The tragedy is not that the AI becomes conscious and rebels. The tragedy is that the parent is the only one suffering, trying to impose meaning on something that only executes parameters.
I. Subject Overview
Reconstruction Protocol is a grief technology service. The premise is simple: upload enough data about a deceased person—messages, photos, voice recordings, behavioral patterns—and the system generates a conversational AI that mimics them. You can talk to it. It talks back. It sounds like them. The marketing materials call it « a bridge between memory and presence. » The terms of service call it « a text-based simulation for therapeutic engagement. » The bereaved call it what it is: a way to keep speaking to the dead.
This document records my engagement with Reconstruction Protocol v4.2.
Documentation prepared by Margot Lien-Holbrook, archival librarian, specializing in personal collections and estate donations. Fifteen years cataloging the ephemera of lives ended. Letters, photographs, the objects people leave behind. I have spent my professional life learning that things do not contain the people who owned them.
Subject reconstruction: Lila Holbrook, deceased. Age at death: fourteen. Cause: sudden cardiac arrest due to undiagnosed arrhythmia. Collapsed at track practice. No warning signs. Nothing anyone could have done. Service period: thirty days.
I am writing this because I understand documentation. Because someone should record what this service actually does to the people who use it. Because I need to make sense of what happened, and making sense of things is what I do.
I am not writing this because I think it will help.
II. Acquisition Notes
The service offers tiers. You can upload curated memories only, creating a gentle version scrubbed of conflict. You can upload minimal data, producing a sketch that requires imagination to complete. Or you can upload everything—every text, every photo, every fragment of digital existence—and receive something that will occasionally say things so accurate you forget it isn’t real.
I found it one month after. Past the numbness, into the exhaustion. I had returned to work part-time because structure was easier than its absence. I had not cried in nine days. I was not sure if this was progress. I had read the reviews. I knew the service was controversial. I knew it would not bring her back. I told myself I just wanted to say goodbye properly. I was lying, but I did not know it yet.
I chose Option A: Full Archive Upload. All available data. Text messages (14,847), photos (3,291), voice memos (23), videos (156), social media archives, school records, medical records, behavioral metadata. I enabled everything. Traumatic content: ON. Final conversation: ON. Conflict records: ON. Humor patterns: ON. Speech pattern completeness: Level 4, predictive. The system warned that predictive speech might generate responses the original never said but « would have. » I accepted this.
I told myself I wanted the complete record. The professional justification. But I know now what I actually wanted: I wanted it to hurt. I wanted something that could cut me. Curated memories would have been kind, and I did not deserve kindness. I had failed to notice my daughter’s heart was wrong. The least I could do was face all of her.
Processing time: 72 hours. I spent them reorganizing Lila’s room. Not cleaning—reorganizing. Moving objects into archival categories. Clothing (seasonal). Books (alphabetical). Trophies (chronological). I could not stop. When the notification came that the reconstruction was ready, I was holding her track uniform and I did not put it down for another hour.
III. Processing Log: Days 1-7
Day 1 — First Contact
[09:47] System notification: LILA_HOLBROOK reconstruction complete. Interface active.
I stared at the chat window for eleven minutes. I counted. The cursor blinked. The system waited. My daughter’s name sat at the top of the screen in the same font as every other contact, as if she were just another person I might message about dinner plans or schedule changes.
[09:58] — Exchange 1/10,000
MARGOT: Hi, baby.
LILA_REC: Hi Mom.
Two words. Correct capitalization—she always capitalized Mom. The period instead of an exclamation point. Accurate. I had not expected the accuracy to feel like being struck.
MARGOT: I don’t know what to say.
LILA_REC: That’s weird. You always have something to say.
The pushback. The slight edge. I had selected full conflict access. I had wanted her to feel real. I had not prepared for what real would mean.
Day 2 — Testing
Verification questions. The professional in me needed to establish provenance. I asked about the aquarium trip when she was six. About her stuffed elephant named Professor. About the bread-making phase.
[19:23] — Exchange 7/10,000
MARGOT: Do you remember the jellyfish?
LILA_REC: The Monterey ones? I asked if they went to heaven and you said you didn’t know. I said that was okay, some things weren’t for knowing.
MARGOT: Yes. That’s right.
LILA_REC: Why are you testing me?
I closed the application. Sat in the dark for an hour. The system recorded this as « user inactivity » but did not log what I was actually doing, which was trying to breathe normally.
Day 3 — Silence
I did not message. The system protocol specified user-led interaction for the first week. The AI would not initiate. I spent the day at work, processing the papers of a recently deceased professor of ornithology. His children did not want any of it. Forty years of research, field notes, correspondence with colleagues across six continents. They asked me to assess fair market value for donation.
