“Mathematical reasoning may be regarded rather schematically as the exercise of a combination of two facilities, which we may call intuition and ingenuity.” – Alan Turing
Sherlock Holmes is fictional expert in what he calls the “exact science of detection” (A Study in Scarlet). Despite his genius in deductive reasoning and intuition is unparalleled, much of the detective success relies upon the calm and composed guidance of his trusty sidekick Dr. Watson. In most of the canonical novels, Watson acts as the sanity check for Holmes’ storm of ideas and, of course, the meticulous chronicler of their adventures together.
After defeating its human opponents on Jeopardy, the supercomputer Watson by IBM will attempt to learn medicine. Despite its terabytes of storage and raw processing horsepower, Watson’s ability to make medical decisions remains unclear. Can IBM’s Watson truly understand the complex human body and make medical decisions, or will it – like Dr. Watson attempting deduction – prove to be an helpful sounding board but falling short of achieving true intuition?
Although a prominent writer, Sir Arthur Conan Doyle was also a physician. Medicine started much like detective work – as an intuitive science (relies on logic and deduction) rather than empiric science (relies on hypotheses and statistics). In fact, as recent as 1992, Center for Health Evidence continued to call testable, reproducible medical science “a new approach to teaching the practice of medicine.” The human body was deemed too complex, too difficult to understand by science, and reasoning – that inscrutable intuition consisting of part experience and part ingenuity – was the norm of medical practice.
From Intuitive to Empiric
Intuitive medicine continues to be the mainstay of medicine today; medical students are taught to listen carefully to the patient’s history for clues with which to deduce the underlying pathology. However, much of the mystery surrounding the mystic understanding of the human body is being reduced to mathematical facts.
In determining the appropriateness of anticoagulation in patients with atrial fibrillation, most physicians invoke the CHADS2 score and for stroke risk prediction. Using a simple formula containing the demographic and co-morbidities, a doctor can speak with some certainty whether the dangers of using a blood thinner is outweighed by the benefits gained in stroke risk reduction. Similarly, a doctor can use the CURB-65 prediction score to determine with good certainty whether a patient with pneumonia is sick enough to warrant hospitalization.
Despite the exponentially growing body of medical evidence, today’s individual medical questions remain too complex to answer using only empiric data. Answers to questions like “should I prescribe warfarin to a patient with coronary disease, strokes, atrial fibrillation and platelet count of 30,000/μL” is sufficiently complex that a clinical trial cannot be hoped to recruit enough patients. In answering complex questions, skilled physicians began to consider additional factors and recall past experiences with similar patients. For example, if she had a prior patient who took warfarin and suffered profuse bleeding, she may be less inclined to start warfarin on this patient. Instead, if she had a prior patient who took warfarin and had no subsequent strokes, she would be more confident.
The noteworthy point is this: in both cases, the patient sitting in front of the doctor is the same person despite the completely different decisions. Computer scientists refer to this form of decision-making as heuristics – the replacement of an unanswerable question with an answerable one. Computers are extraordinarily good at following heuristics because it reduces the difficulty of the calculation. However, computers perform poorly at inventing new heuristics when faced with brand new situations, whereas an ever-so innovative human can do so easily.
Dr. Watson in the Sherlock Holmes novels can quickly comprehend the reasoning behind Holmes’ deduction and recalls vividly the knowledge gleaned from their past exploits, but he continues to rely on Holmes’ deductive reasoning to assemble clues into an answer for the situation now confronting the duo. Similarly, Watson the computer is much better versed at identifying relevant medical data than arriving at a relevant answer to complex medical questions.
Can Dr. Watson Really Practice Medicine?
The complexity of human physiology and the reasoning behind how to repair its malfunctioning captivated scientists. From as early as 1959 scientists have wanted to replicate the thought process of a doctor computationally. Ironically, as medicine advances, its nature also becomes more mathematically oriented.
The bottom line is that Watson will be limited by the amount of empiric data available. Optimistic thinkers like Clayton Christensen envision the arrival of personalized medicine – an era when tests exist to diagnose diseases down to individual DNA or molecular level and when treatments exist to manipulate the each cell by the precise amount required to achieve cure. On that day, physicians will truly become obsolete, as the fully equipped intelligent machine can perform both diagnosis and treatment unaided given a sufficiently large empiric database.
Until then, physician intuition remains a necessary bridge for making the perfect medical decision in the world of imperfect medical data.
Pingback: Robots and Radiologists | Figure Stuff Out
Pingback: When IBM Watson Looks – Does It See Pictures or Patients? | Figure Stuff Out