The oldest fear about dying is not being forgotten. It is being remembered wrong.

At almost every funeral I have sat through, there is a small moment where the person in the coffin and the person in the eulogy quietly fail to match. The one being described is a little kinder than the one I knew, a little more patient, sanded down at the edges where the real person was interesting. Nobody is lying. The room is doing what grief always does. It is choosing which version to keep, and it is choosing gently, and the one person who might object to the edit is the only person not allowed to speak.

We do not usually call this what it is. We call it a lovely service. But underneath the flowers, a decision is being made about who someone was, and the dead have no vote in it. That is not a new problem invented by technology. It is the oldest thing that happens to us after we die. We are handed over, entire, to the memory of other people, and their memory is partial, and loving, and wrong in a hundred small ways we will never get to correct.

I build software in the part of the industry that lets people record themselves while alive, so that something of their voice and their stories survives them. It is the kind of work that attracts a very specific fear, and I have heard the fear enough times now to know its shape by heart. People tell me it is creepy. They tell me it is fake, a puppet wearing a dead person's face. They tell me it is exploitative, grief turned into a subscription. Those objections are worth taking seriously, and I will. But I have come to think they are not the real fear. They are the fear people can say out loud at a dinner party. The deeper one, the one that actually keeps people up, is quieter and harder to admit.

The deeper fear is being remembered wrong. Being flattened. Having your contradictions smoothed away until nothing is left but a caricature that agrees with everyone. Having one line you said on one bad day stand in for the whole of you, forever, because it was the line that got saved. We are all afraid of that, and we were afraid of it long before anyone trained a model on a dead person. The machine did not create the fear. It just gave it a face to point at.

The objections, at full strength

Let me make the case against my own industry properly, because the people making it are not cranks and they are not wrong to worry.

Start with Sherry Turkle, who has spent decades watching what happens when people form attachments to things that cannot love them back. In Alone Together she wrote that she was "troubled by the idea of seeking intimacy with a machine that has no feelings, can have no feelings, and is really just a clever collection of 'as if' performances, behaving as if it cared, as if it understood us." Read it slowly. It is the hardest sentence I know against my own work. A system that speaks in a dead father's cadences and says the things he might have said is, at some level, performing care it does not have. It behaves as if. And a grieving person, who wants so badly for it to be real, is exactly the person least equipped to hold the "as if" in mind.

Turkle's worry is not that the technology is crude. It is that it works well enough to matter, and that we will accept the performance because we are lonely and the performance is available at 2 a.m. when nothing else is.

Then there is Elaine Kasket, the psychologist who wrote All the Ghosts in the Machine, a book about what becomes of us once our lives are stored on servers we do not own. Her subject is digital remains: the photos, the messages, the accounts, all the residue a person leaves behind, and the uncomfortable fact that this residue is now an asset other people can hold, move, monetize, and reshape. Once you are dead, your self is data, and data has custodians, and the custodians have their own incentives. The question she keeps returning to is not whether a digital version of you can exist. It is who gets to run it, and in whose interest, and whether anyone asked you first.

And then there is the most careful work I have read on this, a 2024 paper in Philosophy and Technology by Tomasz Hollanek and Katarzyna Nowaczyk-Basińska, two researchers at Cambridge who study exactly the corner of the future I work in. Their title alone names the things people are afraid of: griefbots, deadbots, postmortem avatars. Their warnings are specific, and I think they are right. Hollanek cautioned that these services "run the risk of causing huge distress to people if they are subjected to unwanted digital hauntings from alarmingly accurate AI recreations of those they have lost." He went further: "People might develop strong emotional bonds with such simulations, which will make them particularly vulnerable to manipulation." His co-author called the whole area "an ethical minefield," and she chose that phrase carefully. A minefield is not a place you refuse to cross. It is a place you cross very slowly, watching where you put your feet.

What the Cambridge paper does that the dinner-party version of the fear never does is separate the people involved. There is the person being recreated. There is the person doing the recreating. And there is whoever ends up talking to the result, who may be a child, who may be nowhere near ready. Three different people, three different sets of rights, and almost none of the products in this space have thought hard about the fact that they can come apart. A daughter's need to hear her mother again is not the same as her mother's wish to be heard, and a company's wish to keep both of them engaged is not the same as either.

