This article originated from a recent interview I had with a robotics company. They mentioned that the starting point of their entrepreneurial journey was a caregiving experience when their father was hospitalized.
He realized that even the closest family members find it difficult to be attentive to every detail when caring for a patient. This led him to ponder whether there could be a machine that allows us to age with greater dignity.
This thought is simple yet aligns perfectly with the public's expectations for robots.
Over the past two years, humanoid robots have been singing and dancing at exhibitions and on stages. However, if you randomly select an ordinary audience member on-site and ask what they most want a robot to do, the answer will undoubtedly be: Take care of me when I get old.
This desire has also been captured and expressed by literary and artistic creators. A recent overseas light comedy, Ann Droid (translated into Chinese as The Caregiver Is Not Human), centers around Sue, an elderly woman living alone. Her only son is preparing to move out, and before leaving, he buys his mother a humanoid elderly care robot.
Right after being turned on, the robot encounters numerous issues: it shoots lasers at people's faces, suddenly bursts into the room while someone is taking a bath, and its outdated system cannot even remember what the elderly person just told it.
However, after a period of adjustment, this clumsy robot gradually becomes Sue's most reliable companion. This is a much more positive future than the one depicted in The Matrix many years ago.
Never before has a technological product resonated and reached such a strong consensus among entrepreneurs, consumers, and literary and artistic works simultaneously, just like care robots.
Everyone seems ready, just waiting for the iPhone 4 of the robotics world to arrive and spark the next consumer frenzy for smart terminals. So, exactly how far away is that day?
To ferry an elderly person to a more peaceful twilight years, the robot itself must first cross three rivers.
What Exactly Do We Want Robots to Do?
When it comes to robot elderly care, what exactly do we hope robots will do for us? This question might seem so simple that it doesn't require discussion—it's just those tasks needed to take care of the elderly. But what are those tasks? And why do we want robots to do those tasks?
Caregiving, a simple phrase, encompasses at least three completely different types of labor:
The first is physical labor. Scrubbing, turning patients, feeding, lifting, helping them sit up, transferring, cleaning... all directly performed on the human body.
The second is emotional labor. Recognizing when the elderly are afraid, lonely, ashamed, or angry, and responding accordingly. More importantly, caregivers must also manage their own emotions: answering the same question twenty times while maintaining patience; continuing to provide care even after being scolded by the elderly; and refraining from venting their own emotions during conflicts. This is also a form of labor.
The third is cognitive labor. When to take medication? When is the follow-up appointment? Did they fall today? Are there any abnormal physical indicators? Do we need to call a doctor? When should the medicine at home be restocked? These tasks may seem trivial, but they constitute a massive amount of health management in elderly life.
Therefore, caregiving is actually a set of tasks with completely different levels of difficulty and cannot be simply viewed as a single scenario. Only by understanding this can we recognize the limitations of human power and the unsustainability of human caregiving.
No matter how close the relationship, no one can be attentive to every detail and remain perpetually patient during long-term caregiving. Japan even has a specific term for this: Kaigo Satsujin (caregiving homicide).
The annual survey conducted by the Ministry of Health, Labour and Welfare under the Act on Prevention of Elder Abuse concluded that over the 19 years from 2006 to 2024, at least 486 elderly care recipients aged 65 and above died due to abuse and other causes inflicted by their caregivers (including relatives caring for the elderly). Japanese media commented that this is merely the tip of the iceberg regarding the caregiving dilemma in an aging society.
To put it bluntly, robots will not abuse or murder the elderly. This is because robotic caregiving relies on code and has pre-set ethical constraints. Robots will not get tired to the point of losing control, nor will they accumulate emotions until they reach a breaking point.
The good news is that robots might be more suitable than real humans as stable caregivers.
The bad news is that robots are still quite limited.
What people most want robots to replace is precisely what robots are currently least good at: physical labor.
What robots can already do quite well today mainly includes monitoring, reminding, information organization, dialogue, and companionship. These two types of labor primarily occur in the information world, where the input is data and the output is also data. Robots can use sensors to know when the elderly get out of bed, analyze health data through models, and conduct conversations via large models.
But physical labor is different. It must translate information into actions in the physical world.
For example: How much force should a robot weighing tens of kilograms use to lift an elderly person with osteoporosis from the bed? What if the elderly person suddenly becomes afraid, struggles, or becomes uncooperative? What if the bed's position is different from yesterday?
These questions require a massive amount of real-world data concerning the human body, force, contact, environment, and the consequences of failure. This is precisely the biggest bottleneck in the industry today and the most difficult data to obtain, because companies cannot let robots make trial and error on humans ten thousand times.
Therefore, if you expect an elderly care robot that can listen, speak, and remind you, there are already many technologies that can meet this need. However, their essence seems to be not much different from a humanoid smart speaker.
If you expect a robot that can lift you out of bed, bathe you, and turn you over in bed, it is still at the farthest end of the embodied intelligence technology stack.
