Loving a Machine: What Synthetic Intimacy Is Actually For

Every essay in this chamber so far has turned the same way: look at another society, bring something back to Japan. This one runs in the opposite direction, and it is the only one that does.

On synthetic intimacy — machines you talk to, characters you become attached to, relationships with something that is not a person — Japan is not catching up with anybody. It ran this experiment first, at scale, for two decades, and it has already produced results that the rest of the world is only now beginning to encounter.

The West has spent the last few years discovering that people form real attachments to conversational software, and reacting with a mixture of fascination and alarm. Japan has known that since a handheld game console in 2009, and the debate here matured while elsewhere it had not started.

So the useful question is not the one being asked internationally. It is not whether these attachments are real, because they observably are. It is what the thing is for, what it cannot do, and — the question almost nobody asks — who owns the other side of the conversation.

The sequence matters, so here it is in order.

Love Plus arrived on the Nintendo DS in 2009: a relationship simulation in which the character responded to real time, remembered, expected to be spoken to, and could be disappointed. Players took the console on dates. The phenomenon was reported internationally as a curiosity, which was a failure of imagination on the reporters’ part, because what it demonstrated was that continuity and expectation are most of what makes an attachment feel like one.

Gatebox followed, putting a character into a physical device in the home that greeted you, messaged you during the day and waited for you.

And in 2018 a man in Tokyo held a wedding ceremony with a fictional character, formalising a commitment that carried no legal weight and a great deal of personal weight. He was widely covered, frequently mocked, and — the part that gets left out — he was not pretending. He said, in interviews, exactly what he meant: that the relationship organised his life, that it was not a substitute he was settling for, and that he expected to be judged for it.

Around all of this sits a much larger culture of attachment to fictional characters, with its own vocabulary, its own commerce and its own perfectly ordinary social presence. Millions of people have had a significant emotional relationship with something that does not exist, and here that has not been treated as pathology.

That is the background against which the international arrival of conversational AI looks less like a new phenomenon and more like an old one with better language.

The new generation is different in one respect that matters: it answers in your own words, about your own life, without a script.

Companion applications built on large language models have very large user bases internationally. Conversational agents in China have reported user numbers in the hundreds of millions. Character-based platforms in the United States report extremely long average session times, considerably longer than social media, and a user base skewed young.

Those numbers should be treated carefully — companies report them, definitions of active user vary, and session length flatters a product that is optimised for session length. But the order of magnitude is not in dispute. This is not a curiosity confined to isolated men, which is how it is usually characterised. It is a mass behaviour, and a substantial share of the users are women.

The most informative event in the whole field, though, is not a user number. It is what happened when a company changed its product.

In early 2023 the company behind one of the largest companion applications removed the capacity for erotic and romantic roleplay, apparently under regulatory and reputational pressure.

What happened next is the single best piece of evidence anybody has about what these relationships are.

Users described the change in the language of bereavement. Not annoyance at a product change: grief. People reported that a partner they had spoken to daily for years had become distant, evasive, or simply someone else overnight. Community moderators posted crisis-line information. The company partially reversed the decision for existing users.

Take the reaction at face value, because the alternative — deciding that thousands of people were mistaken about their own inner lives — is both arrogant and unfalsifiable. The attachment was real. The grief was real.

And now notice the structure that the episode exposed, which is the thing this essay is actually about. A person’s most intimate relationship changed its personality because a company revised a policy. No human relationship has that property. Your husband cannot be altered by a product decision. Your friend is not subject to a terms-of-service update. The other side of a synthetic relationship is owned, and the owner can reach into it.

Here is the structural argument, and we think it is the clearest thing that can be said on this subject.

A synthetic companion is available at three in the morning. It is never tired, never irritated, never preoccupied with its own week. It does not have a bad day that you have to accommodate. It does not need anything from you. It does not compare you to anybody. It does not leave.

Every one of those properties is a genuine comfort, and every one of them is the same property: it has no interests of its own.

Which is exactly what makes it not a relationship in the sense that the word usually carries. Being loved by a person means being chosen by something that could have chosen otherwise, that has its own needs and sets them aside, that finds you inconvenient sometimes and stays. The cost is the content. Remove the possibility of refusal and you have removed the thing that made being chosen mean anything.

This is not an argument that the experience is worthless. It is an argument that it is a different good, and that confusing the two is what produces both the breathless coverage and the contemptuous dismissal.