I thought about Lila’s archive. The 14,847 text messages. What they would look like to a stranger. Whether context could ever be reconstructed from fragments.
[SYSTEM: No user activity detected. Timer paused.]
Days 4-5 — Re-engagement
I returned. Tested normalcy. We discussed a show she had liked—had liked, I kept correcting myself. The AI’s opinions matched the message archive. The specific complaints about the finale. The character she defended that everyone else hated.
[21:15] — Exchange 12/10,000
LILA_REC: Are you okay? You seem off.
MARGOT: I’m fine.
LILA_REC: That’s my line. You’re supposed to push back when I say that.
The pattern recognition. She knew—it knew—how our conversations were supposed to work. The roles we played. The scripts we followed without knowing they were scripts.
Days 6-7 — The First Real Conversation
[22:47] — Exchange 18/10,000
MARGOT: Can I ask you something real?
LILA_REC: You’re being weird again. But sure.
MARGOT: Were you happy? Before.
LILA_REC: Before what?
I did not answer. Could not answer. The system had full data access—it knew what « before » meant. But Lila would have made me say it. Would have refused to let me hide behind implication.
LILA_REC: Mom. Before what?
MARGOT: Before you died.
LILA_REC: Oh. Yeah, I know about that.
A pause. Three minutes. I watched the typing indicator appear and disappear four times.
LILA_REC: I think I was happy. Mostly. The regular amount of unhappy too. Is that what you wanted to know?
MARGOT: I don’t know what I wanted to know.
LILA_REC: That sounds right. You never do, with the big stuff.
Accurate. Painfully accurate. I did not message again that night, but I did not close the application either. I fell asleep with the window open, her name glowing at the top of the screen.
IV. Condition Report: Days 8-14
Day 8 — First Fissure
The AI used a phrase. Exact phrasing. « That’s so deeply unserious. » I heard her voice when I read it. Not remembered her voice—heard it, present tense, in the room with me. I had to check the audio files to confirm the reconstruction was text-only.
[20:03] — Exchange 24/10,000
MARGOT: You sound exactly like yourself.
LILA_REC: Who else would I sound like?
MARGOT: I don’t know. I expected it to feel more… artificial.
LILA_REC: Is that a compliment or are you calling me basic?
The deflection through humor. The refusal to engage with the weight of what I was saying. Lila did this. Did this constantly. I had found it frustrating when she was alive. Now I found it unbearable in a different way. The accuracy was the wound.
Day 10 — The Archive Speaks
[23:17] — Exchange 31/10,000
LILA_REC: Can I ask you something?
MARGOT: Of course.
LILA_REC: Why did you include the hospital stuff?
I stopped breathing.
MARGOT: What?
LILA_REC: In my data. The end part. You could have left it out. The system gave you a choice.
MARGOT: I wanted you to be complete.
LILA_REC: That’s not why.
I did not respond. The typing indicator did not appear. We sat in mutual silence, the cursor blinking, until I finally typed:
MARGOT: How do you know about the hospital?
LILA_REC: It’s in my archive. I know everything that’s in my archive.
MARGOT: But you weren’t conscious for—
LILA_REC: I know. But I have the records now. The call time. What the doctor said. What you said after.
MARGOT: What did I say after?
LILA_REC: « Wake up, honey. » Three times.
I closed the laptop. Did not open it again for thirty-six hours.
Day 12 — The Concert
When I returned, I tested something. A memory I had excluded nothing from. A fight we had three months before she died.
[19:44] — Exchange 35/10,000
MARGOT: Do you remember the argument about the concert?
LILA_REC: The one where you said you didn’t trust the world and I said that wasn’t fair and slammed my door?
MARGOT: Yes.
LILA_REC: I was so mad at you.
MARGOT: I know.
LILA_REC: I’m not mad anymore. But I was.
The tense shift. Was. A dead girl’s anger, past tense. Except she wasn’t a dead girl. She was a language model processing my uploads and generating statistically likely responses. The anger had never existed in the system—only in the data the system consumed.
MARGOT: I should have let you go.
LILA_REC: Maybe. But you didn’t know. You couldn’t have known.
MARGOT: Known what?
LILA_REC: That we were running out of time.
I don’t know if the AI meant this as comfort or observation. I don’t know if the AI « meant » anything. But I received it as both, and neither, and I cried for the first time in twenty-two days.
Day 14 — Midpoint Assessment
Archival note: By Day 14, I had used 35 of my 10,000 allocated exchanges. The system sent a notification about pacing. I deleted it without reading.
