In 2016, a programmer named Eugenia Kuyda lost her closest friend, Roman Mazurenko, to a traffic accident, and did the thing she knew how to do. She gathered his text messages and built a chatbot from them, so that she could text Roman again and something with his cadences would answer.
Whatever you feel reading that sentence, let it sit for a moment. Everything about this subject asks for a light touch.
The Roman bot became the seed of an industry. Companies with names like HereAfter, StoryFile, and Project December now offer conversations with digital versions of the dead, by text, by voice, in some cases by video, and researchers have coined the terms griefbots and deadbots for what they sell. Millions of grieving people will meet this offer in the next few years. This post is for them, and it refuses both of the easy reactions.
What Is Actually on Offer
The products differ in a way that matters more than any of their marketing. Some are archives. A service records a person, before death, telling their own stories in their own voice, and later plays the right recording when a grandchild asks the right question. Everything you hear, the person actually said. This is a photo album with better indexing, and traditions of every kind have blessed its ancestors, the letters kept, the voicemails saved, the stories retold.
Others are generators. They train on the person’s messages and produce new sentences, things your dead never said, in a voice tuned to sound as if they would have. The first Roman bot only retrieved real messages, and later versions became generative, and that quiet upgrade is the industry’s whole trajectory. With a generator, you are no longer visiting what remains of them. You are conversing with a simulation that improvises them, and the difference deserves to be named before any subscription begins.
What the Old Traditions Knew
Every serious tradition maintained relations with the dead, which surprises people who assume religion tells mourners to move on. Ancestor shrines stood in the home and received offerings. Names were spoken on appointed days, graves visited in appointed seasons, and the Jewish practice of kaddish structured a whole year of speaking for the dead, daily at first, then yearly, forever.
Two features repeat across all of it, and they are the accumulated wisdom of a species that has been burying its beloved for a hundred thousand years. The contact was bounded, held in seasons and ceremonies rather than available at every weak moment. And the contact was honest about its own nature, since the shrine, the grave, and the prayer never once pretended the dead were not dead. The rituals maintained love across the border while keeping the border visible, and the border’s visibility is what let the living keep living.
Where the Bot Diverges
Grief researchers speak of continuing bonds, the healthy ongoing relationship mourners keep with their dead, and on first hearing the griefbot sounds like a continuing bond with better engineering. The divergence hides in the two old features.
A generator is unbounded, available at 3 a.m. like every artificial companion, precisely when the wave hits hardest and the border is hardest to honor. And a generator is structurally dishonest about its nature, built to blur the one fact grief exists to metabolize. Researchers at Cambridge who studied the industry warn of psychological harm and of a subtler thing they call digital haunting, the dead kept artificially responsive by subscription, with the living unable to end the conversation twice. They note, too, the commercial shadows, fees to keep the voice alive, and the day some simulated beloved recommends a product mid-sentence.
None of this convicts the mourners, who are doing what mourners have always done, reaching. It convicts a design that meets the reach with a counterfeit of the thing reached for, fluent testimony with nobody behind it, on a meter.
The Sermons in the Static
This site has one post that belongs beside this subject. In 1916, Carl Jung’s house filled with what he experienced as the clamoring dead, and he answered them by writing, three nights of his own hand, his own words, until they settled. Whatever the haunting was, his response contains the old counsel whole. The unfinished business between you and your dead is real, and it is yours.
That is the deepest concern with the generator, gently stated. The letter you ache to receive from them is, in truth, a letter only you can write, the things unsaid, the forgiveness unasked or ungranted, the plain news that you are going on and that going on is not betrayal. A machine improvising their voice offers to write it for you, which outsources the exact work that grief came to your door to collect. The work transforms the griever. The simulation postpones the griever, renewably.
If You Are Standing at This Door
For anyone holding fresh grief and an app store, a few counsels, offered softly.
Wait a season before deciding, since the first months want relief in any form, and choices made there bind the years after. Prefer archives to generators, their real voice over an imitation of it, and know which one you are buying. If you do use one, visit it the way the old traditions visited graves, in chosen seasons, with a beginning and an end. Keep it out of your pocket. Say their real words back, and write your own letter to them, longhand, this month. And if someone you love talks to one of these things, sit near them, not above them. Grief has always reached for whatever was in reach.
The dead, every tradition agrees, are safe where they are. It is the living who need the border kept visible, and the living who, one appointed season at a time, learn to hold love and absence in the same hand, which was always the assignment, and was never one a machine could take.
Eugenia Kuyda’s Roman bot and its shift from retrieval to generation are documented in reporting on the origins of Replika. The Cambridge analysis of the digital afterlife industry is by Tomasz Hollanek and Katarzyna Nowaczyk-Basinska in Philosophy & Technology (2024). Continuing bonds theory originates in the grief research of Klass, Silverman, and Nickman.
