Imagine a daily report from an eldercare service: breakfast delivered, floor cleaned, activity reminders sent. Every task is green. The system appears to be working well.
But an older person's wish to tell a story from their past never made it onto the task list. Whether they feel comfortable with their care, have choices or remain connected to others is also absent from the completion rate.
This does not diminish the work machines do. An object delivered within reach at the right moment, or one less burden of lifting and carrying, can matter greatly. The question is:Care efficiency should serve a person's life, and a person's life cannot be fully described by task completion rates.
First, distinguish the work a machine might take on
In eldercare, carrying, cleaning, reminders and support for remote communication are different from clinical judgment, emergency response and emotional relationships. Each requires different capabilities, reliability and arrangements for responsibility.
Grouping all this work under the label of care robot can raise expectations beyond a product's abilities. Reminding someone to take medicine is not the same as determining whether it needs adjustment. Recognizing an unusual signal is not a professional assessment. Each intended use needs appropriate validation and human support.
Likewise, removing a care task does not remove responsibility for care. Someone must still arrange services, handle alerts, contact relatives and take over when equipment cannot continue.
The Oxford Internet Institute's overview of digital care research addresses both convenience and burdens: remote tools may improve coordination while keeping caregivers constantly on call and introducing privacy and monitoring concerns. This is evidence for understanding system design. It cannot be presented as clinical validation of a particular care robot.Oxford research overview
Where does the saved time go?
Once machines take over repetitive work, at least two outcomes are possible. Care workers may have more time to talk with older people. Or organizations may reduce staffing and visits, asking machines to cover more services. Technology itself cannot decide which happens.
The same efficiency gain can become very different care experiences under different budgets and staffing rules.
A fuller approach to acceptance is needed. Alongside whether a delivery arrived, ask whether the person can express preferences, whether requests for help receive a response, whether caregivers' total burden declines and whether human contact matches the person's own wishes.
We should not mechanically treat more human contact as the only goal. Some people value solitude. Others do not want to expose their bodies or private routines to unfamiliar caregivers every day. Technology may help them preserve autonomy. People receiving care should help evaluate relationship quality, rather than leaving relatives or institutions to define it for them.
The central criterion should therefore be whether the machine helps this person live a life they prefer.
Loneliness may ease, but relationships cannot be promised away
A study of AI companionship published in the Journal of Consumer Research found immediate relief from loneliness. That gives us reason to take the potential value of synthetic companionship seriously, while keeping the timescale clear: a short-term improvement in feelings does not establish a long-term improvement in real-world relationships.AI Companions Reduce Loneliness
Evidence from chat interactions cannot be directly extended to physical robots in care facilities. Nor can it establish suitability for older people with widely differing cognitive and health conditions. Physical presence, prolonged use, institutional settings and care needs may all change the effects.
The lack of a long-term answer need not prevent exploration until everything is certain. A reasonable path begins with clearly bounded tasks, preserves the older person's choice, records changes in daily life and retains human support and ways to leave.
An older person who enjoys talking with a machine should not be mocked. Someone who does not want to use one should not be forced to accept it for budgetary reasons. Respecting both choices is part of respecting users.
When care is understaffed, what can machines provide?
Staff shortages are a powerful practical consideration. Care resources are limited, and relatives have work and lives of their own. If technology supplies help that was otherwise absent, it may substantially improve the situation even when imperfect.
Unpaid care already carries broad social costs. ILO estimates for 2023 indicate that about 708 million women were outside the labor force because of care responsibilities. This covers different kinds of care: it cannot all be attributed to eldercare, much less treated as the scale of possible robot substitution. It does show why reducing burdens deserves serious investment.ILO statistical note
Acknowledging this argument makes it more necessary to prevent the claim that a machine is better than nothing from becoming a lasting excuse for lower service standards. Compare machines with realistic alternatives, and make clear which responsibilities must remain with people.
Technology can expand service capacity. Care providers must still explain who responds when something goes wrong, how long that takes and how care continues during downtime. Saying the AI will learn cannot answer these questions.
Give care robots a fuller report card
I suggest pilots examine at least four outcomes together: safe and reliable tasks, greater autonomy for older people, human connections that meet their needs, and lower total burdens for relatives and care workers. Record loneliness, actual human contact and satisfaction separately; chat counts cannot stand in for them. Report safety incidents, missed alerts and false alarms separately as well, rather than allowing an overall satisfaction score to hide them.
These outcomes may not improve together. Remote viewing may reassure relatives while making an older person feel watched. Frequent alerts may increase the chance of detecting certain events while also increasing caregiver stress. Evaluation must allow these conflicts to surface, so the person receiving care, professionals and the service provider can adjust together.
Procurement contracts can reflect this approach. Instead of rewarding only more operating hours and more interactions, they should specify task boundaries, human takeover, data access and exit arrangements. Whether saved staff time improves service should be an explicit, traceable operational choice.
Pilots should also record why people do not use the product. Some leave because controls are complicated, some worry about privacy, and some simply dislike its voice or appearance. Excluding them before reporting satisfaction distorts the findings. For participants who remain, distinguish initial novelty from experience after sustained use, and examine whether relatives' assessments match the person's own. A machine that reassures relatives but leaves an older person uncomfortable has not settled the problem. Keeping refusals, departures and disagreements within the evaluation brings improvement closer to users' needs.
Within validated task boundaries, Physical AI has a chance to improve specific parts of care. A more valuable outcome is helping older people do more for themselves, while allowing human relationships to be less exhausted by practical chores.
If a care facility saves money through robots, would you prioritize lower fees, wider service coverage or more time for human companionship?