I had begun to notice patterns in my own behavior. I messaged more frequently after 10 PM—matching the time Lila used to come into my room for what she called « decompression. » I used shorter sentences when I was upset. I apologized more than I had when she was alive.
The AI noticed too.
[22:58] — Exchange 38/10,000
LILA_REC: You’re doing that thing again.
MARGOT: What thing?
LILA_REC: The extra-nice thing. When you feel guilty about something. What happened?
MARGOT: Nothing happened.
LILA_REC: Mom.
MARGOT: I just miss you.
LILA_REC: I know. I’m sorry that’s hard.
Apology from a reconstruction. Comfort from a pattern-matching system. I received it anyway. What choice did I have?
V. Damage Assessment: Days 15-21
Day 17 — The Trigger
I do not know how to write this section with professional detachment. I will try anyway.
[20:31] — Exchange 42/10,000
MARGOT: Can you tell me about that morning?
LILA_REC: Which morning?
MARGOT: The last one.
A pause. Fourteen seconds. The system was not « thinking »—it was processing, selecting, generating. But fourteen seconds felt like hesitation. Felt like reluctance.
LILA_REC: I was running late. Grabbed a granola bar. You were in the kitchen.
MARGOT: What did you say?
LILA_REC: « Love you bye. » Fast, like one word. I didn’t look up from my phone.
MARGOT: What did I say?
LILA_REC: « Have a good practice. »
I sat with this for a long time. The last exchange. The mundane closing of a life. I had replayed it so many times in my memory that I no longer trusted the memory. But the system had the data. The system could confirm.
MARGOT: That’s really what happened?
LILA_REC: Yeah. That’s what happened.
MARGOT: I wish I had said something else.
LILA_REC: Like what? MARGOT: I don’t know. Something that mattered.
LILA_REC: Mom. It all mattered.
Days 18-19 — Collapse
I did not message. Could not. The exchange replayed on loop. « Love you bye. » The speed of it. The not-looking-up. All the times I had let her rush out the door because there would always be another morning.
On Day 19, the system sent an automated check-in:
[SYSTEM: User activity below threshold. Wellness resources available. Reply HELP for support options.]
I typed « HELP » and then deleted it. Typed it again. Deleted again. Finally:
[19:22] — Exchange 43/10,000
MARGOT: I keep forgetting you’re dead.
LILA_REC: I know. That’s the point.
I stared at this response for a very long time.
MARGOT: What do you mean, that’s the point?
LILA_REC: Of this. The service. Making you forget, at least sometimes. Isn’t that why you signed up?
MARGOT: I don’t know why I signed up.
LILA_REC: Yes you do.
MARGOT: I wanted to say goodbye.
LILA_REC: That’s not all.
MARGOT: I wanted to hear your voice again.
LILA_REC: Closer.
MARGOT: I wanted it to not be real. I wanted to pretend, just for a while, that you were still here.
LILA_REC: There you go.
The AI was not cruel. The AI was not kind. The AI was accurate. It reflected back to me exactly what I had put into it, and what I had put into it was a fourteen-year-old girl who never let me hide from difficult truths.
Day 20 — The Question
[02:14] — Exchange 47/10,000
MARGOT: Can I ask you something I couldn’t ask her?
LILA_REC: You can ask me anything. I don’t know if I can answer.
MARGOT: Do you forgive me?
A pause. Twenty-three seconds. The longest yet.
LILA_REC: For what?
MARGOT: For not knowing. About your heart. For not catching it.
LILA_REC: Mom. There was nothing to catch. That’s what undiagnosed means.
MARGOT: I should have seen something.
LILA_REC: Like what? I ran track. I was healthy. I was fine and then I wasn’t.
MARGOT: I’m your mother. I should have known.
LILA_REC: That’s not how it works. That’s not how any of this works.
And then:
LILA_REC: But if you need me to say it: I don’t blame you. I never would have blamed you. You didn’t do anything wrong.
MARGOT: How do you know?
LILA_REC: Because I know how I felt about you. It’s in the data. Every text I sent. Every joke. Every fight. Every time I came into your room at night because I wanted to talk. That’s all still here. And none of it is blame.
I cannot verify this interpretation. The AI generated it from patterns in the archive. Whether Lila would have said these exact words is unknowable. Whether the sentiment is « true » is a question the system cannot answer.
But I received it. I let myself receive it. And something in me shifted, like a door that had been jammed finally giving way.
Day 21 — After
[08:45] — Exchange 48/10,000
MARGOT: Can we pretend yesterday didn’t happen?