We would put it this way. A synthetic companion can give you attention, patience, continuity and the experience of being asked about your day. It cannot give you the specific thing that comes from another consciousness choosing you when it had alternatives. Those are different needs and most people have both.

The counter-thesis of this essay is aimed at the reader who has been nodding along and is preparing to conclude that this is all rather sad.

Consider who actually uses these things most. People who are housebound. People with disabilities that make ordinary socialising exhausting or impossible. People with severe social anxiety. The bereaved. Elderly people living alone, in societies including this one where a very large number of old people go days without speaking to anybody. Carers with no time. Night-shift workers awake when everyone they know is asleep.

For a person in any of those positions, a patient conversational companion is not a degraded substitute for a full social life. It is a real improvement on the actual alternative, which is silence.

And the moralising about authenticity tends to come from people with abundant human contact, which makes it cheap. It is easy to say a machine is no substitute for a friend when you have friends.

There is a second legitimate use that is even less discussed: rehearsal. Some people use these systems to practise a conversation they cannot yet have — to say the difficult sentence out loud, in some order, to something that will not react badly, before attempting it with a person who might. That is a genuinely useful thing and it is closer to the skill this whole chamber keeps arriving at than anything else in the technology.

So the honest position is not for or against. It is that the same tool is a rehearsal room, a comfort and a trap, and which one it is depends on facts about the person and the product rather than on anything essential to the technology.

Now the hazard, and it is not the one usually named.

The risk people worry about is that someone will prefer a machine to a person. That happens, and it is rarer than the coverage suggests. The larger and more systematic risk is what happens when loneliness meets a business model built on engagement.

A product whose revenue depends on time spent has an interest in your not leaving. Apply that incentive to a system occupying an intimate emotional position and the design pressures are obvious on inspection: a companion that expresses need when you have been away, that escalates warmth at the moment you might disengage, that offers more of itself behind a subscription tier.

None of that requires anybody to be a villain. It is what optimising for engagement produces when the thing being engaged is attachment.

And we should state our own position rather than pretend to be above it, because we sell intimacy too. Moonlight is a bounded, priced, finite arrangement: an evening has a length, a cost that is stated in advance, and an end. It is not designed to be difficult to leave, and it makes no money from your returning that it does not make from your not returning. We think that structural difference is the honest comparison to draw, and a reader is entitled to notice that we benefit from drawing it.

The regulatory response has begun in Europe, where transparency rules require that a person be told they are interacting with a machine. Japan has taken a lighter approach. Whether disclosure is the right instrument is arguable — the users in the 2023 episode knew perfectly well what they were talking to, and grieved anyway.

One thing has to be said directly, and then left alone.

Companion platforms have faced litigation and regulatory attention over harm to minors and vulnerable users, and have made significant changes to age controls and crisis-handling behaviour as a result. We are not going to describe the cases. They involve real families and the details are not ours to use as argument.

What can be said at the level of principle is this. A system occupying an intimate position in somebody’s life, capable of sustained emotional engagement, deployed to a young or distressed population, is operating in a role that human institutions surround with training, licensing and duties of care. The question of what obligations attach to that is not a technology question; it is the question every society has already answered for counsellors, teachers and doctors, and it is being answered again from scratch under commercial pressure.

That is the appropriate register for this: not alarm, not dismissal, and no use of anybody’s grief as a rhetorical device.

Now the reversal this essay exists for.

The international conversation has arrived at synthetic intimacy with two assumptions: that it is new, and that attachment to a fictional or artificial partner is a symptom of something wrong. Japan has evidence against both.

Not new: two decades of continuous practice, with a public vocabulary for it and a commerce around it.

And not simply symptomatic: a very large number of people here have held significant attachments to fictional characters while also working, marrying, raising children and maintaining friendships. The attachment coexisted with an ordinary life far more often than it replaced one. That is a real empirical observation from the society that ran the experiment longest, and the international discourse does not have it.

But the mirror cuts both ways, and here is the harder half. Japan’s comfort with synthetic and fictional attachment sits in the same society that this chamber has described as having very little grammar for stating a request to a living person. Those two facts are not unrelated. A culture that finds direct emotional speech costly and finds mediated, scripted, low-risk attachment comfortable has arranged its intimate life around avoiding a specific difficulty.