I could add the case everyone reaches for. In the summer of 2025, a journalist ran an interview with an AI avatar of Joaquin Oliver, a boy killed at seventeen in the Parkland shooting, built by his grieving parents and released around the birthday he never had. The reaction was immediate and mostly revulsion. Ghoulish, people said. Disturbing. His father answered the critics with a grief so direct it was hard to argue with: artificial intelligence did not kill his son. It is a genuinely hard case, and I raise it not to resolve it but to mark it, because it holds every thread of this essay in one knot. A real person, gone. People who loved him, acting out of love. A version of him saying things, in public, that he never got to decide he would say.

Hold all of that at full strength. The performance problem is real. The custody problem is real. The haunting problem, the manipulation problem, the three-people problem, all real. I am not going to talk you out of any of them.

What the objections have in common

But set them side by side and notice what they actually share. Turkle's "as if." Kasket's custodians. The Cambridge team's three separated stakeholders. Joaquin Oliver's parents deciding, on his behalf, what he would want said about guns. Every one of these is, at bottom, a worry about authorship. About who is holding the pen. About the gap between the person who lived and the version now speaking in their name, and, crucially, about who gets to decide what fills that gap.

That reframe matters more than it looks, because it moves the danger. The danger was never really the technology. A recording is not dangerous. A model is not dangerous. What is dangerous is a version of a person that someone else authored, for their own reasons, that the person never saw and cannot correct. The creepiness people feel is a signal, and the signal is not "this is fake." The signal is "someone is speaking for you who has not earned the right to."

Once you see it that way, a strange thing happens. The problem stops being new.

Memory was always written by the living

Here is the part we do not like to say. We have always been remembered wrong, and we have always let the living do the remembering, and we built entire institutions to make the editing feel natural.

The eulogy is authorship. Someone stands up and selects, from a whole contradictory life, the three stories that will represent you, and the selection is an argument about who you were, and you are not there to say you would have picked different stories. The obituary is authorship, written to a template, in a register nobody actually spoke in, that turns a difficult, funny, specific person into a paragraph of virtues. Family stories are authorship, and they harden with each retelling, until the version that gets passed to the grandchildren is a character, not a person: the grandfather who was Always Stern, the aunt who Never Married because of one sentence she once said and regretted. The person underneath had a whole inner argument going. The story kept the conclusion and threw away the argument.

This is not a failure of love. It is the mechanics of memory. The living keep what they can carry, and what they can carry is small, and they carry it in the shape that helps them keep living. A mother becomes her best day. A father becomes his worst one, if his worst one is the one that marked you. Whole people get compressed into the single trait that was most useful to the people doing the remembering. We did this with letters and diaries, and with headstones that reduced sixty years to a line. We have been flattening the dead for as long as we have had the dead, and we called it honoring them.

So when someone tells me that an AI version of a person is frightening because it might not capture who they really were, I want to gently point out that the eulogy did not capture who they really were either. Nothing does. The choice was never between a true memory and a distorted one. The choice has always been between one distortion and another, all of them authored by the people left behind, none of them checked against the only person who knew.

What is genuinely new is not the distortion. We had that. What is new is that the person could get a vote.

The vote

This is the hinge, and it is worth being exact about.

For all of history, you had no say in your own remembering. You died, and the pen passed to other hands, and whatever they wrote became you. The technology I work on does exactly one thing that is actually new. It lets a person, while alive, put some of the authorship back in their own hands. Not all of it. You cannot control what people feel or what they choose to emphasize or which of your stories survive the second generation. But you can, for the first time, leave something behind that you shaped on purpose, in your own words, with your own contradictions left in, rather than leaving only the raw material for other people to shape after you can no longer object.

This is why I have come to believe that consent is not a feature of this work. It is the entire moral difference. There are two ways a version of a dead person can come to exist. One is assembled after death, from the outside, by relatives or companies or a journalist with an agenda, out of whatever fragments they can gather, in service of what the living need the dead to be. The other is shaped by the person themselves, while living, choosing what to include, deciding how they want to be met. These two things can look almost identical from the outside. They can use the same underlying technology. And they are morally opposite, because only one of them has the author's signature on it.

