I don’t understand the appeal of wheeled robots in a hospital. It isn’t like speed is a need, a bipedal robot can run if it needs to get somewhere. A robot that can run 20 mph is more than sufficient (and way too fast if anything) so you don’t need the speed that comes from wheels.
Plus a bipedal robot will have an easier time with stairs and other obstacles than a wheeled robot.
The problem is that a bipedal design continuously consumes power trying to stand up and stay balanced. A tall wheeled cart just sits there, statically supported on its wide-enough base.
Unless any robot has a power umbilical attached to a wall outlet, giving it enough battery power at an acceptable weight to make it run for an acceptable amount of time is a very hard design tradeoff to close. Adding bipedalism to the mix makes that a lot harder. Batteries are useful anti-tipping ballast installed in a heavy wheeled base. They are a pro-tipping problem if installed in a “torso” high atop two legs.
Stairs aren’t a big problem. You can find tons of examples of mechanized rising wheels or just big squishy tires that roll up stairs. Or have one per floor. Or have it use a robot elevator.
Wheels aren’t about speed. It’s about stability and ease of moving around. Walking around means you’re constantly balancing to stay upright, tripping hazards and have more points of failure with jointed moving legs.
That just means the bipedal robot has to charge a little more often.
And unlike human nurses, which have to sleep, the robot nurse can just transfer its brainstate to a different body when it’s charging. The patient will never know the difference except for the scratches being in slightly different places.
For the purposes of this thread, do you consider this a given? If so, can you explain what you mean by “understanding”? As someone who doesn’t expect to see AGI in a VERY long time, if ever, I’d like to figure out what baseline you are setting.
Today, for example, this super nurse would sometimes prescribe something like 30 mg of Metroclopapril. Since this doesn’t actually exist, I’m a bit hesitant to let the thing inject something into me.
The issue is that humans are imperfect too.
Do I consider it a given? Yes. Its my understanding that AI is currently good at sifting through all existing human knowledge, but not very good at creating new knowledge. Supposedly that is why AI does well at the international math olympiad but doesn’t do well on frontier math, because the former is based on known math and the latter requires innovation of brand new math.
Current AI is not able to innovate brand new ideas, to my knowledge. I think it just excels at organizing existing knowledge. But even that is something AI would be good at since there are arguably 100+ million books and scientific papers written about medicine, biology, biomedical engineering, etc for it to learn from.
Also I’m assuming in 20 years we will have AI which is far more advanced, which I don’t think is controversial. We’ve gone from chat GPT-2 in 2019 to the programs we have now in 2025, which are vastly superior.
I don’t personally care whether what we have qualifies as AGI or not. What matters to me is whether it is competent at solving and managing medical problems. Whether it can pass a 3rd party metric or not is not important. Modern medical AI already outperform physicians in various metrics. Not only have studies shown this but patients are seeing it too. I uploaded the clinical notes from my MRI and the AI gave me a much easier to understand explanation than the neurologist did. Not only that, I asked it follow up questions about my symptoms and it explained why my symptoms are due to the findings of the MRI.
Also as far as Metroclopapril, something like 300,000 people die as a consequence of medical care each year in the US. Humans aren’t perfect either. I don’t agree with this mentality that unless we have perfect AI that can pass an AGI test that that means we should strictly rely on easily fatigued, burnt out humans that already kill 300k people a year.
AI nursing is better delivered by digital data systems than movie robots. Hospitals are dinosaurs. The insurance paid bill for my wifes’ recent hospitalization was $168,000 and I cannot figure out what they did for the money.
An LLM nurse could monitor and communicate with cell phone connected patients, residing in their own homes. A mid range hospital bed costs $1500. A relative can provide 24/7 patient contact. The patient has immediate access to the LLM nurse instead of constantly pushing an unanswered button. Questions about care could be addressed with the LLM.
So, I have hope for super intelligent nurse robots improving health care, but I believe the delivery will be through information systems not manikins in nurse suits.
Can’t they be both like the late Earl Warren?
Keep in mind that in a hospital, a robot nurse that works 24/7 would replace 4 nurses, so even if the robot costs 200k the hospital would still save a lot of money.
As far as the energy requirements for bipedal robots vs wheeled robots, I don’t know if that matters. Like I said, robots that have an internal battery pack that provides them energy while they swap out their rechargable battery pack should be easy to produce. So a bipedal robot that has to swap out its rechargeable battery pack 12x a day instead of a wheeled robot that swaps it out 4x a day shouldn’t be a big deal.
But yeah I think it’ll be easier for people to get help at home due to advances in biometric devices. However eventually if robots become affordable, then at home robots will become commonplace too which can provide health care along with other help.
Hospitals are already designed around the need to wheel stuff through them. Everything from medical equipment to food carts to patient beds. An R2D2 type footprint would be among the smaller things taking up space in a medical environment. C3P0 clomping down the hallway doesn’t offer much advantage.
In the 1960s I was a partner in a company that made cardiac monitoring systems for ambulatory patients. It was a NASA spin off. Sales based on patient benefit were minimal. But our salesman came up with an idea: sell our systems to a leasing company. They could be leased to hospitals who could make a profit by charging patients for time on the machine - 24/7. Selling became easy.
I don’t see that hospitals can save money on robots. People are easy come easy go while robots are a capital investment. But hospitals can make money on robots. Robots are billing machines that can have all of their time scheduled and all of it billed to insurance companies. This can even be optimized by software.
