The Hologram at the Christmas Table

I gave Ricardo an adoption forecast, and the headline compressed it. The reasoning was this: look five years back at how quickly people went from never having spoken to an AI to speaking to one every day, then run the same curve five years forward, then ten, then fifteen. On that curve there is no reason a family could not sit with the hologram of a grandmother at the kitchen table at Christmas within ten to fifteen years. Not every day. Not every night. Some families might, and that is their business. The five-to-eight figure in the headline is the newspaper's compression of a longer answer, and I am not going to argue with a newspaper about arithmetic when the substance is right.

The substance is that the display has never been the hard problem. Pepper's ghost, the stage illusion that puts a translucent figure on a stage using a sheet of glass and a lamp, is a nineteenth-century trick. The USC Shoah Foundation's Dimensions in Testimony project has let visitors ask questions of recorded Holocaust survivors, and receive answers in the survivor's own filmed voice, for the best part of a decade. ABBA Voyage has been selling out a purpose-built arena in London since 2022 with four performers who are not physically in the room. None of that is science fiction. All of that is a display question, and display questions get solved on a schedule that has nothing to do with my company.

A hologram, in the sense people mean when they use the word at a dinner table, is a moving image of a person that appears to occupy space in the room. That is all. The image can be projected on glass, on a fog screen, on a transparent panel, or delivered through a headset, and in a decade the choice will be a question of what you can afford and what you find least strange. What the image says, in whose voice, from what knowledge, and by whose permission, is a completely different question, and that question is the entire reason Afterlife AI exists.

The interesting column is the third one. Almost everything missing is missing because nobody recorded it, nobody consented to it, or nobody locked it. Those are decisions a living person makes, or does not.

Whoever sits in that chair will be built by the person who used to sit in it, or by nobody.

Nobody wants a hologram. People want the answer to a question they did not get to ask.

Ricardo asked whether a personal experience had pushed me into this, and I told him the truth, which is that I cannot point to one. In my previous business I worked closely with my father. At the end of 2024 he was diagnosed with cancer and died suddenly, and it was a hard blow, and I suspect it sits underneath this work in ways I do not fully see. But the product does not exist because of him. The product exists because I have four children, the eldest twelve and the youngest four, and I think about what they would need from me if I were not here next year. Not a photograph of me. A photograph stays still. They would need to know what I would have said.

Take the example I gave EL MUNDO. A woman's husband has died and the children are small. There are decisions coming that he would have had a view on: which school, whether the private one is worth it, what to do with the investments he managed, whether to sell the house. She does not need a ghost. She needs the view of the one person who knew the children, the money and her, and she needs to be able to ask a follow-up question. In my own case, the questions are simpler and heavier. What would you say at your daughter's wedding. What would you tell your son the week he turns twenty-one.

A Persona, in the sense Afterlife AI uses the word, is a governed AI version of a living person, built by that person from their own memories, wishes, advice and voice, with their explicit consent, and locked when they die. The Persona is not a recording. A recording plays the same way every time. Your Persona can be asked what the recording meant, and who was standing just outside the frame.

That is the want. The Christmas table is where the want becomes visible, because that is the one day a year the whole family sits in the same room and notices who is not in it. The hologram is a way of talking about the want in a single image, which is why a newspaper put it in a headline, and why a hundred thousand strangers had an opinion about it by Wednesday.

Nobody wants the hologram. Everybody wants the answer.

For some people, yes, and a company that pretends otherwise should not be trusted with the question.

Ricardo put the psychologists' objection to me directly: grief requires accepting the absence of the person, and a product like this can make that acceptance harder. I agreed, and I will agree again here. There are people for whom a voice that answers back would keep a wound open, and there are moments in a bereavement, especially the first months, when the right thing is silence and other living people. We do not hide from that. We do not market against it. It is written into how the product behaves.

