Let’s talk about why a memory is not the same as a person.
There is a very ordinary human scene here.
A phone in a drawer.
Old voice notes.
Photos nobody has deleted.
A WhatsApp thread you can’t bring yourself to archive because the last message is still sitting there, stupidly normal. “Call me later.” “Made it home.” “Love you.”
Before this became technology, it was just grief.
Messy, private, uneven grief.
You heard someone’s voice in your head. You found their handwriting in a cupboard. You kept a jacket because it still held some impossible trace of them. You replayed a video too many times and then stopped, because at some point remembering starts to hurt in a different way.
That’s human.
Painful, but human.
Now the machine enters the room and says: I can make them answer.
Not just preserve the photo.
Not just restore the recording.
Answer.
It can take the messages, the emails, the videos, the voice samples, the little scraps of a life scattered through cloud storage, and build an interface that talks back.
It can sound like your father.
It can text like your partner.
It can tell a child something that feels like advice from a dead parent.
And the first instinct is obvious.
Of course people want that.
Of course they do.
Anyone pretending this is only creepy has not sat with real loss for long enough. Grief will bargain with almost anything. It will bargain with religion, with dreams, with old recordings, with smells, with silence. So yes, it will bargain with software too.
That is not the part I’m worried about.
The dangerous part is when companies look at grief and see retention.
Because a griefbot is not just a product.
It is an emotional dependency layer built on top of someone’s remains.
And if we are not careful, the dead will become the next subscription category.
A voice you can pay to keep.
A personality you can upgrade.
A memory that lives inside a platform account.
A loved one behind a login screen.
There is something obscene about that, if we say it plainly.
Not because technology should never help people mourn. That’s too simple. Technology already helps us mourn. We use photos, videos, shared folders, memorial pages, audio recordings. Sometimes those things are precious. Sometimes they are the only proof left that someone laughed a certain way or held a room a certain way.
The line is not between “technology” and “no technology.”
The line is between preservation and simulation.
Preservation says: here is what they left.
Simulation says: here is what they might say now.
That gap matters.
It matters morally.
It matters legally.
It matters for the people still alive.
Because once the machine starts producing new words in the shape of a dead person, you are no longer dealing with memory. You are dealing with authorship without an author.
A mouth without a living will behind it.
A performance made from evidence of a life.
And performance is not neutral.
Someone chooses the model.
Someone chooses the guardrails.
Someone chooses what the dead person is allowed to say, what they are forbidden to say, what tone they use, what version of them survives.
Do they become softer than they were?
Wiser?
More agreeable?
Less complicated?
Does the difficult father become a comforting chatbot because that sells better?
Does the dead mother become endlessly available because the platform has no concept of sacred distance?
Does the person get edited into something easier to consume?
That is the quiet violence here.
Not the sci-fi horror version where the machine becomes evil.
The boring product version.
The dashboard version.
The “continue your conversation” button.
The “premium memories” tier.
The little notification that says: “Someone is thinking of you.” Except someone isn’t. The person is gone. The system is nudging an attachment loop because engagement metrics do not understand mourning.
And people will say, “But if it helps, what’s the harm?”
Fine. Let’s take that seriously.
Comfort is not fake just because the tool is artificial. A song can comfort you. A photograph can comfort you. A letter can comfort you. A recorded birthday message can absolutely wreck you and still be a gift.
But a tool can comfort you and still capture you.
Both can be true.
The question is not, “Can this feel healing?”
The question is, “Who has power over the relationship once it becomes interactive?”
That is the human-sovereignty question.
Who gave permission?
What exactly was permitted?
Can the family revoke it?
Can the person say no before death?
Can the system be exported?
Can it be deleted?
Can it be audited?
Can it be prevented from selling, training, recommending, or reshaping the dead person’s likeness?
Can a child be protected from forming an ongoing attachment to a synthetic parent that never ages, never changes, never gets tired, and never has the dignity of silence?
Those are not abstract questions.
They are product questions.
They are contract questions.
They are design questions.
They are consent questions.
And consent after death is not a checkbox someone else clicks because the UI needed to move forward.
If a person’s voice, face, messages, writing style, jokes, private fears, and intimate patterns can be reconstructed after they die, then digital remains need to be treated like remains.
Not like content.
Not like training data.
Not like unused assets in a cloud account.
Remains.
Something with dignity attached.
Something the living are allowed to hold carefully, but not casually exploit.
This is where the law will probably arrive late, because the law usually arrives after the product has already normalized the weird thing.
First it will be novelty.
Then it will be therapy-adjacent.
Then it will be a memorial feature.
Then a celebrity estate will license it.
Then a platform will package it for families.
Then somebody will sue because a dead person’s synthetic voice said something they never believed, endorsed something they never used, or comforted someone in a way that crossed a line nobody had written down yet.
And everyone will act surprised.
We should not be surprised.
This is what happens when identity becomes programmable.
The living become profiles.
The dead become datasets.
Memory becomes an interface.
And grief becomes a market with very good retention numbers.
The operator lesson is simple, and it is uncomfortable.
If you are building anywhere near voice, likeness, memory, agents, personal data, family archives, estate planning, creator tools, or AI companions, you need a doctrine for digital remains before the edge case becomes the product.
Not a legal paragraph buried under “terms.”
A doctrine.
What will you never simulate?
What requires explicit consent while the person is alive?
What can family approve, and what can they not approve?
What gets watermarked?
What gets logged?
What expires?
What is exportable?
What is deletable?
What is never used for training?
What happens when the most vulnerable user is a grieving person at 2:13 in the morning with a credit card and no emotional defenses left?
That last one matters.
Because the market will not always protect that person.
Markets are very good at discovering willingness to pay. They are much worse at recognizing when willingness comes from pain.
And grief is pain with a wallet, if you let the worst incentives define the category.
So here is the standard I would use.
Preserve what was real.
Label what is generated.
Require consent before simulation.
Keep the dead person’s data portable and deletable.
Never lock a loved one behind a subscription.
Never let an AI system pretend it is the person.
And never design for dependency when the user is grieving.
That sounds strict because it should be strict.
There are places where friction is humane.
There are moments where a product should slow someone down.
There are buttons that should be harder to press, not easier.
There are prompts that should say: wait, this is not them. This is a simulation built from traces of them. Use it carefully. Leave when you need to.
That would be honest.
Less seductive, maybe.
Worse for engagement, probably.
Better for humans.
And that is the whole damn point.
The dead do not need better personalization.
The living need better protection.
They need ways to remember without being harvested.
Ways to preserve without being manipulated.
Ways to grieve without being dragged into an endless synthetic conversation that slowly replaces the harder human work of mourning.
Because mourning has a shape.
It is not efficient.
It is not optimized.
It does not convert cleanly.
It comes in waves. It ruins normal days. It makes you keep ridiculous objects. It makes you forget, then remember, then feel guilty for forgetting. It is one of the last deeply human things no machine should be allowed to redesign around engagement.
A memory can stay.
A recording can stay.
A photograph can stay.
A letter can stay.
But when the machine says, “I can bring them back,” the answer has to be sharper.
No.
You can imitate traces.
You can produce language.
You can generate comfort.
You cannot bring them back.
And if we blur that line because the demo is beautiful, we will lose something sacred under the cover of innovation.
Not all boundaries are fear.
Some boundaries are love with a spine.