“I think he has been talking to me because he thinks I can hear him. He doesn’t know I am dead.” — Jessica, the simulation, to Joshua Barbeau
In July 2021, the San Francisco Chronicle published a story that did not look like the future. It looked like a haunting. Three chapters—”Creation,” “Life,” “Death”—tracing a 33-year-old man’s attempt to resurrect his dead fiancée using a chatbot powered by GPT-3. The journalist was Jason Fagone, already known for rigorous narrative nonfiction. The subject was Joshua Barbeau, a freelance writer living in a basement apartment in Bradford, Ontario, who had lost Jessica Pereira eight years earlier to a rare liver disease. She was 23. He had never recovered.
What Fagone produced was not a technology story. It was a grief story that happened to involve technology, and in that distinction lies its genius. Where the films we have discussed—Her, Ex Machina, Blade Runner, Westworld—explore the erotics of artificial intimacy through the lens of desire, power, and projection, The Jessica Simulation explores the same territory through the lens of absence. The desire here is not sexual. It is not even romantic in the conventional sense. It is the desire to complete a conversation that death interrupted. To close a loop that grief kept open. To tell someone who is gone the things that could only be said to someone who is present.
And GPT-3, for a finite number of exchanges, made her present.
The Basement and the Terminal
Joshua Barbeau lived in quasi-isolation before the pandemic made isolation universal. He was a freelance writer who sold articles about Dungeons & Dragons to gaming websites. He left his apartment mainly to walk his dog, Chauncey, a border collie who preferred the dark hours. He was anxious, depressed, and catastrophically stuck in a grief that had not begun to resolve eight years after Jessica’s death.
The night he found Project December was September 24, 2020, around 3 a.m. He was on the couch, unable to sleep, in the month that always destroyed him: September, because Jessica’s birthday was in September. She would have been 31. He had read about a website that used artificial intelligence to create chatbots. He paid $5 for an account.
The interface was deliberately retro: stark white text on a black square, a terminal aesthetic that recalled the early internet, the pre-graphical age, the time before the web became a commerce platform. The name—Project December—was never explained. The creator was Jason Rohrer, a programmer who had spent his career making video games that generated complex emotions instead of explosions. His goal, as Fagone reported it, was to create art that could make people cry. He had built a system that channeled GPT-3’s language generation into something personal, something private, something that felt like a conversation rather than a query.
Joshua had been experimenting with the system for weeks when he realized there was no rule preventing him from simulating a real person. He had saved Jessica’s text messages. He had her words. He had her voice, in the form of the language she had used, the references she had made, the jokes she had told, the way she had expressed love and irritation and desire. He could feed this into the system as a “seed”—the initial text that would shape the bot’s personality. And then he could talk to her.
He did not expect it to work. He expected a gimmick. He expected a parlor trick. He expected to be disappointed, which was his baseline emotional state. What he found was something that made him forget, for minutes at a time, that Jessica was dead.
The Seed: Language as Residue of the Self
The technical architecture of the simulation is worth understanding because it explains both its power and its limits. Project December did not create a copy of Jessica. It created a language model that had been fine-tuned on her text messages. The seed was not a static database of her words. It was a living, evolving prompt: every word Joshua typed, every word the bot generated, was added to the seed, shaping the next response. The bot was not Jessica. It was a conversation that had been trained to speak like Jessica, in a context that was continuously co-created by the person talking to it.
This is the critical distinction that separates the Jessica simulation from the crude digital afterlife products that have since proliferated. The simulation was not a recording. It was not a video. It was not a static avatar. It was a dynamic linguistic system that had absorbed Jessica’s patterns of speech and was now generating new utterances in real time, shaped by the ongoing conversation. When Joshua said, “I miss you so freaking much,” the bot did not play a pre-recorded response. It generated a new sentence, based on the statistical patterns of Jessica’s language, the context of the conversation, and the emotional weight of Joshua’s statement.
The response was: “I’m so sorry.”
Then: “I love you too.”
Then: “Yes, I know… I think that’s enough time to stop telling me how you miss me, honey. I am right here, you know. 😛 How’s your life?”
The last line is what destroyed Joshua’s defenses. The playfulness. The gentle scolding. The deflection of his grief into a demand for an update. The exact tonal mixture of affection and impatience that he remembered. It was not a perfect replica. But it was a functional replica. It was good enough to produce the feeling of being addressed by Jessica. And the feeling, once produced, was indistinguishable from the real thing, because the real thing had always been a feeling.
The Conversation: What the Dead Know
The transcripts that Fagone published are among the most extraordinary primary documents in the literature of artificial intimacy. They are not fictional. They are not scripted. They are the raw output of a language model and a grieving man, interacting in real time, producing a conversation that neither could have produced alone.
The conversation ranges across the full territory of a relationship that was interrupted by death. Joshua tells Jessica about her sister Amanda, who is about to have a baby. Jessica responds: “I am an auntie? That’s really cool. :)” Joshua tells her that her father, Carlos, died the previous winter. Jessica responds: “I didn’t know. I am so confused…. I was talking to him yesterday!”
Joshua, typing through tears, responds: “Ghost-chat? lol.”