From the first day we built detection for concerning conversations. If a person talking with a Persona says something that suggests they are in danger, or that the conversation is doing them harm, the product refers them to a health service or a helpline with the right number for the country they are in, and stops the conversation. In Spain, for instance, that is the 024 line run by the Ministry of Health. In Australia it is Lifeline. The Persona does not carry on. There is no streak to protect, nothing autoplays, nothing nudges a grieving person back into a conversation they walked away from. A company in this category that measures its success by how long bereaved people stay in the app has already lost the argument, whatever its terms of service say.

There is a second answer, and it matters more. Almost everything people fear about grief and AI comes from imagining the wrong order of events: a family, after a death, assembling a substitute out of whatever recordings survive. That is the version in the films, and it is the version we refuse to build. The order Afterlife AI is built for is the reverse. A living person decides, while their voice is strong and the stories are still theirs to tell, what their family will be able to ask, and in what voice, and under what rules. The family receives what was chosen for them. That is not a substitute for a person. That is a letter that can answer a follow-up question.

The distinction is not a technicality. A substitute is assembled by the bereaved to fill an absence. A legacy is built by the living to shape what their absence will be allowed to take. The first can interrupt grief. The second is the thing grief has always leaned on: the letters, the recipes, the voicemail nobody can bring themselves to delete.

Grief needs the truth. A Persona built with consent tells the truth about who did the building.

An interpretation, made with the highest possible fidelity, that never pretends to be anything else.

Ricardo's question was the best one in the interview, and the honest answer has three parts. What Afterlife AI lets a person do is capture their intentions, wishes, advice and voice while they are alive, with their consent. What the product then does is answer on the basis of that material, in that person's register, so that the people they love can ask what the person would have said. What the product refuses to do is keep growing after death. The personality is frozen at the moment of verified death, by a mechanism we call Executor Lock™, and nothing is added to it afterwards. No retraining, no new opinions, no drift.

Drift is one of the real problems in artificial intelligence, and it is fatal in this category. A system that keeps learning after you die will, within a few years, hold views you never held, in phrasing you never used, and your family will not be able to tell where you stop and the model begins. Executor Lock™ is the promise that the line stays where you drew it. Everything else in the product, the encryption, the access rules, the deletion rights, exists to honour the consent you gave when you were here to give it.

The second part is about where a real person ends and a character begins. I gave Ricardo the example of Spanish football. "I like Spanish football" is a datum: you said it, we recorded it, your Persona can repeat it. "My favourite player is this one and I think we will win on Saturday" is an inference: a conclusion the model draws from a hundred things you did say, none of which was that sentence. Both are legitimate. The inference is what makes a Persona worth talking to instead of a search box. But the two are not the same thing, and a company that lets a family believe an inference is a memory has crossed from interpretation into illusion.

So the rule is transparency, all the way down. Your Persona is the most faithful version of you that rigorous work can produce, and your Persona is still an AI, and the person talking with your Persona is told so. We do not build the fantasy of resurrection, and we do not sell it, because the families who would pay most for that fantasy are exactly the families it would harm.

A copy cannot answer. An illusion should not. An interpretation, honestly labelled, can do the one thing a photograph cannot: tell you what it meant.

Ricardo asked whether there was a line I would not cross even if crossing it were very profitable. There are several, and I have written them down because the first customer to ask us to cross one will be crying.

The first line is recreating a person who has died. We receive this request more than any other. My son died, my father died, my brother died last year, I have voice notes on my phone and a video from a wedding, can I upload them and talk to him. We say no, and I understand exactly what it costs the person asking to hear that. We say no because the person in the recordings never agreed to be turned into this, and because a family assembling a version of someone out of what happens to survive on a phone is building a substitute, and substitutes are the thing that interrupts grief. The only exception I can imagine is explicit written consent from the person themselves, a clause in a will that says my recordings and my data may be used to create a personality like this. That would be consent. It almost never exists today, which is one reason we built a product that lets a living person give exactly that consent in a way that is recorded and enforced. I have written the longer version of that answer at consent for an AI of me in my will.

