A birthday photograph uploaded today is not consumed and discarded. It is retained, indexed, and used to teach systems that will still be running when the child in it is grown. This is an argument about what that permits, written as a scenario, because the mechanism is ordinary and the consequence is not yet.
The oldest rule on the internet
If you are not paying for the product, you are the product. The line circulates with the frequency of a nursery rhyme, and it is worth sharpening, because as usually understood it describes the wrong transaction.
It suggests attention for access: you give a few seconds of focus, you receive a search engine or a social network, and when the screen goes dark the exchange is complete. That would be a fair trade. Attention is spent and gone the moment it is paid.
The actual currency is retained material. A photograph is not consumed by being looked at. It is indexed, tagged and kept, along with the background, the caption, the date, and the people standing next to each other. Free consumer services of the kind Facebook and Google operate are free because that material is not a byproduct of the service. It is the inventory. And inventory is not sold the day it arrives. It is held.
What a photograph is worth to a system
To a machine learning system a family photograph is not a picture. It is a labelled example, and it arrives with labels that would otherwise be expensive to buy.
A face on its own is geometry. A date on its own is a timestamp. A face attached to a date, a place, a group of other faces, and a caption written by somebody who knows everyone in the frame is something else entirely: supervised training data, annotated for nothing by the person best positioned to annotate it correctly.
The value is not in any single image. It is in the series. One photograph is noise. A decade of photographs of the same child records how a face changes with age, who that person grows up around, where they spend their time, and what changes when. That is not a memory of a birthday party. It is a longitudinal dataset with a consent form attached that nobody read.
Fifteen years is not a long time for a file
The asymmetry at the centre of this is temporal, and it is the part people find hardest to hold in mind.
A parent makes a decision about the coming week. Grandparents should see the cake. The photograph goes up. The imagined lifespan of that decision is a few days of circulation followed by a slow fade into irrelevance, the way a conversation fades.
Files do not fade. The imagined schedule and the actual one have nothing to do with each other. The decision was about this week; the object it produced will still be legible, still indexed, still searchable by face, long after the child in it has an adult life and opinions of their own. Nothing in the arrangement expires when the parent stops thinking about it.
A scenario
Nobody in what follows is real, and none of it has happened. It is written to show what the arrangement permits, not to report an outcome.
A parent uploads a photograph from their daughter's third birthday. She is holding a cake. The location is tagged, the date is automatic, and the faces around her are named, because naming them is how the feature works.
Fifteen years pass, which is nothing at all to a file. She is eighteen and deciding what to do next. For most of her adolescence she has been served advertising chosen for her by a system holding a longer and more detailed record of her life than any person in it: what she searched at fourteen, what she stopped searching at sixteen, which friendships lapsed, what time of night she is reachable. The advertising is not crude. It appeals to a wish to be of use, to belong to something, to matter. It arrives steadily, for years, and each piece of it was selected because something in the record suggested it would work.
She enlists. It feels like her decision, and in every sense that can be argued about, it is.
Her parent goes out for a newspaper and sees the photograph, the third birthday and the cake and the small face, printed beside an account of a war she had no part in choosing.
Two words for the same process
We already have language for this, and we apply it selectively.
When a hostile organisation shapes a young person's choices through years of targeted messaging, curating what they see, appealing to belonging and purpose until enlistment feels like their own idea, we call it radicalisation. We call it brainwashing. We treat it as a theft of agency, and we are not much interested in whether the person felt they were choosing freely, because feeling free is what a successful campaign produces.
When an advertising system does the same thing with better data and a budget, we call it engagement. Relevance. Personalisation. Building your future with your own hands.
The objection to brainwashing was never about the identity of whoever was doing it. It was about the process: the sustained curation of someone's information environment until their conclusion arrives feeling like their own. A technique does not become acceptable because a lawful organisation is the one using it.
There is one real difference and it deserves stating rather than skipping. A legal advertiser operates under rules and can in principle be held to them. But that difference is thinner than it sounds once the targeting is automated, individually tailored and continuous. Rules written for a billboard assume everybody sees the same billboard. Nobody else saw what she saw, which means nobody else can say what she was shown, or object to it, or even know that it happened.
What this is not
This is not a claim that any company has done this. It is not a prediction, and it carries no date. I do not know that it will happen, and nothing here should be read as saying that I do.
It is an argument about what the present arrangement permits, made while the arrangement is still cheap to change. Retention is ordinary. Behavioural targeting is ordinary. Children grow up on a fixed schedule. Nothing above requires a new invention, a conspiracy, or anybody behaving worse than they behave now, which is the part I find difficult, and the reason for writing it down.
What a person can do
Not much, honestly, and I would rather say so than offer a checklist that flatters both of us.
Post fewer photographs of children. Leave faces untagged. Strip location data. Prefer services that cannot read what you store. Each of those helps a little and none is sufficient, because the arrangement does not depend on any individual's choices. It depends on almost everybody making the ordinary one.
That advice will mostly not be taken. It asks somebody to be the only parent who does not send the photograph, and to accept a small certain loss today against a large uncertain one in fifteen years. People are not built for that trade, and the arrangement works because they are not. Which is why the answer, if there is one, is not better individual behaviour. It is shorter retention, enforced by somebody other than the party that profits from keeping it.