In the 1990s Garry Kasparov achieved rock star status in the chess world with a freestyle approach that made mincemeat of his opponents. But Kasparov’s reign as world chess master ended in 1997 at the digital hands of IBM’s Deep Blue, making him the first high-profile victim of the modern machine. He would not be the last. In 2011, IBM Watson thrashed human opponents on Jeopardy.
Not long after some people began to wonder if this was the beginning of the end. Stephen Hawking, Elon Musk and Bill Gates have since joined this chorus of fear, warning of a singularity that would give new meaning to the term “machine error.”
But in healthcare at least, AI is humankind’s best chance to meet its potential. In the past it was discovery – not invention – that led to great advances. Vaccines and antibiotics were accidents – Pasteur’s observation that milk maids who contracted cowpox were safe from small pox; Fleming’s discovery that cheese mold in an unwashed petri dish killed bacteria.
In the years ahead, progress will depend on invention, specifically the invention of AI assistants that will be able to process the extraordinary volumes of data that today obscure critical observations.
These inventions, and the opportunities to apply them, surround us. Some took root decades ago in the first applications of AI – rule-based expert systems built from diagnostic algorithms based on human experience. They were part of a grand venture known as SUMEX-AIM (Stanford University Medical Experimental Computer – Artificial Intelligence in Medicine).
MYCIN was the first. Developed in the 1970s, it identified bacteria that caused infections and recommended antibiotics for their treatment.
As a pure research system, MYCIN never made the jump to clinical practice. Nor did the other expert systems in SUMEX-AIM. These early efforts did, however, prove that smart machines could be built and that they could draw medical conclusions.
Today, engineers are working on an AI system that might uncover subtle patterns of disease in medical images, laboratory tests and patient histories. A San Francisco startup called Enlitic has raised $2 million to develop algorithms, imbued with artificial intelligence, that leverage deep learning to find these diagnostic nuances. This software will not render a diagnosis. It will not replace physicians. It will assist them.
Clinical decision support (CDS) systems may be the ideal situation in which to nurture such a human-machine collaboration. The practice of CDS has hardly begun. To date, its use has been mandated for only a handful of on-screen pop-ups in the increasingly EMR-based environment of healthcare. But, CDS is all but assured of eventually becoming a critical part of medical practice. When it does, intelligent CDS will be a necessity to sift through Big Data for the information to keep patients on the rails of best practices.
As each patient presents unique challenges, AI might consider different treatment options, calculating the possible outcomes and presenting them to physicians. In so doing, AI may initially prove its value in health IT, doing for human professionals what they do not have the time to do themselves.
Rather than fretting over how AI might be the ruination of humankind, we should recognize that machine intelligence may be our best hope for a bright future. It’s obvious we need help.
Right now a misguided and determined group of throwbacks – fearful that vaccines cause autism – are choosing not to immunize their children while urging other parents to do the same. In their wake, polio and measles, once thought to be eradicated, are again striking.
The improper use of antibiotics in human medicine and agriculture are producing superbugs that threaten plagues like we have not seen in centuries.
Clearly, human intelligence has had its chance and come up short.
It’s time AI got a shot.