The potential use of artificial intelligence (AI) in medicine evokes both considerable promise and peril. For techno-optimists, AI will replace fallible human intuition with relentless algorithmic logic, informed by ever-expanding data. To techno-pessimists, AI will be plagued with errors and subject patients to autonomous diagnosis, further complicating health care delivery. But a proper vision is more nuanced.
At its best, AI relieves providers of rote tasks, leaving them more time to build the patient-clinician relationship and focus on other aspects of care that truly require a human touch. AI will inform providers who, in turn, will inform further iterations of AI, and so on.
A curious tale told recently on Twitter by video game designer Joel Burgess offers a metaphor for the process.
Burgess was a designer of Skyrim, in which gamers wander over a pseudo-medieval, dragon-infested terrain. Among other things, gamers seek treasure hidden across the landscape. Foxes, whose only role is to flee when startled by gamers, also roam the land. When Skyrim was new, some gamers asserted that if you follow a fox, it will often lead you to treasure. Burgess and his colleagues were mystified. They hadn’t programmed the foxes to seek treasure and didn’t understand why they would.
But AI isn’t a mere computer program, following mechanistic instructions. AI “thinks” and builds on its designer’s wishes.
Fellow designer Jean Simonet solved the mystery. The landscape over which Skyrim’s characters navigate is built from “navmesh”—a digital mesh of triangles. Open landscape has few details—a rock here, a tree there—requiring only a small number of large triangles. Treasure sites are highly detailed, requiring lots of small triangles. The AI that drives foxes’ behavior measures distance in triangles—not meters or feet. Foxes don’t care about treasure, but treasure sites offer the quickest way to put lots of triangles between themselves and gamers.
Gamers sensed this pattern and started following the foxes to treasure. In effect, designers, AI-foxes, and gamers melded into a synergistic system, accumulating greater knowledge and efficiency than any alone would have possessed.
At its best, AI operates similarly in medicine. Programmers create software that discovers new things (or appears to do so), and intuitive providers discern previously unknown patterns in illness and treatment. Machines don’t replace providers; they enhance them.
Some years ago, an official at the electronic medical record company, Cerner, told one of us (Graboyes) a medical anecdote that parallels the AI-fox case. Cerner had developed an algorithm, St. John Sepsis Surveillance Agent, to detect pathogens in hospital rooms, enabling the staff to isolate these dangerous sources of infection before they spread across the hospital. Somehow, according to the official, users noticed over time that this algorithm could also indicate the onset of PTSD or suicidal impulses, neither of which are pathogen-related.
Just as AI-foxes aren’t really interested in treasure, St. John Sepsis Surveillance Agent may not have been “interested” in suicide. But somehow, its behavior uncovered something in the direction of suicide, and caregivers came to recognize this correlation.
There are three broad reasons why AI will come to the forefront in coming years as an essential tool of health care.
First, expanding volumes of useful data burden providers with information overload. Second, with incomes rising and people living longer, the demand for health care providers will outstrip supplies in the coming decades. Third, new health care technologies emerge every year, and providers can’t keep up with the capabilities or even existence of all such innovations.
AI is uniquely capable of lessening all of these problems. Computers can process vastly more data than any individual provider or group of providers. It can relieve physicians and others of the rote, repetitive tasks that divert them from their patients. And AI can effortlessly stay current on new innovations and, with superhuman powers of pattern recognition, discern heretofore unimagined capabilities of those innovations. Like Skyrim’s virtual foxes, the outpouring of AI-generated information will allow providers and patients to discover previously unseen opportunities to improve care.
AI won’t replace doctors and other health care personnel. Rather, AI offers the possibility of achieving the elusive trifecta of better care, more care, and lower unit costs for care.
