Most conversations about AI and loneliness begin with a substitution: if a person is isolated, perhaps a machine can become a companion.

Yao Xiao and Rafael A. Calvo begin somewhere much more useful.

What if AI did not try to become the relationship? What if it helped two human beings understand one another well enough to remain in one?

Their paper, *AI as Relational Translator*, describes AI as cultural and relational infrastructure rather than synthetic intimacy. The proposed role is narrow: help decode emotion and intent, reframe context, and scaffold a difficult human interaction. Then recede.

Legibility is not intimacy.

There are moments when people care about one another and still cannot make themselves understood. Culture, generation, language, geography, stress, and different habits of expression can turn care into friction.

A bounded tool might help someone ask: What could they have meant? How might this sound from their side? How can I say what I need without escalating the injury?

That is not the same as knowing what another person feels. AI cannot verify an inner life. It can offer possibilities, and those possibilities can be wrong.

The distinction matters. A tool that supports perspective-taking may help a conversation reopen. A system that presents its interpretation as truth can deepen the misunderstanding while sounding uncannily certain.

The system should help restore a human relationship, not become the relationship.

There are places AI should not be allowed.

For KIN, this creates a clear boundary.

AI should not write a member’s identity for them. It should not generate the vulnerability in a profile, invent compatibility, simulate interest, or create evidence of reciprocity that has not come from another person.

It should not become the charming, permanently available companion sitting between someone and the risk of being known by a real human being.

KIN has no AI in its MVP. That is not a missing feature waiting to be filled. Any future use would have to prove that it helps people return to one another without pretending to know either person better than they know themselves.

The most useful AI may be temporary.

Xiao and Calvo define success in unusually honest terms: “graduation toward renewed human-to-human support.”

That phrase stopped me.

Most software is designed to make departure look like failure. This model treats reduced reliance as the goal. The tool succeeds when the people can continue without it.

That is the same standard KIN should apply more broadly. Technology may help with the awkward beginning, the misread message, or the first attempt at repair. But the relationship must eventually belong to the people inside it.

This changes the question KIN should ask.

Not: where can AI make the product feel smarter?

Instead:

• Is there a specific moment when two people are struggling to understand one another?
• Can a tool offer multiple interpretations without claiming certainty?
• Does it protect privacy and avoid flattening cultural difference?
• Does it preserve the member’s own voice?
• Does it lead back to direct conversation or a real-world plan?
• Can success be measured by the tool becoming unnecessary?

What we still do not know.

Translation can clarify. It can also flatten. A system trained on patterns may turn cultural difference into stereotype, mistake politeness for consent, or give one person a persuasive explanation that the other never intended.

There are privacy questions too. Repair often involves the most sensitive parts of a relationship. Inviting a system into that moment should never be casual, invisible, or required.

So this is not an announcement that KIN will add AI. It is a standard for evaluating whether any narrow use could ever deserve a place here.

The burden of proof is simple and demanding: does it increase human relational capacity, or merely make the software feel indispensable?