History used to be a story someone chose to write down. A diary, a ledger, a letter, a chronicle: the record was sparse, and what survived was the signal. It told us that a battle happened and roughly who won. It did not tell us the temperature each hour, or each soldier’s pulse, or what any of them were afraid of.
We filled the gaps with interpretation, and we knew we were doing it.
What we are building now is a different kind of record. It is dense, it is mundane, and nobody chose it. It is made of braking patterns, thermostat settings, trip times, and the quiet logs of devices that were only ever meant to make something a little more convenient. And I think we are early. The record is still thickening.
The obvious reading is that more data means more truth. I want to argue the opposite is at least as likely, and that the gap between the two is where the interesting problem lives.
How the record gets made
I want to start with how the record gets made, because it did not take a conspiracy.
In October 1999, Coca-Cola’s chairman, Douglas Ivester, told a Brazilian magazine that the company was working on a vending machine that would raise the price of a drink as the weather got hotter. His logic was textbook economics: a cold drink is worth more to a thirsty person on a hot day. Reports at the time said the technology had been tested in Japan; Coca-Cola said none of the machines were in use and that it was still evaluating the idea. People hated it anyway, and the company was soon stressing that the same machine could lower prices off-peak.
In February 2024, Wendy’s chief executive told investors the company would test “dynamic pricing and day-part offerings” as early as 2025. Headlines called it surge pricing, and within days Wendy’s said it would not raise prices at peak demand.
What those episodes share is not the pricing logic, which is ordinary. It is that people could see the mechanism. When extraction is visible, it is resisted. So the incentive for anyone who wants the data is to stop being visible.
That is where friction takes over. People weigh what a service gives them against what protecting their privacy would cost in time, effort, and attention, and often the hassle wins. That is not carelessness. It is an ordinary judgment about a real burden, and a lot of the burden is designed: how many screens stand between you and the setting you want, which features vanish if you decline, which option is already selected. Businesses notice, and compete to remove it. But each reduction in friction also makes the data easier to gather and more valuable to use, which gives firms a reason to build the next layer of collection into the product itself.
Preferences shape incentives, incentives shape infrastructure, and infrastructure reshapes the choices people face next. Collection that is invisible gives no one a reason to object, and collection that is built in makes objecting expensive. No single step is sinister. But the person who values privacy at the start of that loop is not choosing under the same conditions as the person who values it at the end, even if their preferences never changed. What changed is the price of acting on them.
The result is the record: dense, continuous, and mostly unchosen.
I will call the place where this record lives the Thing Realm. It is not the world, and it makes no claim to be. It is a representation, assembled continuously by things observing other things: cars, appliances, phones, meters.
Every point in it was captured from something real and in motion, which is why it feels like looking at the thing itself. But it is a snapshot, retaken constantly, and like any snapshot it is neither true nor false in itself. Sensors can fail, but the deeper point is about meaning, not measurement. What the picture shows becomes true or false only when someone reads it and says what it means. The record does not say anything. Readers do.
The paradox
Here is the part that I think gets lost. Dense does not mean true.
Take a hard stop on a highway. The data says the car decelerated sharply at 11:42 p.m. That is a fact. What it means is not in the data. It could be recklessness, or a deer, or someone driving to a hospital. The measurement is exact and the meaning is open. Multiply that by every trip, every appliance, every device, and you do not get certainty. You get a record that supports many different readings, each of them built from facts.
That is what the old record could not do. A sparse history forces humility, because everyone can see the gaps. A dense one invites confidence, because the gaps are small enough to forget. And the confidence is not evenly earned. The record is thick where people have connected cars, smart devices, and reliable connectivity, and thin where they do not. Whoever reads it later will inherit a very detailed picture of some lives and a faint one of others, and may not know which is which.
So the paradox is that the more precisely we record what happened, the more room there is to argue about why. And someone, or something, will be doing the arguing.
How a reading becomes an outcome
The mechanics matter here, because this is where an interpretation becomes a consequence.
A system reading that record does not have to understand you. It only has to be right on average, and to price accordingly. A single sharp stop means almost nothing. A million of them, run through a well-built model, reveal trends. Older data collection sampled people sparsely, a credit check each quarter or a loyalty card scanned on a grocery trip, and filled the gaps with demographic guesses. Continuous collection fills those gaps with behavior. The system never needs to know why you swerved. It needs to know how people like you tend to behave.
When it is right, nobody notices. When it is wrong, the error costs the system almost nothing and can cost the person a great deal. A premium rises, a claim is questioned, a line of credit tightens. The person meets the penalty without ever seeing the reasoning that produced it, and often without any way to supply the missing context.
That is roughly what some GM drivers alleged in 2024. According to reporting by the New York Times, driving data from GM’s OnStar Smart Driver feature was reaching insurers through the data brokers LexisNexis and Verisk, and GM confirmed that it shared summaries of enrolled drivers’ hard braking, hard acceleration, high-speed driving, and drive time. Some drivers said their premiums rose. Some said they had never knowingly enrolled, and that any consent was buried in fine print. GM was sued over the practice and denied that it shared data without consent, but it ended the arrangement with both brokers in March 2024.
One complaint against the company made the point that stays with me. Because the data was collected without drivers’ knowledge, it said, they never had a chance to explain what it showed, such as a sudden stop. I do not know how a court will see it. But look at what is being asked for. Not privacy, exactly. Recourse: the chance to say “that is not what it looked like,” and have someone listen.
Who gets to read it
Now put the two ideas together. The Thing Realm is a dense, mundane, unchosen record, and it supports many readings. And the systems that will do the reading are getting more capable, and in some cases harder to inspect.
A single moment in the record is a frame. A system reading it rarely looks at one. It watches the whole film: every stop, every thermostat change, every trip, in sequence. It can infer things no single frame contains, like a routine drifting or a pattern that tends to come before a claim. What reaches you is one frame of that reading: a score, a rate, a flag. The rest stays out of view. That is why “why” is such a hard question to ask. Asking it means asking to see the film.
I do not want to predict how far that goes. Researchers are working on making these systems explainable, and they may succeed. But if the machinery gets harder to read faster than we get better at reading it, then a premium that cannot be explained today is a small version of something larger: a record nobody chose, interpreted by reasoning nobody can follow, about people who cannot be asked what any of it meant. A historian reading a diary can at least guess at the author’s intent. A system reading a decade of your driving can infer things about you, and you may never get to say that it was wrong.
None of this is inevitable, and that is the part I would hold onto. Friction is engineered, so where it sits is a choice: what gets collected by default, how clearly consent is asked for, what refusal costs, and who gets to ask why. Those can be decided differently.
History was never the same as what happened. It was what someone chose to keep and how someone chose to read it. We are now keeping nearly everything, and choosing almost nothing.
The question is not whether the record is true. It is who gets to interpret it, and whether, when they are wrong about you, anyone, you included, can still ask why.