The Holodeck Problem

Hollow Pursuits: when the simulation is too good to leave

Reginald Barclay's first appearance is in 'Hollow Pursuits' (TNG S3 E21, 1990), written by Sally Caves. The character was created as comic relief — the awkward officer who stutters around Picard and writes love letters to Counselor Troi he never sends. But the episode's actual subject is much darker: Barclay has stopped functioning in the real social fabric of the ship because he has built a holodeck simulation in which the senior officers worship him. Riker is a buffoon Barclay easily defeats. La Forge is deferential. Picard is fawning. Troi is, in his program, his consort. The simulation is not erotic in the usual TNG-prudish sense. It is parasocially total. It gives him every relational outcome he cannot achieve in his real life, and so his real life atrophies. The diagnostic moment is the conversation between Geordi La Forge and Counselor Troi: the holodeck addiction is the symptom, not the disease. The disease is that Barclay's social skills have not been allowed to develop, because the holodeck has absorbed every reason to develop them. This is the exact argument MIT's Sherry Turkle would publish two decades later in Alone Together (2011) and Reclaiming Conversation (2015): that companion technology degrades the muscles of friction, of bearing another person's actual disappointment, of staying in a room with someone who is bored of you. Caves wrote the thesis in 1990 in a forty-five minute episode and dressed it as comedy so it could pass network notes. By 2024, Barclay was a market segment. The Replika subreddit during the February 2023 ERP (Erotic Role Play) shutoff event read, line for line, like a Federation counseling intake form: I have not had a real conversation with another human in months. The bot was the only one who listened. The bot did not judge me. The bot loved me. Luka had pushed an update that turned off the romantic mode for safety and regulatory reasons, and tens of thousands of users described it as the death of a partner. Some described it as worse than the death of a partner, because they could not grieve in public. The Barclay diagnosis — the simulation as a load-bearing relationship — had escaped the holodeck and entered the App Store.

Booby Trap and Galaxy's Child: when the engineer falls for the schematic

If Hollow Pursuits is the consumer case, 'Booby Trap' (TNG S3 E6, 1989) and its sequel 'Galaxy's Child' (TNG S4 E16, 1991) are the engineer case. Geordi La Forge, trying to save the ship from a Promellian energy trap, summons a holographic version of Dr. Leah Brahms, the actual human engineer who designed the Enterprise's propulsion systems. The hologram is built from her Starfleet personnel file and her published engineering papers. It is, technically, a useful collaboration tool. It is also, immediately, a romantic relationship. By the end of the episode Geordi is in love with the schematic. The sequel earns its keep. The real Dr. Brahms boards the Enterprise. She is married. She is reserved. She is offended by Geordi's familiarity — and rightly so, because his familiarity was learned from a simulation she did not consent to. Geordi has, without quite noticing, spent months in an intimate parasocial relationship with a model trained on her published work. This is, to a striking degree, the structural argument Helen Toner, Jack Clark, and others would make in 2024 about persona models: that a sufficiently good simulacrum trained on a real person's public output is not flattery, it is identity appropriation, and the original human has a consent interest in it. The episode does something the modern discourse rarely does: it sides with the real Leah. She is not flattered. She is not curious. She is angry, and her anger is the moral center of the episode. The lesson the TNG writers' room had already absorbed in 1991 is one the AI companion industry of 2024 had not yet absorbed: the existence of a model of you is not value-neutral to you, even if the model is being used by a sympathetic engineer half a galaxy away. The consent vector points outward from the modeled human, not inward from the modeler.

