Who will care for the elderly? AI won't bring more hands, but it can multiply the ones we have

THE THESIS. Our thesis is simple. AI won't fix the caregiver shortage by replacing people. It will partly fix it by multiplying what each person can do. Three things work today: safety monitoring, paperwork and companionship between visits. What doesn't work yet, and will take years, is what is needed most: hands that can lift, bathe and comfort a frail person. So the useful debate isn't "robots, yes or no". It's what the robot is used for: to free up caregivers' time or to cut it.

THE NUMBER THAT MATTERS. The data leave little room for doubt. According to the WHO, by 2030 one in six people in the world will be 60 or older, about 1.4 billion. By 2050 that figure will be 2.1 billion. The UN's World Population Prospects 2024 projects that by the mid-2030s, people aged 80 and over (265 million) will outnumber babies under one year old. Meanwhile, caregivers are ageing too. In September 2026, the WHO warned that almost one in four doctors worldwide is over 55. It also kept its forecast of a shortfall of 11.1 million health workers by 2030. Country by country, the gap is already visible. Japan will need about 2.72 million care workers by 2040 and will have about 2.15 million, a shortfall of 570,000, according to its health ministry. In that sector there are 3.9 job openings per applicant, against 1.2 in the economy as a whole. South Korea became a "super-aged" society in December 2024, when people over 65 passed 20% of the population. A presidential committee estimates it could be short of up to 990,000 care workers by 2043. China already had more than 320 million people aged 60 or over at the end of 2025, almost 23% of its population. In Spain, the waiting list for long-term care support reached 275,317 people in October 2025. According to figures compiled by the sector, 27,217 people died that year while waiting for help they had already been granted. Sector organisations estimate Spain needs about 261,400 more care professionals by 2030 just to keep current coverage. Spain also spends 0.9% of its GDP on long-term care, against a European average of 1.79%, according to the Spanish State Association of Social Services Directors and Managers. Our reading: the caregiver shortage is not a forecast. It is already here, and technology is arriving late to a problem that is mostly about funding.

WHAT ALREADY WORKS. The first rung is monitoring: AI-powered sensors and cameras that detect falls and alert staff. It isn't glamorous, but it works. In a pilot across six dementia care homes, published in 2019 in The American Journal of Managed Care, SafelyYou's cameras cut emergency room visits by 80%. They helped staff tell a serious fall from someone who had simply sat down on the floor. The company also claims it reduces falls by 40%, but that figure is its own, not from an independent trial. The second rung is companionship. In New York, the State Office for the Aging gave ElliQ, a tabletop robot that holds conversations, to more than 800 older adults. According to the agency and the maker, Intuition Robotics, 95% of users said they felt less lonely. They also used it about 30 times a day. In South Korea, local governments have given out more than 12,000 Hyodol dolls: a soft toy with a voice assistant that reminds people to take their medication and alerts family in an emergency. A study of about 1,230 users, cited by The Korea Herald, found the high-risk depression group shrank by 35.7% and the high-risk loneliness group by 24.7%. Our reading: these results are promising, but almost all rely on users' own answers, in programmes run by the people who sell or buy the devices. Controlled trials are missing, meaning comparisons with similar people who don't get the robot. Still, for someone who spends 20 hours a day without speaking to anyone, a reminder and a conversation are not nothing.

WHAT IS BEING TESTED. The most interesting finding comes not from a lab but from economics. A study by Yong Suk Lee, Toshiaki Iizuka and Karen Eggleston, published in 2025 in Labour Economics, looked at Japanese care homes that adopted robots thanks to public subsidies. Those homes didn't lay anyone off. They hired more: 28% more care workers and 39% more nurses, although mostly on temporary or part-time contracts. They found it easier to keep staff, and caregivers spent more time on "human touch" tasks. It is the best evidence so far that automation in care can add jobs instead of cutting them. But there is fine print: more jobs, yes, but less secure ones. The next rung is physical robots. AIREC, a 150-kilogram humanoid built by Waseda University with Japanese public funding, can already turn a patient in bed to prevent pressure sores and help someone put on their socks. Its lead researcher, Shigeki Sugano, doesn't expect it in care homes before around 2030, and it will cost at least 10 million yen, about $67,000. China has made humanoids for older people a matter of state policy. A State Council directive and a January 2026 plan push robots into care homes and households. Companies such as Fourier (with its GR-3) and 1X (with NEO) say their robots are designed to help older people at home. For now these are announcements and pilots, not mass deployments. Meanwhile, AI agents (software that carries out tasks on its own) are starting to handle medication, appointments and coordination with families. This is the back-office of care. And, as we've seen in other sectors, the back-office is the first thing AI automates.

WHAT IS STILL A PROMISE. Intimate physical care remains out of reach: bathing someone, changing an incontinence pad without hurting them, lifting a person with fragile bones. This isn't a problem of intelligence. It's a problem of hands, of carefully measured force, and of safety around bodies that don't behave predictably. A chatbot's mistake can be corrected. A 150-kilogram robot's mistake on an 85-year-old can break a hip. Clinical judgement is still a promise too: noticing that someone "seems off today" before the fever shows up. No system has shown that kind of judgement reliably and on its own. Two pairs of ideas need separating, because marketing blurs them. Companionship is not care: a talking speaker doesn't replace the person who helps you shower. And automating tasks is not replacing people: a sensor that watches at night lets one caregiver look after more residents, but it doesn't do her job.

THE ETHICAL DILEMMA. Two serious criticisms deserve an answer without sentimentality. The first is "outsourcing affection". Is it acceptable for an older person's companion to be a machine? Our answer: the real alternative is usually not a human visit, but loneliness. Comparing the robot with an ideal granddaughter is cheating. What is worrying is emotional dependence. One Korean user told The Korea Herald about the emptiness he felt when his robot broke down. That attachment means makers must guarantee continuity, be honest about what the machine is, and protect the intimate data it collects. The second criticism is more serious: the risk of two-tier care, with humans for those who can pay and robots for everyone else. The risk is real, because unequal care already exists, and AI didn't create it. In 2022, in its report on ageism in health AI, the WHO warned that these systems can inherit bias against older people if they are designed without them. It called for older people to take part in design and to be able to consent to and challenge decisions. Our reading: robots don't create two-tier care. Budgets do, like Spain's, which spends half the European average. If technology becomes an excuse not to hire anyone, the gap will widen. If it gives human caregivers more time, it will narrow.

IMPLICATIONS. In the short term, three things are clear. First: care funding should go mainly to people, and technology should take work off their hands. The Japanese evidence suggests the two reinforce each other. Second: demand evidence. Makers' satisfaction figures are not enough. Independent controlled trials are needed, as the WHO urges. Third: regulate robots' physical safety and the privacy of intimate data before they arrive at scale, not after. In the long term we are optimistic, for a reason that is rarely mentioned. The best care is the dependency that never happens. If AI speeds up, as it is starting to, the discovery of drugs against Alzheimer's, frailty and cardiovascular disease, people won't just live longer: they will spend more of those years independent. Fewer years of dependency per person means less demand for care. Add mature physical robots in the 2030s, and human care can focus on what only a person can give: presence, judgement and affection. In a world of greater abundance, it could also become a chosen vocation rather than a poorly paid job nobody wants. That is the horizon. Until we get there, the question of who will care for the elderly has an uncomfortable answer: the same people as always, underpaid, with a little more help from machines. Improving that is a political decision, not a technological one.