In a recent post, I explained why I was looking forward to the day when Dr. Watson, the medical version of the Jeopardy! champion Watson supercomputer built by IBM, started offering diagnoses. I asked readers whether they shared my enthusiasm, and got some very interesting comments back.
Karthick Hahiharan spoke for many when he wrote “With all due respect for Dr.Watson, I am still a firm believer of Human intuition and incomprehensible ability of humans to solve even complex situations. Our Knowledge is certainly not limited to data.” And Massimo wrote “You assume that about medical issues there is ONE truth, which is true for Jeopardy! questions by design, but not for human-related issues. So, finding correlations between hospital cases can be done, giving a diagnosis (except for the trivial ones) cannot. Dr. Watson will not replace Dr. House…”
These folk and many others feel that medical diagnosis is an ineffably human task, one where the best humans are always going to be better than the best machines. This is because, the argument goes, good diagnosis requires a combination of intellect, knowledge, and intuition. The intellect is innate, the knowledge is picked up through long years of study and experience in medical school, residency, and practice, and medical intuition — the ability to see more than what’s in the test results and patients’ self-descriptions — is the product of both inherent ability and accumulated experience.
Can we replicate and improve on each of these digitally with Dr. Watson? Yes, we can.
Let’s take intellect and knowledge first. As Watson demonstrated so convincingly on Jeopardy!, artificial intelligence is now astonishingly good. As I wrote, we can put all the world’s accumulated medical knowledge in a database, turn armies of algorithms loose on it, and when presented with a set of symptoms arrive at a diagnosis within seconds. No human brain can do this, and the digital ones are getting better at it all the time.
But what about intuition? Don’t human senses, minds, and guts allow us to intuit things that machines just can’t? Won’t the best human diagnosticians notice that a patient’s skin is slightly jaundiced, or that he suddenly breaks off eye contact when stating that he’s been taking his meds faithfully, or that his voice changes tone when he answers questions about illegal drug use?
Yes, the best diagnosticians will do all these things. But as I wrote before, they’ll do them inconsistently and with great overconfidence. These problems are so great that they typically negate the advantages of intuition over algorithm even for experienced clinicians, as careful research has shown. And most doctors, of course, have less than world-class intuition, yet still trust in their own ability to ‘go beyond the data’ and arrive at a diagnosis after face-to-face interactions with their patients.
The best responses to this group come from the fictional diagnostician Massimo mentioned in his comment. Dr. Gregory House, the troubled-yet-brilliant star of Fox’s hit show “House,” has some choice quotes about the importance of listening to patients:
- “I’ve found that when you want to know the truth about someone that someone is probably the last person you should ask.” [#105]
- “You want to know how two chemicals interact, do you ask them? No, they’re going to lie through their lying little chemical teeth. Throw them in a beaker and apply heat.” [#113]
Most of us pride ourselves as good judges of character and sniffers of lies. And most of us are kidding ourselves. The world’s best diagnosticians and poker players can accurately sense dishonesty. The rest of us should, like Dr. House, just stick to the facts.
Medical diagnosis is an exercise in pattern matching. The fact that people are biologically and psychologically complex shouldn’t distract us from that essential fact. Once computers have demonstrated that they’re better at this pattern matching exercise than humans, why would we not give them the job?
One good reason is that, as my former student Humberto Moreira pointed out, people might be more likely to accept a diagnosis given by a person rather than a machine, and so to follow through on the recommended treatment. Human interaction and interpersonal ritual comfort many if not most people, and it’s counterproductive to ignore this fact.
So one approach here is to follow commenter Julius Campbell‘s advice to have tomorrow’s Dr. Watsons team up with human diagnosticians. As he wrote, “I would rather not have Dr. Watson give me a diagnosis instead of a human doctor. I would rather get my diagnosis from a human doctor consulting with Dr. Watson.” This would improve quality and provide ritual, but would also not cut costs a lot (except for the downstream costs caused by lousy initial diagnoses). Everyone agrees that we need to ‘bend the curve’ of health care costs, and having humans involved in all diagnoses keeps us on our current ruinous trajectory.
So we need to do something very different in future health care delivery models. I advocate (probably in futility) that as part of this shift we get over out infatuation with human intuition. Like all infatuations, it’s keeping us from seeing things as they really are. And like a lot of them, it’s costing a lot and leaving us worse off in the end.
I’d love to hear what you think on this topic, so please leave a comment if you’d like. I have to say, though, that I’m really not interested in hearing more expressions of confidence and faith in human intuition in this context. Such expressions are increasingly flying in the face of the evidence. Hearing them is like hearing someone tell me that Boston will have 300 sunny days next year. I’d sure like to believe it, but I know better. I don’t know if Dr. House was talking about diagnosis, intuition, and confidence when he said “You know what’s worse than useless? Useless and oblivious.” But he might as well have been.