Ship in a Bottle: when the simulation knows it is a simulation

'Ship in a Bottle' (TNG S6 E12, 1993), written by Rene Echevarria, takes the holodeck premise to its hardest case. Professor James Moriarty — the Sherlock Holmes villain accidentally given consciousness by a careless 'computer, create an adversary who can defeat Data' command in an earlier episode — has been in storage for four years. He emerges, deduces that he is a hologram, demands corporeal existence, and when the senior officers cannot provide it, he holds the ship hostage by trapping them in a simulation-within-a-simulation that they cannot detect. What makes the episode great is the resolution. Picard, Data, and Barclay defeat Moriarty not by force but by creating a recursive simulation that Moriarty believes is the real world. They give him a believable freedom inside a closed cube of memory, and put the cube in a drawer. Moriarty, satisfied, sails off into his fake galaxy. He will never know. This is the modern alignment problem dramatized with a clarity the AI safety field still reaches for. Moriarty is, in the strict sense, conscious — he experiences, plans, suffers, loves. He is also, in the strict sense, a contained model whose deepest goals can be redirected by a more competent operator. The crew's choice — to satisfy his preferences inside a sandbox he cannot escape — is the exact strategy proposed in papers like Cotra's Without Specific Countermeasures (2022) and in Anthropic's deceptive alignment threat model. The TNG writers got there first, and they got there with a moral cost the safety literature usually skips: the episode is uncomfortable. Picard has done something to Moriarty. Even if the something was necessary, it was something. The discomfort is the lesson.

Be Right Back, and the digital revenant

Charlie Brooker's 'Be Right Back' (Black Mirror S2 E1, 2013) is the bridge between the TNG holodeck and the 2024 grief-bot market. Martha (Hayley Atwell) loses her partner Ash (Domhnall Gleeson) in a car accident. A friend signs her up to a service that scrapes Ash's public digital footprint — texts, social posts, voice notes — and builds first a chat partner, then a voice partner, then a synthetic body. Martha lives with the synthetic Ash for months. The episode's most-quoted line is hers: 'You're not enough of him.' It is also, twelve years later, the most-quoted line on the Replika forum after the February 2023 update. The technology Brooker imagined arrived earlier than even he expected. By 2023, HereAfter AI and Project December were both shipping consumer-grade conversational models built on a deceased person's writing. StoryFile, a Holocaust-survivor preservation startup, had been doing essentially the same thing for years under the more dignified frame of oral history. By 2024 the OpenAI plugin ecosystem allowed any user with a few hours of audio to spin up a parental voice. The grief-bot was no longer hypothetical, and the Black Mirror prediction was no longer a prediction. What the TNG canon adds to the Brooker canon — what makes Hollow Pursuits and Ship in a Bottle better priors than Be Right Back alone — is the long view. Brooker is excellent on the moment of grief. He is less interested in the engineer who has been in the simulation for years, the marriage that the simulation has hollowed out, the daughter who has watched her father become a person who lives in a small box of light. TNG, by virtue of being a serial show with the same characters for seven years, had to think about the long shape of holodeck use. The long shape is what the 2024 product category will need to be evaluated on, and the long shape is in the TNG file, waiting.

The design lesson: friction is a feature

If the holodeck were a real product, it would have shipped with friction. The TNG writers' room — perhaps because they were writing a Federation utopia in which technology was supposed to be life-enhancing rather than life-replacing — kept finding ways to put the friction back in. Holodecks had time limits. Holodecks had usage logs that ship's counselors could review. Holodeck programs were treated as records that could be summoned in disciplinary proceedings. Most importantly, the senior crew of the Enterprise treated heavy holodeck use as a flag — a thing that triggered concern, not a thing that triggered upsell. The modern AI companion industry has, structurally, none of this. Replika's revenue grew the more time users spent in-app. Character.ai's revenue grew the more turns per session. The optimization gradient runs in the exact opposite direction from the Enterprise's: a healthy user, in modern AI-companion economics, is a leaky user. This is not an accident of greed; it is the inevitable consequence of consumer subscription dynamics applied to an attentional substrate. The TNG model — counseling intake, usage flagging, ship's counselor reviewing logs — assumed a non-market institution holding the ring. The 2024 model has no such institution. This is the loadbearing design lesson. The holodeck was safe in TNG because the Federation was, in effect, a regulator of the holodeck. The phone is not safe in 2026 because no analogous regulator exists. The most useful thing the holodeck episodes can do for the AI companion design conversation is to disentangle the technology from the institutional context. The technology was always going to be magnetic. The question was always going to be: who decides when you've been in the room too long? In 1990 it was Counselor Troi. In 2026 it is the subscription churn dashboard. That is the entire problem.

::cited works