The Japan Institute for Promotion of Digital Economy and Community published, in late May 2026, the results of a survey conducted online in mid-January. The survey asked 1,449 Japanese residents between the ages of 18 and 79 whether they would prefer to consult a human or an artificial intelligence about interpersonal relationship issues. The overall results were unremarkable. 45.8 percent of respondents said they would prefer to consult a human. 36.5 percent said they would prefer to consult an AI. The remainder gave various other responses.
The unremarkable overall results, on close examination, obscured a single demographic finding that the wider cultural register has, in the days since, started to take more seriously than the standard absorption of survey data ordinarily allows for. The finding is that one specific demographic group reversed the overall pattern. 47.8 percent of Japanese women in their sixties and seventies said they would prefer to consult an AI rather than a human about relationship issues. The 47.8 percent figure was the highest AI preference of any demographic in the survey. The figure was also, on close examination, higher than the proportion of women in the same age range who preferred consulting a human, which came in at 37.3 percent.
The structural feature worth attending to is that this is the only age group in the survey that, on the available data, made this choice. Men in the same age range preferred humans by 57.0 percent to 25.2 percent. Younger demographics across both sexes also preferred humans overall, with the exception of teenage girls, who showed elevated AI usage for personal advice but in a different structural pattern from the elderly women.
What the standard framing is missing
The standard cultural framing of this finding has tended to absorb it as a piece of mildly surprising technology adoption data. The framing has been calibrated to the assumption that AI preferences correlate with age in a particular way, with younger users more comfortable with the technology and older users less so. The finding contradicts the assumption. The standard framing has accordingly registered the contradiction, expressed a small degree of surprise, and moved on.
What the standard framing has not, on the available evidence, adequately attended to is what the finding actually implies about the structural condition of elderly women in contemporary Japanese society. The finding is not, on close examination, primarily a piece of technology adoption data. The finding is, more accurately, a piece of social structural data, in which the technology preference is the visible signal of an underlying condition that the wider register has been considerably less willing to engage with.
The underlying condition is the well-documented elderly isolation crisis in contemporary Japan. According to the National Institute of Population and Social Security Research, 19.4 percent of Japanese seniors aged 65 and older currently live alone, and the number of elderly single households is projected to increase by 47 percent by 2050. Among Japanese people who die alone, approximately 59 percent of women are aged 60 or older. About 47 percent of Japanese people, in a Cabinet Office survey, reported feeling lonely “always” or “sometimes.” The structural conditions are real. The structural conditions have been deepening.
What Atsushi Nakagomi’s observation actually points at
The most cited expert response to the survey, from Atsushi Nakagomi, an associate professor at Chiba University who studies the intersection of AI and human health, is worth attending to closely. Nakagomi said, in response to the finding, that he was surprised that it was elderly women who were more likely to prefer AI. He added that AI “makes people feel more comfortable about opening up, as they might feel free to seek advice without worrying about how their comments will be perceived.”
The observation is, on close examination, doing more interesting work than the standard absorption of the quote tends to allow for. The observation is, more specifically, pointing at the structural feature that makes the AI preference legible in the demographic where it appears. The feature is the freedom to disclose without the social cost that disclosure to other humans, in the specific cultural context, structurally imposes.
The cultural context matters here. Japanese society has, by every available measure of comparative cultural analysis, considerably stronger structural norms about the management of social presentation than most Western cultures do. The norms produce, in the Japanese population, particular structural costs to the disclosure of difficulty, vulnerability, or relational conflict to other humans, particularly to humans within the social network the disclosure would concern. Research published in Psychogeriatrics using data from 13,766 Japanese adults aged 65 and older has documented that loneliness and social isolation in this population are associated with specific structural features including socioeconomic disadvantage, depressive symptoms, and physical health problems. The features are real. The features are also, on close examination, the kind of material that the standard Japanese conversational register is least equipped to engage with substantively.
The AI does not impose the structural cost. The AI has no social position. The AI has no relationship to the people the elderly woman might be in conflict with. The AI cannot, by structural design, gossip, judge, or alter its relational position to the woman as a result of what the woman discloses. The AI is, in some real way, the first available interlocutor in many of these women’s lives that they can disclose to without the disclosure incurring the social cost that disclosure to other humans has, across decades, been structurally imposing.
What this connects to, in the wider context
The structural context this finding sits inside, on close examination, is worth attending to. Japan has, by every available demographic measure, one of the most aged populations in the world, with approximately 29 percent of its population aged 65 or older as of 2022. Previous research indicates that 10 to 30 percent of Japanese older adults experience social isolation, with 2 to 10 percent facing severe social isolation, particularly among those who have contact with others less than once a week.
The wider social infrastructure that historically supported the elderly population, including the multi-generational household structure and the various community-based support networks, has, across the previous several decades, been progressively eroded by urbanization, declining birth rates, and the various structural changes of late-twentieth-century Japanese economic life. Research from a nationwide Japanese survey of 5,351 older adults has documented that single-person household residents score significantly higher on the UCLA Loneliness Scale and significantly lower on the Lubben Social Network Scale than multi-person household residents. Living alone is, by every available measure of the research, the strongest predictor of isolation and loneliness in older age. The number of older women living alone in Japan has been rising for decades.
The erosion has produced, by every available measure of elderly well-being, a population-level isolation problem. The problem is not, in most cases, the absence of contact with other humans in the abstract. The problem is, more specifically, the absence of the particular kind of substantive engagement that the eroded social structures had, in previous generations, structurally produced. The elderly women who are now reporting their preference for AI advice are, on close examination, the demographic most directly affected by this erosion. Many have outlived spouses, watched their adult children move to different cities, and aged out of the workplace structures that had previously provided their primary social engagement.
The AI is, in this context, not primarily a technology preference. The AI is, more accurately, the available alternative to the structural condition the women are sitting in. The condition is the absence of a substantive interlocutor who can be addressed without social cost. The AI provides such an interlocutor. The provision is what the survey is, on close examination, measuring.
The acknowledgment this article wants to leave
The survey finding from the Japan Institute for Promotion of Digital Economy and Community is, on close examination, considerably more substantive than the standard cultural framing has registered. The finding is not primarily about technology adoption. The finding is, more specifically, a piece of social structural data in which 47.8 percent of Japanese women in their sixties and seventies have, by their reported preferences, indicated that the available human alternatives in their lives are not producing the substantive engagement they are looking for, and that an artificial intelligence has become, in their accounting, the preferable interlocutor for the relationship advice the standard cultural register would assume should be coming from other humans.
The underlying condition is the well-documented elderly isolation crisis that contemporary Japan is, by every available measure, currently in the middle of. The condition is real. The condition has been accumulating across decades. The condition has been the subject of considerable institutional concern and various policy responses, none of which have, on the available evidence, adequately addressed the underlying erosion of the social structures that had previously supported elderly engagement.
The AI preference is, accordingly, not the problem the survey is identifying. The AI preference is, more specifically, the visible signal of the underlying problem. The underlying problem is that a significant fraction of Japanese elderly women have, on the available evidence of their own reported preferences, concluded that an artificial intelligence is a better interlocutor for their substantive concerns than the available humans in their lives. The conclusion is being reported in survey data. The conclusion is, in some real way, what the wider cultural register has been considerably slower to absorb than the survey data itself has been to produce.