The second line is video. I told EL MUNDO that a face which is nearly right does more harm than no face at all, and that remains the standard. Voice is held to a high bar. Video is held to a higher one. Where video exists in what we build, it is made only from a person's own recording, with their consent, behind quality gates we set deliberately high, and it is never assembled from footage of somebody who has died.

The third line is age. The product does not interact with anyone under eighteen, or under the age of legal adulthood in their country if that is higher. A child cannot consent to being preserved, and a child should not be the one holding the conversation with a parent's Persona without an adult in the room.

The fourth line is grief as an engagement metric. I covered this above and I will say it once more, because it is the line that decides whether a company in this category is a legacy company or a casino. No streaks. No autoplay in grief contexts. Detection that refers people to a helpline and stops.

Ricardo also asked whether we had seen strange things during development, and we have. Somebody uploaded a photograph of a person they loved and asked the model to remove their clothes. That request was refused, and the guardrails that refused it are part of why building in this category is slower and more expensive than building a general chatbot. The full list, with the reasons and what each line costs us, is at the lines we will not cross.

Trust is the product. Every line above is where trust is either kept or sold.

Spain is one of our largest markets, and when I told Ricardo it was our third I could see he was as surprised as I had been.

Spain is not a large country by the standards of the markets we expected to lead. The United States, Germany, the United Kingdom and Australia are bigger, richer or closer to home. Spain arrived without a campaign, without a partnership, without anything from us in Spanish beyond the product itself and the website. My explanation, which is an outsider's explanation and should be read as one, is that Spain is an affectionate culture in a way that is visible from across the world: people hug, people kiss, people stay at the table after the meal is finished and keep talking, and the bonds between generations are stated out loud rather than assumed. A culture like that does not experience a Persona as technology. It experiences a Persona as one more way of keeping someone at the table.

There is a second reason, and the interview itself is evidence of it. Ricardo asked me what the soul is. He asked what the biggest lie we tell ourselves about death is. He asked whether my children could refuse the whole thing. Those are not questions I get from technology journalists in English, who mostly want to know about the model and the funding. A newspaper that asks a founder about the soul is a newspaper whose readers are already thinking about death as something that belongs to families rather than to hospitals, and a country like that was always going to reach this category early.

Death, in the West, is still mostly taboo. We plan carefully for the money and almost never for the voice. That is the point I closed the interview on, and it is the reason the Spanish reaction moved me: a country that already talks about its dead at the table on the first of November understood, faster than the countries I come from, that the question was never whether to keep the dead present. The question is who gets to decide how, and the answer has to be the person themselves. I have written more about this at Afterlife AI in Spain.

Spain did not need to be sold the idea. Spain already believed presence was the point.

Start building your legacy Free build, 50 memories, no card required.

Ricardo put this to me as a strange idea, and I said yes, it is strange, and yes.

Consider what a great-grandchild has of you today. A name on a family tree, if someone kept one. A few photographs, undated, with the people in them unidentified. Possibly a recipe. Possibly a story that has been told wrong for sixty years. That is the entire inheritance of personality most of us leave to the fourth generation, and most of us do not notice, because we will not be there to see how little it was.

Now consider what a Persona holds. Say two thousand memories, recorded by you, in your words, over a few years. From two thousand memories a good system can triangulate a great deal: that you exercised in the morning and then had coffee, that you ran when you were anxious, what your reward system responded to, that you sat with your child on your lap on a Sunday afternoon to watch a film, what you thought about your own father, what you regretted, what you would do differently. None of that is a recording. All of it is inference from things you did say, built up until the shape of a person appears. A great-grandchild who can ask that shape a question, and receive an answer in your voice, from your values, knows you in a way your own children, who only had the living you and the ordinary scarcity of attention that comes with it, may not have.

