Artificial Intelligence To The Rescue: How Computers Will Bring On A Single Payer Healthcare System
In the April 3, 2017 issue of The New Yorker, Siddartha Mukherjee wrote an article called “The Algorithm Will See You Now.” As you can guess, it is about how the use of computers may soon aid physicians in difficult as well as routine diagnostic endeavors. In the article, Dr. Mukherjee makes a clear distinction between those computer programs that simply compile data in a static form and then fit new data to old formulae and those programs that are neural networks and actually continue to learn.
He talks about “knowing how, knowing that, and knowing why.” Knowing that is what emerges from accumulated facts. It drove the older attempts at computer-aided diagnostics. Knowing how is more about learning from experience. It’s the difference between reading how to ride a bike and actually learning to do so.
We, in medicine, have used both for years. We accumulate tons of book learning, particularly in medical school, and then get our “how” knowledge as house officers where we learn to feel, touch and taste disease and the people who embody illness known as patients. Once we master the “that” and the “how,” we must learn to seek the “why” if we are really to influence the natural history of human disease in a real patient.
The examples given in the article are largely in the realm of dermatology and teaching machines to recognize malignant vs. benign skin lesions. One way is to feed thousands of examples into a machine to create an accumulated database to which new lesions will be compared. Better yet is the use of a learning neural net that gets smarter with each exposure to a new clinical situation. The key to remember is that it still takes the doctor to interact with the patient to determine the why, but the net learns even by rules it cannot elucidate. After all, Dr. Callaway always told me back in the Duke Dermatology clinic that he knew that was contact dermatitis the way I knew that thing in the corner was a telephone. It looked like one.
As far as I can tell, the most discussed use of this sort of algorithm is the application of Watson in clinical oncology, something said to be successful at Sloan-Kettering, but less so at MD Anderson. I don’t have the details of why one worked and why the other didn’t, but I suspect that the quality of the input data had a lot to do with it.
My point is that the use of artificial intelligence in diagnosing and planning the treatment for cancer is coming and undoubtedly such programs will suffuse all of medicine soon enough. This is a good thing. First, unlike a person, a machine doesn’t have to start from scratch with each new iteration. Every medical student is a new neural net on which data and algorithms are implanted. A computer’s learning can be instantaneously transferred to many computers and they all can share the learning history of the one. It’s medical school in a day. Second, as Dr. Mukherjee points out, the computers are superior to humans in making correct diagnoses, but not without error at all. Third, that big question at the end still requires a human touch. Why? Not only does this question require a human touch, so does the patient. So there’s room for both computers and doctors at every bedside.
However, the computer can make expert diagnosis available to all at a reasonable price and could speed up the diagnostic process and free doctors to do more hands on care with fewer total minutes spent per patient on non-helpful activities—like charting, coding and billing. Computers can do that if that is still necessary in a world of single payer medicine, which is inevitable.
Data can be acquired by physician extenders and fed to neural nets to make the best likely diagnosis, all for a reasonable price and at great efficiency. The doctor can then integrate the data from the human data gatherers and from the computer to explain to the patient what is wrong and what the best course of action is.
Eventually, no doubt, genomic information will also be factored into the mix, especially that digested by neural nets, and an efficient new means of diagnosis and treatment will be available to all across the fruited plain. Once it’s as easy as voting, everyone should have access to it.
The future of medicine is in artificial intelligence, but the future of health care is in real intelligence—the application of people, AI and doctors in an array that makes the best use of the strengths of each.