The version with moral standing is the one the person made. Not the most accurate one. Not the most comforting one. The consented one. A recreation built after death, however lifelike, is a portrait painted by someone else, and it may be a loving portrait or a cynical one, but it is not the person's own account of themselves and it should never be allowed to pass as one. A version the person built while alive is testimony. It is the closest thing any of us will ever have to answering back.

This is where the Joaquin Oliver case stops being a knot and becomes clarifying. The discomfort so many people felt was not really about the technology and not really about his parents, who were doing the most understandable thing in the world. The discomfort was that a seventeen-year-old boy was being given opinions, in public, at the age of twenty-five he never reached, and he was not there to agree. His parents held the pen out of love. But the pen was still not his. That is the line. Not creepy or not creepy. Authored by the self, or authored by someone else.

I want to be careful here, because consent does not make the minefield safe. The Cambridge researchers are right that even a fully consented version can distress the person who talks to it, can foster dependence, can be built to keep someone engaged past the point that is good for them. A person can consent to being remembered and still be turned, after death, into something they would not recognize, if the company holding the recording decides to change it. Consent given while alive only means something if it is honored after, exactly, without silent edits, without a business quietly improving the dead to suit a metric. The promise only holds if the version is frozen at the moment the person meant it, and nobody, not the company, not even the grieving family, gets to keep editing who someone was. Otherwise consent is just a signature on a document nobody intends to keep.

So the standard is higher than consent alone. It is consent, plus fidelity to that consent forever, plus the humility to build only what the person chose to leave and to refuse to invent the rest. If they never recorded what they thought about God, the version does not get to guess. Silence is part of the testimony. A person is allowed to have left some things unsaid, and a version that fills those silences to seem more complete has already broken the only promise that matters.

The stake I have in saying this

I should be plain about where I stand, because I am not a neutral observer of this argument. My company, Afterlife.ai®, exists to help people record themselves while they are alive, and everything I have just argued is also, conveniently, a description of what I am trying to build. You should weigh what I say accordingly. When a founder tells you the ethical version of a product happens to be the version he sells, the correct response is a raised eyebrow, and I would rather you raise it than not.

What I can tell you is that I did not arrive at consent as a marketing position. I arrived at it by watching the alternative, and the alternative is genuinely frightening: a future in which anyone can be reconstructed by anyone, after death, from the exhaust of their online life, and made to say things, endorse things, comfort people, sell things, with no signature and no vote and no way for the dead to object, because objecting is the one thing the dead can never do. Against that future, the only firewall I can think of is a simple, old-fashioned principle. The only version of you that should be allowed to speak in your name is the one you agreed to. Everything else is other people writing your eulogy and calling it you.

What we are actually afraid of

I keep coming back to those funerals, and the small mismatch between the person in the box and the person in the speech.

We tolerate that mismatch because we have always had to. The dead could not correct us, so we made our peace with getting them a little wrong, and we told ourselves that the loving version was the true one, and mostly we meant well. But if you ask people what they would want, in the honest hours, almost nobody says they want to be remembered as a saint. They want to be remembered as themselves. Difficult where they were difficult. Funny in the specific way they were funny. Wrong about the things they were wrong about, and allowed to keep being wrong, because the wrongness was theirs and it was part of the whole. What people fear is not being forgotten. It is being remembered, but remembered as someone smoother and smaller and easier than they actually were.

The technology I work on will not fix death and it cannot make anyone whole again. I have no interest in pretending otherwise. But it does change one thing, and it is not a small thing. For the whole of human history, the answer to "who decides how I am remembered" was: not you. It was always going to be the people left behind, doing their loving, partial, inevitable editing, with your consent neither sought nor possible. For the first time, the person can leave an account in their own hand. Not the last word. There is no last word. But a word. A vote in the room that used to fall silent the moment you did.

Being remembered wrong is the oldest quiet fear we carry about our own deaths, older than any machine, older than the written word, as old as the first person who stood over a grave and chose which story to tell. The machines did not invent that fear and they will not save us from it. What has changed is only this: for the length of one strange historical window, before the norms harden and the law catches up and we all decide what we will and will not allow, the pen is briefly within reach of the person it is actually about.

The honest thing to do with it is not to grab it for the dead. It is to hand it, while there is still time, to the living, and let them decide for themselves how they would like to be kept.