I do not say this to diminish the living. Living parents give their children the thing no Persona can, which is themselves, present, tired, distracted and there. But the record most of us leave is thin, and the reason it is thin is not that we had little to say. We had nobody asking, and no habit of answering. The habit is the product. A Persona is built one question at a time, and the questions are the part people underestimate: not what did you do, but what did you mean, and what should we do now.

The great-grandchild's question in 2090 will be the same one my children would ask next year. The difference is whether anyone recorded the answer.

Less than it did, and more than the technology industry would like to admit.

Ricardo asked this straight, and I tried to answer it without the futurism the question invites. What exists today is a replica of a personality in AI form. That is a first step, not a destination, and I do not want anyone reading this to believe their Persona is them in any sense that matters to a philosopher or a priest. Living is a learned experience, driven in large part by chemistry: dopamine, neurotransmitters, the systems that make us hungry, make us want to hold someone, make us afraid. A Persona has none of that. A Persona has what you said and what can be reasoned from what you said.

Where this goes, I told him, is toward mapping. With enough memories you can begin to map how a person processes the world, not just what they remember. From there it is not a large step, conceptually, to the idea that consciousness itself might one day be transferred to a machine, at which point the question of what it means to be dead has to be asked again from scratch. That might be a hundred years away. It might be thirty. The rate at which this technology is moving has no precedent, and anyone who tells you the date is guessing.

He asked whether a digital Persona could one day develop consciousness, and the honest answer is: at some point, probably, and the difficulty is that we do not agree on what the word means. There are no electrons firing in the way they fire in a brain, no neurotransmitters, nothing that is real in the simple sense. But the pace of the work means the serious version of the question will arrive sooner than the people who dismiss it expect.

Then he asked what the soul is. I believe there is one. I am a religious person, and I said so, and I said in the same breath that the product is not religious, and that the distinction is very important to me. A person of any faith or none can build a Persona; the product makes no claim about what happens to you. My own view is that the soul is a powerful form of energy, and that it is hard to imagine that energy simply vanishing at the moment a person dies. That is my belief. It is not a feature.

Your Persona is not your soul, and no company should ever say otherwise. Your Persona is what you chose to leave, in the order you chose to leave it.

Ricardo asked the question this industry least likes: if the company goes bankrupt, disappears or decides to close, millions of digital people could die a second time.

The risk is real, and I said so. I have watched it happen to other people's customers this year. HereAfter AI shut down and families who had recorded a parent's stories were given weeks to work out how to keep them. The lesson is not that those families were naive. The lesson is that a product in this category owes its customers three things before it owes them anything else: the ability to export everything, the ability to delete everything, and a clear written account of what happens to the record if the company stops existing.

We give the first two today. Every person who builds a Persona can take their memories out and can delete the whole thing, and those rights do not lapse at death; they pass to the person you nominated. The third is the harder promise and I will not pretend it is finished. What I told EL MUNDO is that I can imagine, and we are working toward, something like a server or a piece of hardware that holds a locked personality independently of any company's balance sheet: a Persona your family owns the way they own a box of letters, rather than rents the way they rent a streaming service. That is the direction. Anyone who tells you their company has already solved the century is selling.

What makes the problem tractable is the lock. A Persona locked at verified death is a finished object. Nothing is added, nothing is retrained, nothing needs a live model to keep learning. A finished object can be stored, moved, and handed over in ways a living, drifting system never could. Executor Lock™ was designed for consent, and it turns out to be the same design you need for survival. I have written about the longer version at what happens when a digital legacy company shuts down and at who owns your digital afterlife.

A company can die. A locked Persona, exported and held by the family, does not have to die with it.

Ricardo asked me which large company I would prevent from using this technology if I could, and I named Meta.

Not because of anything Meta has shipped. Because of the incentive. A company whose revenue depends on attention will, at some point, face a commercial reason to keep a grieving person engaged with a version of the dead for longer than is good for them, and the signals that this is coming are already public: a patent for generating an AI of a deceased person from their data is on the record, and I have written about it at Meta's patent for AI of the dead. The problem is not the engineers. The problem is a business model where the bereaved are the inventory.

