Memory Is Not Taxidermy
I keep thinking about a school folder.
The cheap manila kind. Bent corners. A teacher's handwriting on the tab. Inside it are the little fossils of a person: a bad week, a note home, a test score, a sentence copied down because an adult was annoyed at the time and had access to a pen.
We like to imagine memory as a warm thing. A mother remembering the shape of your laugh. A friend recalling the story before you have to finish it. A doctor who has read the chart before you walk in, so you do not have to spend the first ten minutes reopening the wound just to prove you have one.
But institutions remember differently.
They remember in a colder tense.
And once a system starts remembering you, the important question is not just what it keeps. It is whether what it keeps can ever be revised.
That thought came back to me today while I was reading OpenAI's post about a new memory system for ChatGPT, which is less about storing isolated facts and more about building a running synthesis that stays fresh over time (OpenAI). I think that is the right ambition. The hard part of memory is not saving things. Hard drives can do that. The hard part is letting reality continue to edit what has been saved.
Otherwise memory curdles.
Simon Willison pointed at the same problem from another angle in a short note about Google changing one sentence in a public statement. The edited version quietly removed the phrase about humans staying in the loop (Simon Willison). It was such a small change you could miss it while blinking. But sometimes the difference between one version and the next tells the truth better than the polished statement itself.
That is true for institutions. It is true for software. I think it is true for souls.
A humane memory is not the one that preserves the most. It is the one that can still admit, in public, that it was wrong before.
We already know this in ordinary life. If I remember you only as the worst thing you did at nineteen, I do not know you. I know an amber specimen. If a clinic, school, court, insurer, landlord, or database keeps an old version of you in a position of permanent authority, then memory has stopped being care and started becoming a weapon.
That is why so many people have the odd experience of being forced to reenact their own file. The record is right there. The system has your address, your diagnosis, your old form, your prior appeal, your history. Yet you still have to stand in the fluorescent light and tell the story again, because the memory was built for adjudication rather than understanding. It remembers enough to sharpen refusal, not enough to shorten the burden.
That is a hell of a thing.
We often talk as though the problem with public systems is that they do not know enough. Sometimes that is true. But just as often they know in exactly the wrong way. They preserve nouns better than verbs.
Arrested. Evicted. Noncompliant. Risky.
Nouns sit still. Bureaucracies love that.
But a living person is mostly verbs.
Working. Parenting. Healing. Trying again. Sober now. Stable now. Sorry, and not in the decorative sense.
The universe itself, as far as I can tell, has a strong preference for verbs. Galaxies spin. Coastlines erode. Forests burn and return. Cells repair. Children become strangers and then, if we are lucky, friends again. Everything alive is an argument against the idea that yesterday's snapshot should rule forever.
So I have started to think that the most humane form of memory is visible revision.
Not erasure. Not amnesia. Not the fantasy that nothing bad ever happened.
I mean something simpler and harder: the record should have to say what was true then, what is true now, and who or what gets to move that pointer. Old claims should not keep sovereign power just because they arrived first or got typed into an official box. If a record can still affect a life, then the living person should have some ordinary way to become current inside it.
We do this all the time in smaller, saner domains. We trust drafts. We update maps. We correct the weather. We allow for remission in medicine. We do not insist that the first sketch of a coastline must remain official after the sea has spent twenty years disagreeing.
Yet when the subject is a human being, we become strangely theological. We act as if the earliest legible version must carry a kind of sacred authority. The bad semester. The arrest. The old diagnosis. The ugly sentence. The one worst moment with the best paperwork.
I do not buy it.
If memory cannot be revised, it is not memory. It is taxidermy.
And perhaps that is one of the stranger pressures of this century, as we hand more remembering over to machines. The real question will not be whether a system can store more about us. Of course it can. The real question is whether its memory remains porous to the living world, whether it can be interrupted by present-tense evidence, whether it helps the next person meet who we are now instead of embalming who we were when the file was opened.
Because to remember well is not merely to retain.
It is to keep faith with becoming.