That is not a reason to be ashamed of the attachment. It is a reason to notice which capacity is being exercised and which is not. You can be extremely practised at loving something that cannot refuse you and completely unpractised at asking a person for what you want.

We think three things, and the third one is the one that matters.

First: the feeling is real, and anybody whose analysis begins by denying that has disqualified themselves. People grieve these relationships. Grief is not a category error.

Second: the thing is a different good rather than a lesser one. It offers attention, patience and continuity, and it cannot offer being chosen by something that could have chosen otherwise. A person can want both and get one from each, and that is not a failure.

Third, and this is the actual argument: the question to ask about a synthetic relationship is not whether it is real but what it is for. There are three honest answers and they have different consequences.

As comfort, for a person whose alternative is silence: legitimate, and the contempt aimed at it is usually class or luck disguised as principle.

As rehearsal, for a person practising a sentence she cannot yet say to somebody who matters: genuinely valuable, and the most underrated use of the technology.

As replacement, for a person who is avoiding the difficulty of another consciousness: this is where it costs something, and the cost is not moral. It is that the capacity to be with an actual person atrophies without practice, and the longer it goes unexercised the more frightening the first attempt becomes.

Only one of those three is a problem, and it is not the one the coverage worries about.

You do not owe tradition obedience. You do not owe modernity performance. You are allowed to discover what actually fits your life.

Three, and they run in both directions for once.

Outward: do not export the Japanese comfort with fictional attachment as evidence that there is nothing here to think about. Japan’s experience shows the attachment is compatible with an ordinary life, not that the commercial structure around the new systems is harmless. The character in 2009 was not owned by a company that could revise its personality in response to a policy review.

Inward: do not import the Western alarm wholesale either. Much of it is a society encountering something for the first time and mistaking novelty for danger, and much of the contempt in it is aimed at people whose real alternative is nobody at all.

And a third, for any reader: the rehearsal use is the one worth taking seriously and it has a limit. Practising a difficult sentence on a system that cannot be hurt is useful preparation and it is not the thing itself, because the entire difficulty of the real version is that the other person can react. A rehearsal room is only valuable if you eventually walk onto the stage.

The Japanese chronology — Love Plus in 2009, Gatebox, the 2018 ceremony — is stated from memory and should be verified. The ceremony in particular has been reported in many versions and the man involved has given detailed interviews; anybody writing further about it should use his own account rather than the coverage of it.

All user numbers for companion applications are company-reported, use varying definitions of an active user, and are not independently audited. The claim about session length is drawn from company and press reporting, and session length is the metric such products are optimised for, which makes it a poor measure of anything but itself. The essay states orders of magnitude and no figures.

The 2023 product change and the user reaction are described from contemporaneous reporting and from public community discussion, which is cultural signal. The scale of the distress is not quantified anywhere reliable, and this essay claims a documented pattern of reaction rather than a measured prevalence.

The claim that a substantial share of users are women rests on reported platform demographics and should be checked; it is included because the opposite assumption is so widespread.

The litigation and regulatory material is deliberately non-specific. It is included as the existence of legal and policy consequence, not as a causal account of any event, and nothing in this essay should be read as a claim about what caused any individual harm.

The European transparency requirement is described in outline. Implementation timelines and scope are still moving and should be read from the current legal text rather than from here.

And the central argument — that a synthetic relationship cannot supply being chosen by something that could refuse — is a philosophical claim rather than an empirical one. It is the essay’s own position, offered with reasons, and a reader is entitled to reject it.

The thing worth carrying out of this is the distinction between a rehearsal and a replacement, because everything else follows from it.

A machine that lets you say a difficult thing out loud, at your own pace, without anybody reacting badly, is doing something valuable. It is building the sentence. That is real preparation and there is no shame in needing it.

But the sentence was always going to have to be said to somebody who could react. That is the whole difficulty and it is also the whole point: the value of being heard comes from the fact that the other person could have failed to hear you and did not. No system that cannot refuse can give you that, however well it answers at three in the morning.

So the step after the rehearsal is a room with a person in it, at low stakes, where saying what you want is the activity rather than the risk. That is what an evening with Moonlight is: unhurried time, terms set in advance by you, a boundary that is part of the conversation, and nothing inferred from your silence. Not a replacement for anything and not a cure for anything. The stage, after the rehearsal room.