Proof of Person

The message arrived the morning after your father’s funeral, from a college roommate you had not heard from in six years. Four sentences, and every one of them landed: the memory it reached for was specific, the tone was pitched exactly between grief and gratitude, and it closed on the one detail about your father that a stranger could not have known. Ten years ago you would have kept it. You almost did — until a quieter thought arrived first, unbidden, the way surprise always arrives: did he write this?

Notice what the question did. It did not accuse your friend of anything. It did not even lower your opinion of him — he remembered, he reached out, he cared enough to send something. What the question did was cut the wire between the message and the man. Four sentences that would once have been evidence about a specific person became evidence about nothing in particular, and the warmth you were about to feel had nowhere to land. Yesterday I argued that trust between a person and a system is retired attention — a prediction whose error rate fell low enough that attention withdrew. Today the harder case: what happens to trust between two people, when the sentences that carried it stop being handmade.

For nearly all of history, a paragraph was proof of work. Not proof of good work — proof that a specific mind spent specific minutes on you. The letter could be clumsy and still be precious, because the clumsiness itself certified the hours. Words were expensive to produce, so the words carried information beyond their content: who made them, how carefully, in what state, at what cost.

The people you know are, under the Reality Equation, models you run. Your friend exists in your head as an Expectation — a compressed prediction of what he will say, how fast, in which words, with which characteristic mistakes — and every message he ever sent you was training data. The way he reaches for the wrong word in exactly the same way each time. The two-day silence that means he is thinking, not sulking. The joke placed one beat too early. When a message surprised you, the surprise was news about him, and your model updated: he is angrier than I thought, kinder than I thought, further away than I thought. Attention paid the fee, and collected the update. This is what knowing someone is. There is nothing else it could be.

The signal could always be faked — Cyrano under the balcony, the secretary drafting the senator’s condolences — but forgery was expensive, rationed, one letter at a time, and so rare that the default assumption held: the words came from the person whose name they wear. Ten thousand years of correspondence rested on that default. It was never written down anywhere, because it never needed to be.

That default is now false, and it failed in the space of about three years. The stack drafts the difficult email, softens the apology, punches up the toast, finds the specific memory in a photo archive and sets it perfectly in a condolence note. The cost of a polished paragraph has fallen to nothing, and a signal that costs nothing certifies nothing. This is the familiar half of the problem, the half everyone has noticed: polish no longer proves care, fluency no longer proves thought, and the beautiful message is exactly as likely to mean everything as nothing.

The deeper half is quieter. When your friend’s stack writes in your friend’s name, the words still arrive, your model still updates — but it is updating on the output of a different entity. You are learning the machine’s manners and filing them under his name. Do this for a year and the person you know diverges from the person who exists: your Expectation of him tracks a system he merely approves, smoother than he is, prompter than he is, incapable of his particular failures. And his failures were half the signal. The surprise channel between you — the channel that knowing him was built from — goes quiet, not because he stopped changing but because his changes stopped reaching the wire.

You are updating a model of your friend with the output of someone else’s machine — and filing the result under his name.

So the signals invert. For ten thousand years, polish was the costly signal and roughness was noise. Now polish is free and roughness is expensive — expensive because it must be risked. The typo, the awkward sentence, the message that arrives too fast to have been engineered, the voice that cracks on a live call: these become the signature, because they are what the smoothing layer would have removed. People already feel this without naming it. The unedited voice memo lands differently than the composed text. The wrong word used the way only he uses it is worth more than the right word used the way everyone does.

But hold the celebration of the authentic typo, because it will not survive either. The moment imperfection becomes collateral, imperfection gets counterfeited — the models write “authentic” roughness on request, complete with hesitation and a misplaced comma. Any signal that can be transmitted can be forged, which means every static proof of person eventually falls. What survives is only what must be performed live, against unrehearsed reality, at conversational speed: presence, and the responses no one could have staged because no one knew the question was coming. Surprise, once again, is the only honest witness.

Here is the forecast, and it is checkable. Within a few years, proof of person becomes an explicit layer of ordinary life — social, economic, and eventually technical — the way proof of identity became one when strangers started transacting at distance. Not because anyone legislates it, but because trust between people is too valuable to run uncollateralized, and the old collateral is gone.

You can already watch the layer assembling. Schools drifting back toward the oral exam and the supervised essay, not out of nostalgia but because the take-home essay stopped proving what it was assigned to prove. Hiring processes adding live, unscripted rounds. The etiquette hardening in private life: there are messages — the condolence, the apology, the declaration — that people increasingly sense must not be delegated, where the labor is the message, and outsourcing it is not a shortcut but a lie about attention paid. The word for that offense does not exist yet. It will.

Already true

Polished prose has stopped functioning as evidence of effort in any domain with stakes — teachers, editors, and hiring managers have quietly re-weighted it to zero. Live and in-person formats are returning wherever proof matters. The unedited channel — voice memo, live call, handwriting — already carries a premium that nobody negotiated.

What has to happen

The premium becomes explicit. Institutions name the live round as the round that counts. Somebody ships person-attestation for messages and it matters less than presence does. Delegating the intimate message acquires a name and a stigma, the way plagiarism did — a word for lying about attention paid.

Where I am probably wrong

The opposite equilibrium is available: we may simply stop expecting text to carry the person at all. Messages become logistics, nobody is insulted by the drafted condolence because nobody reads it as personal, and personhood retreats entirely into presence without any market for proof ever forming. If in five years the drafted apology is as unremarkable as the printed invitation, this post named a transition and mistook it for a destination.

Either way, one asset survives the repricing. The channel that cannot be delegated is the one I keep arriving at from every direction this month: attention itself. A stack can write your words, hold your calendar, answer in your name — but it cannot sit in a room and be surprised on your behalf, and the other person can tell. The words were only ever a receipt for attention paid. Now that anyone can print the receipt, the payment has to be witnessed. The last proof of person is the one thing that cannot be sent.