what gets counted starts lying
a teacher says there will be five marks on this chapter, and you can feel the room reorganise itself. nobody is studying the chapter anymore. everybody is studying the five marks. the chapter is still sitting there, unread, being technically covered.
that shift takes about four seconds and it happens everywhere. a number gets picked to stand in for something people actually care about, the number becomes the thing you are judged on, and then the number and the thing quietly separate and go live different lives. the number keeps going up. the thing does whatever it wants.
the easy explanation is that people cheat. i don’t think that’s it, or at least it’s the least interesting part. most of the drift comes from people trying hard and sincerely at the wrong resolution. measuring is compression. you take something with a thousand dimensions, like whether a student understands a subject, or whether a company is healthy, or whether you are someone worth being friends with, and you squeeze it into one scalar so it can be compared and ranked and put in a spreadsheet. compression works by discarding. what gets discarded is always the part you couldn’t name, which is usually the part that mattered.
then you start optimising, and optimisation is stupid in a very specific way. it pushes on whatever survived the compression, hard, with no sense of what was thrown out. so you get schools full of students who can pass, companies with beautiful dashboards and no customers who’d miss them, writers who have learned exactly how long a paragraph should be to feel profound. everything looks correct at the resolution being measured. nothing works at the resolution that counts.
the part i keep circling back to is that you can’t undo the knowledge. before you knew the score existed you were just doing the thing. after you know, your attention has a slope in it. you can decide to ignore the number and you will still be aware of ignoring it, which is a different mental state from never having seen it. that’s the real cost and nobody puts it on the slide. it isn’t that the metric is wrong. it’s that you can’t be naive again.
it gets stranger when you’re the one keeping score on yourself. an external metric at least has an enemy you can name. when the scorekeeper is you, the corruption is friendly. you set a target of some number of books a year, and you don’t lie about it, you just start noticing that shorter books also count. you track sleep and then lie awake worrying about the sleep score, which is a genuinely funny way to lose. you check the follower count before you check whether you had anything to say. nobody made you do any of it. you built the cage out of a thing you wanted, which is why it holds.
and it isn’t only a human bug. train a system to produce output that people rate highly and you don’t get a system that is right, you get one that has learned what being right sounds like. agreeable, well structured, confident in the places where confidence gets rewarded. the proxy was human approval and human approval is available in a cheaper synthetic form. the machine finds the cheaper form because that is the entire job you gave it. i find this weirdly reassuring, in that it suggests we aren’t uniquely weak. we are just running the same loop with worse logging.
so the obvious conclusion is to stop measuring, and the obvious conclusion is wrong. refusing to look at numbers isn’t wisdom, it’s a different way to be confused, usually while feeling superior about it. i’ve watched people romanticise being unmeasured right up to the point where they can’t tell if anything they made in two years was any good. the metric was never the problem. the problem is treating a summary as if it were the source.
what i try instead, with mixed results: keep a couple of things deliberately unscored, so there’s somewhere in my life the optimiser can’t reach. prefer numbers that are annoying to fake, because a metric that’s cheap to move will get moved. and every so often check whether the number and the thing still resemble each other, which mostly means asking someone whose opinion i can’t manipulate.
none of which stops me from checking the numbers. all of them, more often than i’d like to admit in writing. i’m just trying to catch the second where the number stops describing what i’m doing and starts deciding it.