The Algorithms Take Over

This was going to be a link for a Friday Links, but it quickly expanded from a comment to a full post.

§

A glimpse of the future. Oh, no, it’s algorithms!

(tl;dr a nurse’s union warns against using “algorithms” for diagnostic, saying that you should ask for a nurse’s judgement)

This is heavy on the tropes used against evidence based medicine in general [1], with the basic argument being that “we cannot rely on conclusions from statistics because each individual case is different”.

The point that is more directly related to computers (or algorithms) is that “algorithms are simple mathematical formulas that nobody understands”. Naturally, the “simple” was sarcastic, simple mathematical formula being an oxymoron.

In fact, some of the algorithms can use a complex formula that nobody understands. The systems get so complex that it starts to make sense to talk about the computer’s intution. The leaps of logic to arrive at a correct solution are too complex to be followed. The system cannot be comprehended. This is true not just “in practice”, but, for practical purposes, “in principle”. By this I mean two things: (1) the amount of effort required to retrace the computations becomes too large. When google returns a specific page in return to a query, the amount of computation that would be required to reproduce that by hand might be longer than the life of the universe. In addition, (2) even in cases where the result can be traced, there might be no there there, no insight. We’d be inside the Chinese Room [2]: just many numbers being crunched together.

We cannot, however fall for the easy Nirvana fallacy and thus reject this mechanical intuition. We currently already often rely on uncomprehensible intuition, just that it is human intuition. But the exact reasoning of a doctor is not always retraceable: this case matches some explicitly learned rule or some previously observed cases (this is exactly what the computer does by the way). A little species pride makes us want to be better at this game than computers are, but that should remain an empirical question, liable to be tested and disproved.

§

In fact, in order to move towards personalized medicine, we probably need more computers and automated diagnostic. Nurses can nurse and the basic sort of medicine that we have today can be provided by humans (although it can become more costumer service than reasoning [3]), but only hyper-specialized doctors will have a shot at the sort of reasoning that is necessary to be able to truly deliver personalized medicine.

Dr. Smith, here are the patients DNAs from a wide body-site sampling [4], a full record of the patients biological activity over the last two weeks. The full history is, of course, available from the online system. Which of these 100 drugs should the patient take and at what dosage? The full record of the patients response to previous drugs is available from which you may wish to infer his P-450 profile in case the genomes are ambiguous.

I mean, seriously, this is well beyond human cognitive abilities. There is nothing wrong with this. Humans are just not very good at computing. I, for one, welcome our new algorithmic overlords as much as I welcome our mechanical overlords in taking me from point A to point B.

If you really want to be treated as an individual, you will need more, not fewer computers.

§

Also note this public choice story [5]Observation is the latest fad in large part because Medicare reimbursement penalties for patients re-admitted within 30 days for the same illness do not apply if the patient was discharged from an observation unitGoodhart’s Law strikes again.

[1] There are issues with evidence-based medicine, yes; but they are not as bad as the issues with the alternative, guess-work-based medicine.
[2] Like many philosophical arguments, the Chinese Room is a bad argument, but an useful metaphor.
[3] One person one genome is so 2010.
[4] We already see with the rise of nurse-practicioners that the system is catching up to the notion that having MDs perform customer service (“Let me read what the results say on the computer. Now let me read this prescription for me.”) is wasteful.
[5] The whole thing is classic public choice: a special interest group trying to derail the public good for private benefit (last time I checked, American healthcare was very expensive; but these adds portrait attempts at cost savings as mean-spirited). It’s also interesting that nurses have more recently greatly benefitted from the trend of replacing more expensive doctors in many contexts (but that cost cutting was not mean-spirited).
Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s