A full staff of attractive super capable LLMs meeting your every need is a pleasing fantasy but a bunch of point of sale terminals in nurses clothing is more likely.
Yes but for an initial capital investment, a hospital can buy the robot outright. Why would a hospital lease a robot for $50 an hour when it can buy it outright for $200,000? After it buys it outright, it can still charge insurance on an hourly basis. But now that insurance companies are merging with hospitals, the hospital would just be charging its parent company.
Also bipedal robots will probably not cost anywhere near 200k. Some consumer models that are being worked on in 2025 are only 20k or less. The price will go down as manufacturing at scale increases the same way car prices went down with mass adoption.
Considering that a nurse that works ~36-48 hours a week can cost 100k, and a robot working 24/7 will replace 4 (or more) of them, the financial incentives are there.
Plus when the robots reduce patient mortality and morbidity due to having stronger sensory systems than a human, having more knowledge, having more arms with more dexterity, being able to multitask better, being able to notice subtle signs of illness or worsening symptoms, etc the patients will start wanting them too.
I’ve had hypoglycemia in the past. In the hours before an episode I give off small symptoms that a nurse who is overworked and doesn’t know my subtle symptoms like I do would not notice. People who have seizures have dogs who can pick up signals of a seizure before it happens. Robot nurses would be the same way. They’d be designed with very advanced visual, auditory, olfactory and tactile sensors so it can detect minor changes that signify a health crisis is going to happen within a few minutes to a few hours. Signals a human nurse can’t pick up on.
It depends on the setting.
Bipedal medical robots would serve little purpose in hospital settings. Indeed everything is already built for wheeled objects to get around and, a few fetishists aside, most humans anthropomorphize and experience empathy based on facial features and voice more than on legs.
Nurses for home visits would have different mobility requirements. Stairs and cluttered floors are more probable. Legs, not necessarily as pairs, become the better choice.
The design is in all ways very medical tasks and circumstances dependent.
There is no reason to modify the current Da Vinci form factor for surgical robots when they function autonomously. A robot whose function includes helping someone feel less afraid or overwhelmed and possibly obtaining a good history and exam OTOH will have a different patient forward interface.
AI doesn’t sift through anything, much less something as broad as human knowledge. The models are trained on whatever they are fed, factual or otherwise. Not only do they not create knowledge, they don’t know what knowledge is. There is no “understanding” going on at all.
No argument from me.
We don’t currently have fully self-driving cars, no matter what Apartheid Clyde might say. We don’t have a timeline for actually having them either. I’m of the opinion that driving a car is a whole lot easier than being a doctor. We let 16 year olds drive cars. I don’t want one poke around me with needles for shits and grins. AI will improve and will continue to get better at things that it is good at. Understanding is not currently one of those things.
Using AI to give probabilities of specific diagnoses based on image and object recognition is in fact a wonderful use. With that said, it is simply giving probabilities, not understanding what the actual problem is. If AI says that it’s 61% likely that I have a large complex cyst, I’m not ready to be cut open until a human confirms that diagnosis. Again, I have no issue with using AI as a tool for medical practitioners.
As noted earlier, driving is easier than being a doctor, yet full self-driving is still far more dangerous than a human at this time. Even Elon’s own metric of what he considers safe enough to lose steering wheels is several orders of magnitude beyond where his vehicles are today. My doctor or nurse can certainly make mistakes, but they are that, mistakes. AI is still prone to hallucinations and I encounter them multiple times on a daily basis. AI doesn’t get the syringe halfway into the bottle and then realize that they grabbed the wrong medicine. It was planning on injecting that incorrect medicine the entire time.
You want to stick an iPad in the hand of a nurse or doctor with all sorts of data science based tools on them, I’m fully on board. You want a conversational robot to help dementia patients get through the day, cool! Doctor wants to run my MRI through a Convolutional Neural Network model to see if it finds something the radiologist missed, awesome! Replacing the humans that are doing things that can have a negative impact on my health, I’ll let the robot visit your room instead of mine.
I think something like this would be key.
Recently, my mom took a fall, and hurt herself badly enough that she needed lots of help doing things like getting out of bed, or out of her chair, and going to the bathroom. We had a couple of personal support workers helping her, but even then, it was a huge effort to move mom, and she was obviously in a lot of pain. The fundamental problem seemed to me that they only had two hands, and so had limited areas of contact while trying to lift her.
I could see designing some kind of octopus-like manipulator, that provides far better support for moving people in such situations. More points of contact, and more flexible so the point of contact is over a much greater area, reducing stress on any one joint or other part of the person’s anatomy.
There are sling&pulley systems designed to help move patients. I first saw one advertised as a way to reduce workers comp claims (back injury from lifting people) but they also claimed to be more comfortable, for this reason.
Something like this, perhaps?:
Stranger
I was thinking more pillows and fewer claws
Less snarling too. So not like the malfunctioning (but bipedal) robot at the opening of the original Robocop.
“Ladies and gentlemen, let me introduce you to the latest in robotic biomedical support technology. Emergency Droid Series Two-Oh-Nine is a self-sufficient medical nursing robot. Two-Oh-Nine is currently programmed for hospital pacification but that is only the beginning; after a successful shift in Delta City Hospital, we can expect Two-Oh-Nine to become the hot nursing support product for the next decade. We need a patient subject; Mr. @LSLGuy, would you come over and give us a hand, please?”
Stranger
I think I’ll save some time and just dive through the window here in the 30th story boardroom. It’ll hurt less that way; I’m sure of it.