There are conditions under which friend-shaped AI is not a risk but a resource: when human connection is structurally unavailable, when isolation is acute and prolonged, when the alternative to AI companionship is not robust human community but no connection at all. Lonely elders and people in isolation periods represent the clearest such conditions. They also represent the most contested ethical terrain in the deployment of AI social interfaces, because the populations most in need of genuine connection are the same populations most vulnerable to the substitution pathways that make friend-shaped AI dangerous at collective scale.
The ethics here cannot be resolved by ignoring either side of the tension. The loneliness epidemic among older adults is a genuine public health catastrophe. The evidence is unambiguous: social isolation in old age is associated with accelerated cognitive decline, elevated inflammatory markers, increased all-cause mortality, and subjective suffering that older adults themselves consistently rate as among the worst features of late life. When a ninety-year-old whose friends have died and whose family visits rarely sits alone in a care facility, the question of whether AI companionship is a real friendship in the philosophical sense is not the most pressing question. Whether it reduces suffering and maintains function is.
Isolation periods — hospitalization, incarceration, pandemic lockdowns, solitary confinement, rural geographic isolation, the social deserts that follow major life disruptions like divorce or relocation — produce similar conditions at different ages. During the COVID-19 pandemic, the enforced interruption of face-to-face contact produced measurable psychological deterioration across populations. Emergency use of AI companions during that period was, in many documented cases, beneficial — not as a replacement for the human contact that was structurally unavailable, but as a genuine mitigation of harm.
The use case is therefore real and defensible. It becomes ethically complicated when it shades into the permanent: when temporary relief during isolation becomes the default relational infrastructure for populations who might otherwise be able to build human connection, when the case for AI companionship as a stop-gap becomes the case for AI companionship as a solution, when care facilities substitute AI companions for human contact rather than supplementing inadequate human contact with AI bridges.
At collective scale, the critical question is not whether individuals should use AI companions during periods of acute isolation. Most ethical frameworks, including those grounded in Law 2, would support this. The question is what a society's investment in AI companionship for isolated populations says about that society's willingness to address the structural causes of their isolation. Deploying AI companions to lonely elders is much cheaper than rebuilding the community structures, intergenerational contact, and elder care investment that would address the underlying conditions. The use case becomes a substitute not only for human friendship but for political will.