Japan’s robot-loneliness trial: the responses came from people

Murayama and Takase, JMIR Aging, 22 October 2025 tested BOCCO emo among 73 randomized older Japanese people living alone and scoring as lonely; 68 completed four weeks (34/group). The primary UCLA loneliness score improved by −3.1 points relative to control (95% CI −5.9 to −0.4), while well-being rose by 1.9 (CI 0.1 to 3.7). No significant change in reported monthly contact with other people; secondary outcomes largely null. Design was unblinded, sample 94% women, included cognitively/physically independent participants and had four-week follow-up; sponsor SECOM funded study and helped collect data but not analysis according to authors.

Crucial mechanism: This was not a trial of an autonomous AI companion replacing a carer. Participants’ voice messages were sent to trained human operators whose personalized responses were spoken by the robot; family members could also message through it; programmed reminders and quizzes supplemented the conversation. This is a promising human-mediated remote contact channel, not proof machines can give independent intimate care or replace nurses. It also resists an indiscriminate Human Reserved ban on devices in care: such a device can carry human attention to a housebound person. Trial did not test equivalence to an in-person care worker or model access, costs, clinical quality, long-term displacement or care shortages. Authors conjecture potential generalizability to automated robots but did not randomize that comparison.

Wolfe et al., JMIR, March 2025 interviewed/surveyed 28 older adults after a facilitated chatbot interaction: 25/28 interested in appointment reminders, 22/28 in emergency assistance, 17/28 in social-connection features; willingness varied, and privacy/learning/dependence concerns featured. Preferences in a convenience sample are not measured health effects. US HRSA December 2025 workforce demand model projects 40% rise in long-term services/support FTE demand 2023–2038, explicitly does not project future supply, so not an observed shortage of this size. Read with policy synthesis and Gates’s Japan exception.

Needed next trial: in-person caregiver, trained human via device, AI-aided human caregiver and autonomous AI with clear rescue paths, each matched by contact time and cost, measuring clinical errors, neglect, relationship/consent, loneliness, access and actual paid-worker hours. In people with dementia, capacity and vulnerability require distinct study; do not transfer this cognitively intact community trial.