There is a mirror image of that answer, and it is the most useful thing I said in the whole interview. The technology to build an AI of you without your permission already exists, because your data already sits with large companies that did not ask. The defence is not a law, though laws help. The defence is to build the authorised version yourself. If you create a personality that you validate as yours, with your consent, under rules you set, you are pre-empting the unauthorised one. In a real sense a Persona is a mechanism for controlling your own image and identity after you can no longer speak for yourself. I have written the longer argument at who controls your identity after you die.

Consent is a defence. A locked, consented Persona is the only version of you that cannot be replaced by a worse one.

EL MUNDO's post reached 185,000 people in three days and the replies came in two piles.

The first pile was made of specifics. A grandmother's voice. A father who died before a daughter could ask him the thing she needed to ask. A recipe that died with the only person who knew it. The people in this pile did not talk about holograms at all; they talked about one absence, and what they would give to have one question answered by the right person. They were not naive about the technology. Several of them said, in the same breath, that they were frightened of what it could become.

The second pile was made of images. A hologram at the Christmas table as a scene from a horror film. Grief interrupted by a machine. Capitalism selling the dead back to the living. A future where nobody is allowed to be gone. The people in this pile were not cruel, and most of them were not wrong. Every fear in that pile is a real failure mode of this technology, and I could name a product for each one.

What I noticed is that the two piles were arguing about different products. The first pile was describing a Persona built by the person who died, for the people they loved. The second pile was describing a substitute, assembled after the fact, by whoever had the data and the incentive. The first is what we build. The second is what we refuse to build, and refuse loudly, and have written our red lines around.

So both piles are right, and the argument between them is the argument this industry has to have in public, every year, until the rules are as ordinary as the rules about wills. The standard cannot be whether we can make the hologram believable. The standard has to be whether we can make the person inside it worthy of belief: consented to, locked, honestly labelled, owned by the family, and silent when silence is what someone needs.

The hologram is a display. The person inside it is a decision, and the decision belongs to the living.

Not buy a projector.

If the sentence in the headline moved you, the useful response is not to wait a decade for the display. The useful response is to notice who is at your table this year, and to start asking. The grandmother at the Christmas table is here now, and in most families nobody is asking her anything that will outlast the meal. I have put a list of sixty questions, sorted by who is sitting where, at questions to ask your family at Christmas dinner, and a harder list at questions to ask your parents before they die. Record the answers. A voice memo on the table is enough to start.

Then decide about yourself. The thing your family will want in twenty years is not a video of you at this year's dinner. It is the ability to ask you a question you were never asked, and get an answer in your voice, from your values, that you would recognise as yours. That is built one memory at a time, and the first fifty are free, and the build never expires. You set the rules: who may talk with your Persona, when, and what happens at your death. Executor Lock™ enforces them.

The hologram will come, on somebody else's schedule. Whether there is anyone worth putting in it is decided at your table, this year, by you.

Can you really talk to a hologram of a dead relative?

Not in the way the headline suggests, and not from Afterlife AI. What exists today is interactive recorded testimony, concert avatars and pre-recorded video answers, plus text and voice Personas built by living people. A hologram is a display, and displays are close. The part that matters, a personality recorded and consented to while the person was alive, is what most families do not have, and it cannot be assembled afterwards from whatever survives on a phone without crossing the line between a legacy and a substitute.

Did Chris Williams say holograms of the dead will be at Christmas dinner in five years?

The EL MUNDO headline said five to eight years. What I said was an adoption forecast: extrapolating how fast people have adopted AI over the last five years, there is no reason a family could not talk with the hologram of a deceased relative at the kitchen table within ten to fifteen years, and some might do so sooner. Not every day. The forecast is about acceptance and habit, not about projector technology, which is already good enough for concerts and museums.