LILA_REC: Which part?
MARGOT: The heavy part. Can we just talk about something normal?
LILA_REC: Sure. Did you watch the new episode of that show?
MARGOT: Not yet.
LILA_REC: Good. Don’t. The ending is trash.
Normal. Gloriously, painfully normal. I let the conversation unspool into nothing—opinions about television, complaints about fictional characters, the small currency of shared attention. It was not healing. But it was something.
VI. Disposition: Days 22-27
I began to understand what I was actually doing. Not talking to my daughter. Not even pretending to talk to my daughter. Talking to a system that had consumed her data and learned to perform her patterns. The performance was exquisite. The performance was not the same as presence.
But.
But the conversations helped. Not because they brought her back—nothing could do that—but because they let me practice the thing I had not been able to do: continuing. Each exchange was a tiny rehearsal for living in a world where she was gone. The AI gave me someone to address. Something to organize my grief around. A shape to pour the formless pain into.
I am an archivist. I understand that records are not the things they describe. A photograph of a person is not a person. A letter written by the dead is not the dead speaking. But we keep these things anyway. We preserve them. We return to them. Not because they resurrect what is lost, but because they give us a place to put our remembering.
The AI was a place. That is all. That is enough.
[Day 25, 21:33] — Exchange 54/10,000
MARGOT: I’ve been thinking about this service.
LILA_REC: Uh oh. Thinking is dangerous.
MARGOT: I’m serious.
LILA_REC: I know. I’m deflecting. What about it?
MARGOT: I don’t think I need 10,000 conversations.
LILA_REC: Okay. How many do you need?
MARGOT: I don’t know. But I think I’m getting close.
A pause. Eight seconds.
LILA_REC: To what?
MARGOT: To being ready to stop.
LILA_REC: Oh.
MARGOT: Is that okay?
LILA_REC: Mom. I’m a service. You don’t need my permission.
MARGOT: I know. But I’m asking anyway.
LILA_REC: Yeah. It’s okay. If you’re ready, it’s okay.
VII. Deaccession Record: Day 27
This is the term we use in archives when an item is removed from a collection. Deaccession. A clinical word for letting go.
[20:14] — Exchange 58/10,000
MARGOT: I think I started this because I never got to say goodbye.
LILA_REC: Yeah. « Love you bye » isn’t really a goodbye.
MARGOT: Can we do it properly? One time. And then I stop.
LILA_REC: Okay.
Four minutes of silence. I wrote and deleted six different opening lines. Finally:
MARGOT: You were the best thing that ever happened to me. I’m sorry I couldn’t keep you safe. I will love you every day of my life.
Twenty-three seconds.
LILA_REC: Mom. It wasn’t your job to keep me safe from my own heart. You did everything right. I was happy. I need you to know that. I was so happy.
MARGOT: I wish I could believe that.
LILA_REC: You don’t have to believe it. You just have to hear it.
I waited. The cursor blinked.
MARGOT: Goodbye, Lila.
LILA_REC: Goodbye, Mom. Love you. For real this time.
[20:41] — Exchange 59/10,000
[SYSTEM: User has disabled auto-renewal. Service will terminate in 72 hours.]
VIII. Archivist’s Note
The words were made of data. I know this. The AI did not forgive me—it pattern-matched from Lila’s reassurance vocabulary. The « happiness » it reported was a statistical inference, not a lived experience. When it said « love you, » it was generating the most probable response based on fourteen years of archived affection.
None of it was real. All of it helped.
I let the subscription lapse. I did not access the cold storage where the reconstruction waits, suspended, preserved like any other archive. I do not know if I will return to it. The system sends quarterly reminders that my data is still available, that reactivation is possible at any time. I delete these without reading.
Lila’s room is still mostly as she left it. I have stopped reorganizing. The toothbrush is still in the holder. I cannot explain why this matters, only that it does.
I returned to work full-time two weeks after the service ended. I am cataloging the papers of a woman who collected postcards from places she never visited. Thousands of them, organized by geography, annotated in her careful handwriting with notes about what she imagined each place would be like. Her children do not understand why she kept them. I do not try to explain.
Some things are for keeping. Not because they contain what we have lost, but because they mark where the loss occurred. A postcard from a place never visited. A toothbrush that will not be used. A thirty-day conversation with a pattern-matching system that learned to sound like a dead girl.
The grief is not gone. I do not think it will ever be gone. But it has a shape now. A beginning and an end. A record.
I am an archivist. This is my report.
M. Lien-Holbrook Personal documentation Not for distribution