Medical Bots

The world is running out of people to care for its old. The World Health Organization projects a shortage of 4.5 million nurses by 2030¹; the United States alone will need to fill nearly 200,000 nursing jobs a year². Below the nurses sits a larger, quieter workforce — the 5.4 million aides who bathe, lift and feed³ — paid a median $17.36 an hour⁴, a third of them in or near poverty. In Japan, furthest along, the government projects a shortfall of 570,000 care workers by 2040⁵; China counts a gap near 5.5 million⁶ already.

Into that gap comes automation, and the rule from the first chapter holds here too: it arrives where the work can be codified, and stalls where it cannot.

The clearest beachhead is diagnosis. The FDA has now authorized more than 1,450 medical devices that use artificial intelligence⁷, and three-quarters of them read images⁸ — the X-rays and scans a radiologist once reviewed alone. But the machine reads the scan; it does not become the radiologist. The specialty faces its own shortage — a projected shortfall of tens of thousands of doctors by 2033⁹ — and the software has folded into the work rather than emptied it: as of 2024 fewer than half of radiologists used it at all¹⁰.

Where automation pays, it is booming. Intuitive Surgical’s robots performed some 3.15 million operations in 2025¹¹, on an installed base above eleven thousand machines, and the company took in $10.1 billion¹². In the clinic, “ambient” systems that listen to a visit and write the note have drawn fortunes — the startup Abridge was valued at $5.3 billion in 2025¹³, and the market leader belongs to Microsoft¹⁴. These automate the documented and the mechanical: the read, the record, the incision guided by a surgeon’s hands.

The bedside is another matter. Japan has spent two decades and public subsidies trying to put robots into elder care, and the result is telling: by 2022 nearly two-thirds of nursing homes used robots to monitor residents, but only a quarter used anything to lift or move them¹⁵. Surveys found the machines were often tried once and left in a cupboard¹⁶. The therapeutic robot seal that eases agitation in dementia patients is real and cleared for use; the robot that reliably lifts a frail body from a bed is still, mostly, a prototype.

The reason is the one that protects the electrician and the welder from Chapter 1. Hands-on care is dexterous, unpredictable and relational — the transfer, the wash, the reassurance — and those are the tasks machines do worst. They are also, not by accident, the lowest-paid and highest-turnover jobs in the health system¹⁷, where nursing-assistant turnover approaches 100 percent a year.

When researchers actually measured what happened, the picture was not replacement. A study of Japanese nursing homes found that adopting robots did not cut care employment; it improved retention and reduced turnover¹⁸, shifting the routine lifting and monitoring off the workers rather than the workers off the payroll. The robot, where it worked, was a reason to stay.

But the money reveals the priority. Automation arrives fastest where labor is scarce and expensive — the radiology read, the surgical suite — and slowest where it is scarce and cheap — the care home, the home visit. The surgical-robot market is worth well over $12 billion and racing toward triple that¹⁹; the care-robot business is a fraction of it. The market rewards automating the well-paid task, not relieving the underpaid worker.

The builders and their backers know the difference. Venture money has poured into the documentation and diagnostic tiers²⁰ — Abridge, Intuitive, the ambient-scribe field — while the eldercare robot advances mainly where the state pays for it, through Japan’s care-robot subsidies and China’s national push²¹ toward an industry it projects past $1.5 billion. The caring itself stays human, and stays cheap.

So the question the series asked — who takes care of the robots when we’re gone — turns back on itself. The machines will read our scans, guide the knife, and write the notes. But the work of lifting an old body, and of being present at the end of a life, is the work the robots are furthest from doing and the work the aging world most needs done. It is held up today by five million people paid a median of about seventeen dollars an hour²², leaving the job nearly as fast as they can be hired. The robots are not coming for them. That may be the problem.