Social isolation now rivals smoking and obesity as a risk factor for premature mortality, yet scalable, low-cost interventions remain scarce. The question of whether artificial intelligence can serve as a genuine social surrogate — not merely a distraction — matters enormously for aging populations, remote workers, and anyone whose social networks have thinned in the post-pandemic era.
Two sequential experiments with healthy Chinese adults tested whether shared activities with AI partners could meaningfully shift loneliness metrics. In the first experiment (N = 180), participants imagined three paired activities — singing, running, or medical diagnosis — with either a human or AI partner. Anticipated loneliness reduction, social presence, companionship, and perceived empathy were assessed. Co-singing and co-running with AI partners both generated positive anticipated outcomes, though consistently weaker than human equivalents; task type mattered, with expressive and physical activities outperforming an instrumental one. The second experiment (N = 60) moved from imagination to reality: participants actually sang with either a human or AI partner in physical or virtual settings. AI co-singing produced a statistically significant reduction in reported loneliness. Crucially, while human partners delivered stronger social presence and empathy gains in physical versus virtual modes, AI partner efficacy was entirely unaffected by modality.
This modality-independence finding is arguably the most clinically interesting result. It suggests AI-based co-experience interventions could be deployed at scale without requiring physical presence infrastructure. That said, this work carries notable limitations: the sample is exclusively healthy young Chinese adults, meaning generalizability to elderly, clinically lonely, or cross-cultural populations is unestablished. Both experiments are also relatively small, and the loneliness measures are self-reported in the short term. Whether repeated AI co-singing sessions produce durable reductions in chronic loneliness — or whether novelty effects drive initial gains — remains an open empirical question. Still, for researchers designing digital health interventions, this represents a credible proof-of-concept that AI-facilitated expressive activities may occupy a legitimate rung on the social support ladder.