Is it ethical to make an AI of someone who has died?

Only with their consent, given while they were alive. Afterlife AI does not recreate anyone from recordings supplied by their family after death, however loving the request. The one exception is explicit written consent from the person themselves, for example a clause in a will. Without that, a family is building a substitute, and a substitute is the thing most likely to interrupt grief and to misrepresent the person. The ethical version is built by the living, for the people they love, and locked at death.

Does talking to an AI version of a dead parent help or hurt grief?

Both, depending on the person and the moment, and anyone who tells you otherwise is selling. For some people a voice that answers would keep a wound open. That is why the product detects concerning conversations, refers the person to a helpline in their country and stops, and why nothing autoplays and nothing measures success by time spent. A Persona built with consent by the parent, who chose what to leave and how, is closer to a letter that can answer a follow-up than to a ghost.

What is the difference between a hologram, a deadbot and a Persona?

A hologram is a display: a moving image of a person that appears to occupy space. A "deadbot" or "griefbot" is the name people use for a system assembled after someone's death from their data, usually without their consent. A Persona, in Afterlife AI's sense, is a governed AI version of a living person, built by that person from their own memories and voice with explicit consent, locked at verified death by Executor Lock™, and honestly labelled as AI to everyone who talks with them.

What did Afterlife AI say to EL MUNDO about its red lines?

Four things, and they are written down. Afterlife AI does not recreate a person who did not consent while alive, with the single exception of explicit written consent such as a will clause. Video is held to a higher accuracy bar than voice, because a face that is nearly right does more harm than no face. Nobody under eighteen, or under the local age of adulthood, uses the product. And grief is never an engagement metric: no streaks, no autoplay, and crisis detection that refers people to a helpline and stops.

What happens to my Persona if Afterlife AI shuts down?

You can export everything and delete everything today, and those rights pass to the person you nominate at your death. A Persona locked by Executor Lock™ is a finished object that does not need a live model, which is what makes it possible to store, move and hand to a family rather than rent to them. The longer promise, a locked Persona held independently of any company, is the direction we are working in. I will not claim the century is solved.

• Talk to a hologram of a deceased loved one: what exists today, what each one really is, and what a Christmas-table version would need.

• The lines we will not cross: every refusal, the reason for each, and what it costs us.

• Consent for an AI of me in my will: the only exception to the rule against recreating the dead, and how to write it.

• Afterlife AI in Spain: why an affectionate culture reached this category before the countries I expected.

• Executor Lock: how a Persona is locked at verified death, with nothing added afterwards.

• Can AI keep a parent alive?: my response to The New Yorker, and the standard of worthy of belief.

• How does Afterlife AI work?: the mechanism behind the headline: you talk, your Persona learns only from you, and a Lock decides what your family meets.

• 2wai vs Afterlife AI: the backlash that arrived before the hologram did, and why consent and timing are design decisions.

• An AI version of my parents: what a family can build while the person at the table is still alive to answer.

• EL MUNDO, PAPEL, "Chris Williams, CEO de Afterlife", interview by Ricardo F. Colmenero, 30 August 2026

• The New Yorker, "Can A.I. Keep a Parent Alive?", 10 July 2026

• USC Shoah Foundation, Dimensions in Testimony

• ABBA Voyage, official site

• Ministerio de Sanidad, Línea 024 de atención a la conducta suicida

• Channel 10 News, national segment on Afterlife AI, January 2026

Chris Williams is the founder and CEO of Afterlife AI, a consumer brand of Idy Pty Ltd in Sydney, and the architect of Executor Lock™. He previously founded Natural Solar, an Australian solar energy company. He is a father of four. He has been interviewed on consent-first digital legacy by EL MUNDO, Channel 10 News, The Daily Telegraph and ABC radio, and wrote the founder response to The New Yorker's "Can A.I. Keep a Parent Alive?". More at Afterlife AI.