any antibodies. And type one diabetes is an autoimmune disease. So that was suspicious,” Cooper said. His family had a history of diabetes, but things didn’t make sense.
Shortly after moving to Austin, he began to look closer at his diagnosis.
His blood sugar with low, despite being fitted with an insulin pump. “When blood sugar started going down, that’s a big concern. People could be driving, people could be sleeping and have low blood sugars, that can be a safety concern,” said Dr. Keta Pandit, Cooper’s doctor, with Texas Diabetes and Endrocrinology.
Dr. Pandit said Cooper’s symptoms weren’t adding up. Curiosity getting the better of him, Cooper looked for an answer by asking… ChatGPT.
Diagnosis: ChatGPT
Cooper typed this prompt into ChatGPT: “If my father is a skinny type 2, my grandfather is a skinny type 2 and I am a skinny type 1 with no antibodies, then could we have a different kind of diabetes?”
ChatGPT, a generative search engine that uses AI to create answers, came up with MODY (Maturity-Onset Diabetes of the Young). A rare form of diabetes that has a genetic origin, MODY works by causing the pancreas not to realize it needs the release insulin. The pancreas still creates enough insulin, but doesn’t use it.
“MODY is often overlooked simply because of just the common conditions that we see so frequently. And the prevalence is just so high of Type-1 and Type-2,” Dr. Pandit said.
MODY also appears in younger people that Type-2 typically does. It also doesn’t require insulin to treat, just a pill.
The AI proved correct. Cooper was diagnosed with MODY. He now no longer has to monitor his diet as closely, take insulin or have the pump installed. He simply takes the daily medication.
“It basically took me from like one of these people that was like counting every single thing I eat, to just eating whatever I want,” Cooper said.
The risks of using AI with medicine
AI is still a fairly young technology and comes with all the risks associated with new technology. Risto Miikkulainen, a professor and AI expert at the University of Texas, said that this case shows some of the potential of the technology.
“People who are really highly trained, can use it as a second opinion,” Miikkulainen said.
He said that there are some downsides. AI typically uses the most common answers, so finding rare diseases is unlikely. “It’d be difficult to get it to identify something that’s unusual and exceptional, something that the doctors might overlook,” Miikkulainen said.
AI can also combine information that doesn’t make sense called hallucinations. AI, however, can be trained to not do this, but the technology isn’t there yet.
Until then, Miikkulainen and Dr. Pandit recommend using AI like a resource.
“If that information is utilized in conjunction with your healthcare team, and to expedite care, or diagnosis or evaluation, I think that can be very usefu,” Dr. Pandit said.