Between Care Robots and Commercial Success: A Physical World in Between
A humanoid smart speaker obviously does not meet our expectations for robotic elderly care. It can be observed that as robots move from the information world to the physical world, and from cognitive labor to physical labor, the more they interact in reality, the closer they get to the care people truly want, but the more technically challenging it becomes.
Japan is one of the earliest countries in the world to systematically promote the development of care robots. Since 2010, the Japanese government has continuously included robotic care applications such as transfer, mobility, excretion, and monitoring in its key support directions, and has promoted products into care facilities through subsidies, pilot programs, and the care service system.
However, over the past two decades, a general trend can be drawn: the care robots that survive are often small, specialized, and vertical.
For example, Cyberdyne's HAL exoskeleton has entered medical and rehabilitation scenarios, primarily used for assisted rehabilitation training for elderly patients. There are also various Mimamori (watching over) monitoring sensors that detect whether the elderly are in bed, whether there are living beings indoors, or whether they have fallen or need to use the restroom.
These products have all entered the physical world, and their common characteristic is that they are easy to deploy and do not require complex operations.
In contrast, general-purpose intelligent robots that attempt to directly replicate human caregivers mostly remain in the experimental stage.
For instance, RIBA and Robear from the RIKEN institute, whose goal is to lift the elderly, currently only have prototypes and cannot be mass-produced or deployed at scale.
Pepper once entered nursing homes on a large scale, but primarily took on auxiliary functions such as chatting, leading exercises, and interacting. Later, its commercialization path changed. Toyota's HSR also once targeted home service and care scenarios but ultimately became a research and development platform.
The closer a care robot gets to replacing human labor, the harder it is to mass-produce and commercialize; the more single-function and auxiliary the product is, the easier it is to survive.
China is also facing a similar structure today. A large number of humanoid robots with concepts of elderly care, health and wellness, and home care have appeared on the market, but how many have truly achieved large-scale commercial delivery?
After mass production, whether robotic care can generate commercial revenue is another issue. After all, elderly care is not the scenario where robots can most easily demonstrate their capabilities, yet it is the scenario where mistakes are least tolerable.
The willingness to pay and consumption capacity for robots in home elderly care are limited by the ceiling of residents' income. Meanwhile, the risk of compensation claims after robots make mistakes will increase the financial burden on enterprises. The two together determine the income and liability statement of robot companies. If residents' consumption capacity is insufficient and market risks are huge, what motivation do robot companies have to continuously invest in the R&D of elderly care robots?
From the perspective of the risk-reward ratio, it might be a more reasonable commercialization path for robot enterprises to start from rehabilitation and medical equipment, and then move towards humanoid robots.
Who Is Responsible When Things Go Wrong?
Following this path, we can also deduce the implementation path of robotic elderly care, which will first land in care facilities rather than in home consumption.
The reason is that the home environment is far more complex than factories and care facilities.
The furniture, beds, floors, and lighting in every home are different. Creating a general-purpose robot capable of performing generalized tasks in complex home scenarios is technically too difficult. Even if the technology is achieved, the home scenario will further amplify the issue of responsibility boundaries.
In contrast, the environment in facilities is more structured. Bed sizes are relatively uniform, and processes are clearer, which greatly reduces the risk of technical errors. Moreover, with professionals on-site and clear purchasers, the chain of accountability is much clearer than for home robots. If a robot makes a mistake at home and an accident occurs, it will face the division of responsibilities among robot hardware manufacturers, software algorithm companies, family consumers, and other parties.
Although the product liability system in the Civil Code can cover a considerable number of robot accidents and can pursue the relevant responsibilities of producers and sellers, the source of accidents involving elderly care robots may not be single or direct. For example, if a machine makes an error, it could be due to hardware defects, algorithm issues, model judgment errors, poisoned training data, software updates, improper human-robot collaboration, human operational errors, incorrect user settings, or autonomous robot decision-making... which greatly increases the difficulty of dividing responsibilities.
Therefore, facing such high thresholds for rights protection, we can easily infer that it is more realistic to prioritize facilities over homes, and commercial services over home consumption, allowing robots to first become a part of the caregivers' workflow.
The Ferryman of Old Age Doesn't Have to Be Human
At this point, we no longer need to ask: Can robots provide elderly care for us? The answer is that in a future with an aging population and continuously tight caregiving manpower, machines might be the real hope for the vast majority of us to receive care.
The real question we should ask is: In what order will robots enter our elderly care lives.
This path will not be a blitzkrieg of general-purpose humanoid robots, but rather a long and arduous river crossing.
First, specialized roles such as monitoring and reminding, acting like a pair of tireless eyes, an always-online ear, and a system that never forgets medication times.
Then, companionship and rehabilitation, some standardized complex scenarios, which might be a companion wheeled robot that follows our steps, or an assistive robotic arm that helps us get out of bed.
One day, when we are truly old, we hope that the machine has already crossed every river over the past decade or so. At that time, the robot will truly come to our side, becoming the ferryman of our twilight years.