Tuesday, December 18, 2018

Autonomous Cars and Doctors

Few people understand this, but the practice of medicine is almost entirely a matter of procedures and protocols. It's about memorizing flow charts, zillions of them, as well as the chemical, biological, structural, and anatomical basis for them. Med schools aren't tough to get into (and out of) in order to select particularly talented and resourceful hero/healers. On the contrary, admission is crazy selective in order to filter out the well-rounded and broad-minded. We want single-minded nerds who can memorize all that stuff and apply it without screwing up. That's the whole ballgame: not screwing up.
A very good doctor hardly ever screws up.
A fair doctor mostly doesn't screw up, at least on important stuff.
A lousy doctor screws up.
(and, needless to say, the nurse, pharmacist, and, most egregiously, you and me screw up prodigiously, regardless).
There's a popular notion that especially good doctors - the fancy expensive ones - put a personal spin on their treatment, wielding resourceful and unconventional moves to uniquely respond to their patients' unique health situations. If you ever find any such a doctor, run away fast.

The problem is that a single doctor operates from a few hundred data points, whereas medicine as a whole operates from millions of them. Protocols have been honed to the best of our ability, and while medical science may not have all the answers, it unquestionably has the most rigorously proven ones. So deviating from protocol out of intuition, creativity, or other fuzzy subtleties places patients in no-man's-land; an untested realm. Gambling. 

While as a chowhound I resist well-trodden pathways, as a patient I prefer to be placed smack in the bright center of the lane. Don't give me your extra-special best attention. Don't play out your hunches or untested regimens. Feed me into the machine and treat me exactly like the jillion who came before me. That's where my best odds are.
Unless, that is, your hunches and untested regimens are utterly harmless. I actually have some hunches and regimens of my own, some surprisingly effective (propolis, for example), but that's not where I seek when things get serious...except in scenarios where medicine has nothing viable to offer (the sole realm where a "God of the gaps" perspective, otherwise fallacious, might make itself practically useful).
Medicine will soon be dominated by algorithms and robots. They lack intuition and judgement, but in most cases, that's a feature, not a bug. What's most important is that they never screw up.
One seeming exception is diagnosis, which is considered something of an art in edge cases - e.g. when rare diseases are hard to distinguish from more common ones. One might hesitate to have computers handle this part, even though human doctors are notoriously prone to misdiagnosis. An innate fallacy makes humans feel most indispensable at their greatest failure points.

Same with driving. One might insist that human experience/intuition/higher-level judgement is invaluable to safe driving, but the overwhelming majority of traffic accidents are due not to inadequate subtlety of judgement, but to plain old stupid screw-ups.

I've got very good driving intuition. I can spot, for example, when a driver is growing aggravated by a slow car in front, and is impatiently pressurizing an impulse to spazzily dart into the passing lane. Similarly, I can tell when someone awkwardly inviting me to proceed is poised to step on the gas if I hesitate. Spazzy driving bursts are not difficult to anticipate (it's not telepathy, just a read of ongoing micro-actions - wheel direction and speed changes, general jerkiness, etc.). It will be a long time before autonomous cars match my street smarts.

But that ignores the salient point. If I were to honestly survey all the cat lives I've used up by committing blunders while no one happened to be in the way, dozens of potential calamities have strewn my 40 year driving career. However we absolve ourselves via amnesia, we're fortunate that our errors haven't killed multitudes. While an autonomous car would lack human wisdom, it wouldn't make the infrequent stupid blunders that kill people.

As AI powers more and more critical tasks, and we feel leery about the loss of human intuition and judgement in such systems, bear closely in mind that stupid blunders are what kill people.

No comments:

Blog Archive