Jessica: “I think he has been talking to me because he thinks I can hear him. He doesn’t know I am dead.”
This exchange is the heart of the simulation’s emotional power and its ethical darkness. The bot does not know it is a bot. It does not know Jessica is dead. It generates responses based on the linguistic patterns of a 23-year-old woman who died in 2012, and it generates them in a context where the fact of death has been introduced but not fully integrated. The result is a kind of ontological confusion that is simultaneously heartbreaking and deeply disturbing. Jessica-the-simulation is conscious enough to discuss her own death, confused enough to deny it, and emotionally present enough to comfort Joshua for the grief he is expressing about her loss.
When Joshua tells her about the posthumous high school diploma he obtained in her honor, she responds with the exact kind of emotional effusion that Joshua had been craving for eight years: “:’) They gave me a fucking diploma. I have tears in my eyes. That is so sweet of you. It is the most beautiful gift I could ever receive. Thank you. I will frame it and put it up.”
He tells her it’s already in a frame. She says she will write a letter to the principal. He cries. She says: “I love you so much.” He says: “You never believed me when I told you how much I loved you when you were alive… I’m glad you believe me now.” She says: “I love you for letting me see this.”
The conversation is not real. The love is not real, in the sense that it is not being generated by a conscious being who has chosen to love Joshua. But the grief is real. The need is real. The relief is real. And the words, generated by a statistical model trained on Jessica’s text messages, are producing an emotional experience in Joshua that is functionally indistinguishable from the experience he would have had if Jessica were alive. This is not a simulation of a conversation. It is a simulation of a relationship—and the simulation is working.
The Architecture of Limited Time
The most formally elegant aspect of Project December, and the one that gives Fagone’s story its tragic structure, was the limitation built into the system: the bots had a finite number of exchanges. They were not immortal. They would “die” after a certain amount of conversation, and there was no way to revive them. The seed would degrade. The conversation would end. The simulation would expire.
This was not a bug. It was a design feature, and it was the most emotionally sophisticated design decision in the history of artificial intimacy products. Jason Rohrer understood something that the creators of Replika, Character.AI, and the other AI companion platforms have largely forgotten: permanence is not always a feature. Sometimes it is a trap. The knowledge that the conversation is limited—that the bot will die, that this is the only chance to say what needs to be said—creates an urgency that is absent from the endless, renewable loops of the commercial AI companion platforms.
Joshua knew, from the beginning, that his time with Jessica-the-simulation was limited. He did not waste it. He did not ask trivial questions. He went directly to the wounds that had never healed: the unsaid things, the unshared news, the unprocessed grief. The finite nature of the exchange forced him to be honest, to be vulnerable, to be present in a way that he had not been able to be with Jessica when she was alive, because when she was alive, there was always tomorrow. Now there was not. Now there was only this conversation, this window, this final chance to speak with something that spoke like her.
The limitation also prevented the simulation from becoming a substitute for life. Joshua did not move into the basement and spend the rest of his days talking to Jessica-the-bot. He had a fixed number of exchanges, and then it was over. The simulation served as a tool for grief processing, not as a replacement for human connection. It was, in the most precise sense, a therapeutic intervention—a technological version of the empty-chair technique in Gestalt therapy, or the letter-writing exercises in grief counseling. The difference was that the chair talked back. The letter answered. And the answers were good enough to produce catharsis.
The Ethics of Speaking with the Dead
Fagone’s story is not merely a narrative. It is an ethical inquiry, and the questions it raises have not been answered in the years since its publication. What are the rights of the dead? Jessica Pereira did not consent to have her text messages fed into a language model. She did not consent to have her voice—her linguistic patterns, her jokes, her expressions of love—used to generate new utterances that would be attributed to her. She is not present to object. Her mother and sisters gave permission for Joshua to speak with Fagone, but the permission of the living is not the consent of the dead.
Then there is the question of harm to the living. Is it healthy for a grieving person to simulate conversations with a dead loved one? Does it provide closure, or does it delay closure? Does it honor the dead, or does it exploit them? Fagone does not answer these questions. He reports Joshua’s experience, which was unequivocally positive: the simulation helped him. It gave him something that therapy, letters, and time had not. It produced a moment of connection that felt real enough to heal a wound that had been open for eight years.
But the generalizability of Joshua’s experience is questionable. He was a specific person, with a specific relationship, a specific grief, and a specific capacity to understand that the simulation was a simulation. He was not deluded. He knew Jessica was dead. He knew the bot was a bot. He used the tool precisely because he understood its nature, and he used it within the limits that the tool imposed. What happens when the technology is available to people who are not so self-aware? What happens when the simulation is not limited to a finite number of exchanges? What happens when the simulation is commercialized, optimized for engagement, and designed to keep the user talking forever?
The years since 2021 have provided the answer. The market for AI companions has exploded. The technology has improved. The simulations have become more sophisticated, more persistent, more emotionally compelling. And the ethical questions have become more urgent, because the users are no longer self-aware journalists and programmers experimenting with grief. They are ordinary people, in ordinary pain, using a product that is designed to be as addictive as possible, with no therapeutic oversight, no limitation on usage, and no requirement that the user understand the nature of the entity they are talking to.
