We leave behind thousands of messages, hours of voice recordings, photos, texts. That is more than enough to build a system that answers the way we would have answered. This possibility already exists, and services are offering it. The question is therefore no longer whether it is feasible, but whether it is desirable, and who should decide.
We have covered delegating memory and attachment to artificial companions. Here is the point where these two topics converge, and it is more serious than the others.
What these systems actually reproduce
Let us be precise about what is at stake, because the vocabulary used by these services is often misleading.
A system trained on a person's traces reproduces their surface regularities: their vocabulary, their syntax, their verbal tics, their tone of voice. That is enough to create an impression of immediate and overwhelming recognition.
What it does not reproduce is what made that person who they were: their ability to change their mind, to be surprised, to react to something they had never encountered. It produces what they would statistically have been likely to say, not what they would have said.
This is the mechanism we constantly describe: a model produces the plausible continuation. Applied to someone who is no longer here, that means it produces an average of what that person was, frozen at the day of their death.
What it can offer
It would be easy to condemn, and that would miss something real.
Hearing a voice, rereading a way of phrasing things, can help. Human funeral rites have always involved maintaining a form of presence: photos on the mantelpiece, letters reread, the grave one visits and sometimes talks to. A recording is not different in nature.
And for certain situations — a parent leaving a message for a child too young to understand, someone who knows they are dying and wants to pass something on — the gesture can be deliberate and peaceful.
A photograph does not answer back. That is precisely what allows you to look at it knowing the person is no longer here. A conversational system answers, adapts, produces something new. It occupies the place of a presence instead of marking an absence. Yet the work of grief, as described by those who accompany it, largely consists of integrating that absence rather than filling it. A device that fills it is not neutral for that process.
The question of consent
This is the most delicate point, and it is almost always sidestepped.
A dead person has not consented to being reproduced, unless they explicitly planned for it. Their traces belonged to their private life, not to a posthumous use.
And there is a troubling asymmetry: the one who builds is not the one who is reproduced. A family can decide to recreate someone who would have hated it. A service can offer a feature the deceased would never have accepted.
The law says almost nothing on this point. Data protection concerns the living. Image rights largely lapse. We are in a void we have already described regarding responsibility: the technology arrives before the framework.
What it tells us about traces
There is a consequence that concerns the living, and it may be the most important one.
We produce a considerable quantity of traces every day, without ever thinking about what they will enable. A message sent to a friend in 2019 was a conversation. It has become, without anything changing, potential training material.
This invites a question we never ask ourselves: what do I want to happen to what I leave behind? Most people have an opinion on what will become of their body and none on what will become of their words. The latter is now as concrete as the former.
What to take away
We have no firm position to offer, and it would be indecent to have one on other people's grief.
What seems right is to distinguish two gestures. Preserving — recordings, texts, a voice — is what humanity has always done, and it helps. Simulating a presence that answers is something else, and it requires at the very least that the person concerned was able to decide on it.
There remains one thing these systems cannot produce, and it may be what matters. Someone we miss is missed because they could have said something we did not expect. It is that possibility which dies with a person, and no amount of traces can reconstitute it. What remains of us is not what we said: it is what we could still have said.