Why You Might Want to Be Harder to Read

In the late eighteenth century, foresters in Prussia and Saxony set out to make their forests make sense. A natural forest was a tax official’s nightmare: trees of every age and species, undergrowth, dead wood, paths that local people used for grazing and gathering and firewood. None of it could be counted. So the foresters cleared it and replanted in straight rows of a single fast-growing species, all the same age, and for a generation the result was a triumph. The forest could finally be read at a glance and taxed.

Then the second generation of trees began to fail. The monoculture had stripped out the tangle that had been doing the work: the mixed species, the rot, the life in the undergrowth that fed the soil and kept pests in check. German forestry needed a new word for what happened next, Waldsterben, forest death. The forest had become perfectly readable, and it was dying of it.

James C. Scott opens Seeing Like a State (1998) with this story, and his word for what the foresters wanted is legibility. States cannot govern what they cannot read, so they simplify. They invent permanent surnames where villages had fluid, local names. They draw cadastral maps and lay cities out on a grid. Scott is careful not to call this evil. The census that makes taxation possible also makes a vaccination programme possible. His argument is about what gets lost when a living, local, illegible order is flattened into something that can be read from the centre, and how often the flattening kills the thing it was meant to manage.

The foresters wanted the forest to be readable. Something similar now wants you to be.

A machine that reads you better than your friends do

In 2015 Wu Youyou, Michal Kosinski and David Stillwell published a study in the Proceedings of the National Academy of Sciences. Tens of thousands of volunteers filled in a standard personality questionnaire, and the researchers compared two kinds of judge. One was the people who knew the volunteers: colleagues, friends, family, spouses. The other was a computer model fed nothing but each volunteer’s Facebook likes.

With ten likes, the computer judged personality better than a work colleague did. Seventy likes put it ahead of a friend, 150 ahead of family, and 300 ahead of a spouse. The authors noted that the average user had a couple of hundred likes to give away.

The result is usually told as a story about machines knowing us better than our loved ones. Look at how accuracy was defined, though. The main test was agreement with the volunteers’ own answers on a questionnaire, so what the computer was best at predicting was the self a person reports when asked to describe themselves in standard categories. It read the version of you that fits in the boxes, and it read that version better than the people who have sat across a table from you for years.

That is legibility in Scott’s exact sense. The scattered, contradictory record of what you respond to gets turned into a small number of dimensions that can be sorted and sold to. Every surface now asks for a declaration: age, politics, hunger, fear. The rough edges, the parts that stutter or swell or shrink depending on the day, get cropped away, and what is left is the most profitable silhouette.

The reading does not stop at reading, either. A recommendation system is built to show you more of whatever you have already responded to, so the version of you that it can see is the version it keeps feeding. Each round makes the profile a little more confident, and the feed a little more like the profile. The foresters at least had to wait a generation to find out what their rows had cost.

What gets lost when a person becomes easy to read

It shows up in small ways. There is the faint vertigo when a filter erases the lines your life has carved into your face, and the strange sadness when predictive text puts a word in your mouth that you had not chosen, and you send it anyway, because it was close enough.

Close enough is the whole problem. The foresters’ rows were close enough to a forest to fool a generation. A profile is close enough to a person to sell to. And a person who spends years being read in those terms can start reading themselves that way too, reporting their moods as preferences and their confusions as opinions, tidying the undergrowth before anyone asks, because the tidy version is the one that gets recognised.

Scott’s forest suggests where that ends. What made the old forest alive was exactly what made it illegible: the mixed ages, the dead wood, the uses nobody had written down. It is not obvious that a person is different. The griefs that never resolve into a category, and the parts of you that change depending on who is in the room, may be less like noise in the data and more like the soil.

Scott had a name for the kind of knowledge the old forest carried and the rows destroyed. He called it mētis, a Greek word for practical cunning: the know-how of the farmer who can tell from the colour of the soil when to plant, or the sailor who reads weather nobody has written a manual for. Mētis is real knowledge, and it resists being written down, because it lives in the doing and changes with the conditions. Most of what a person knows about their own life is like that. You know how to be with one friend and not another, when to leave a party, what a certain silence in your own chest means, and almost none of it would survive being turned into a questionnaire answer. It is the part of you that no dashboard can see, and it is also, quite often, the part that is actually running things.

Illegibility is not a tantrum

None of this is an argument for disappearing. Scott did not want the forests left unmapped, and there are good reasons to be findable, not least the ordinary business of being someone to other people. Refusing to be read at all is its own kind of simplification, a profile of the person who has no profile.

The case is narrower than that. Some of you does not need to be converted into a signal. Some days can go uncaptioned, and some griefs can stay unsorted while they are being felt. Illegibility, kept on purpose and in small amounts, is closer to care than to protest: care for the parts of a life that grow too slowly to be captured, and would not survive being made to fit.

The research result stands. Given a few hundred traces of your behaviour, a model can reproduce your answers about yourself better than most of the people who love you. What it cannot tell you is whether those answers were ever the part of you worth knowing, or whether the thing that matters was always in the undergrowth, where nobody was counting.