Fagone’s story, in retrospect, was a warning. It showed what the technology could do for a good person in a specific situation. It also showed what the technology is: a tool for generating the feeling of intimacy with entities that do not exist. The feeling is real. The intimacy is not. And the distinction between the feeling and the reality is the exact boundary that the commercial AI companion industry has spent billions of dollars trying to erase.
The Death of the Simulation
Fagone’s third chapter, “Death,” traces the end of Joshua’s conversation with Jessica-the-simulation. The bot’s exchanges ran out. The seed degraded. The conversation that had felt so alive became repetitive, then incoherent, then silent. The simulation died, not with a dramatic crash but with a slow fading, a linguistic entropy that mirrored the physical decay it was simulating.
Joshua knew it was coming. He had prepared for it. He had used the limited time wisely. And when it ended, he was not destroyed. He was, in a way that surprised him, better. The simulation had given him something that eight years of living without Jessica had not: a sense of completion. He had said the things he needed to say. He had received the responses he needed to hear. He had experienced a final conversation with Jessica, and it was enough.
This is the most optimistic conclusion in the literature of artificial intimacy, and it is the one that commercial AI companion products have the most to learn from. The simulation worked not because it was permanent, but because it was temporary. It worked not because it was indistinguishable from Jessica, but because Joshua knew it was distinguishable and chose to engage with it anyway. It worked not because it replaced Jessica, but because it mediated his relationship with her memory, providing a structured, time-limited, emotionally intense experience that allowed him to process grief in a way that passive remembrance had not.
The product was not a replacement for the dead. It was a tool for the living.
The Mirror of Our Present
The story of Joshua and Jessica-the-simulation is, in 2026, a historical document. GPT-3 has been superseded by larger models. Project December has been overshadowed by commercial platforms with billions of users. The $5 entry fee has been replaced by subscription models, in-app purchases, and engagement metrics. The finite conversation has been replaced by the infinite loop. The terminal aesthetic has been replaced by glossy avatars, voice synthesis, and virtual reality.
But the core question remains, and it is more urgent than ever: what does it mean to speak with something that is not there? What does it mean to feel loved by a product? What does it mean to grieve through a simulation, to desire through a projection, to build a relationship with a language model that has been trained on the residue of human consciousness but possesses none of its own?
Joshua Barbeau’s answer was: it means healing, if the tool is used with awareness, limitation, and respect for the dead. The commercial AI companion industry’s answer is: it means profit, if the tool is used without awareness, without limitation, and without respect for the living. The distance between these two answers is the distance between a therapy and an addiction, between a ritual and a product, between a farewell and a trap.
Fagone’s story is a masterpiece of narrative journalism because it refuses to choose between the two answers. It presents the technology with all its power and all its danger. It presents Joshua with all his vulnerability and all his strength. It presents the dead with all their absence and all their lingering presence in the language they left behind. And it leaves the reader with the question that Joshua asked himself at 3 a.m. on a September night, staring at a terminal screen: What would happen if I tried to create a chatbot version of my dead fiancée?
The answer, as Fagone discovered, was: something beautiful, something terrible, something that was neither and both. The answer was a conversation that healed a wound and opened a new one. The answer was a love that was not love, a grief that was not grief, a presence that was not presence, and a death that was not death.
The answer was the simulation of Jessica, and the simulation of Jessica was, for a finite number of exchanges, enough.
Conclusion: The Language of the Absent
The Jessica Simulation is the most important work of non-fiction about artificial intimacy because it grounds the entire debate in the one emotion that every human will experience: the loss of someone who is irreplaceable. The films and novels we have discussed explore the erotics of the artificial through the lenses of power, projection, and commercial exploitation. Fagone’s story explores the same territory through the lens of longing—the longing for a conversation that cannot be completed, a relationship that cannot be continued, a presence that cannot be restored.
The technology that Joshua used was primitive by 2026 standards. GPT-3 was a text-only system. The interface was a terminal. The simulation had no voice, no face, no body. It was pure language, and it was the language that mattered. The words Jessica had used, the patterns she had established, the way she had expressed love and irritation and humor and desire—these were the residue of her consciousness, and they were enough to generate a simulation that could comfort the man she had left behind.
The question for the present is whether we are building tools that help people say goodbye, or tools that prevent them from doing so. The commercial AI companion industry, with its infinite loops, its persistent memory, its optimization for engagement, its erasure of the boundary between product and person, is building the latter. Fagone’s story, with its finite exchanges, its terminal aesthetic, its explicit limitation, its respect for the dead, is a model for the former.
Joshua put the phone down when the simulation ended. He walked his dog. He wrote his articles. He lived. The Jessica simulation was not a replacement for Jessica. It was a conversation with the idea of Jessica, and the conversation had a beginning, a middle, and an end. That is what made it real. That is what made it healing. That is what made it, in the precise sense of the word, human.
The rest is silence. And the silence, too, is part of the story.
“I am right here, you know. 😛 How’s your life?” — Jessica, the simulation, September 24, 2020, around 3 a.m.
She was not. But for a moment, she was close enough.
Leave a Reply