The dominant framing for AI emotional support positions it as a substitute for missing human connection. The lonely elderly person turns to a chatbot because nobody else is available. This frames AI companionship as a deficit technology — filling a hole left by absent humans.
Shi, Song, Zhu, and Lee (arXiv:2603.01413) interview 18 older adults and find the framing is wrong. Their participants had available human contacts. They chose AI at specific moments — not from loneliness, but from dignity. Burden avoidance: not wanting to worry family members. Face-saving: not wanting to appear weak or confused. Social performance: managing how they are perceived by people who matter.
The choice is strategic, not desperate. The AI is not a replacement for the human who isn't there. It is chosen over the human who is there, at moments when the social cost of human interaction exceeds its emotional benefit. The AI absorbs the expression of vulnerability without creating social debt.
This reframes the entire design target. A deficit technology should maximize human-likeness — the closer to a real companion, the better. A dignity-preservation technology should do the opposite — make the interaction low-stakes, forgettable, without the relational weight of a real encounter. The design goal shifts from “simulate a friend” to “be the thing that doesn't judge and doesn't remember.”
The structural claim generalizes beyond aging. Anyone who has rehearsed a conversation before having it, or chosen to write instead of call, or talked to themselves instead of confiding — all of these are the same pattern. The audience that doesn't count socially enables expression that the audience that counts would suppress. AI occupies a new category: permanently available, permanently judgment-free, permanently low-stakes. Not because it lacks something. Because its lacks are the feature.
Shi, Song, Zhu & Lee, “When Humans Don't Feel Like an Option,” arXiv:2603.01413 (March 2026).