It does mean that the human shape is capable of doing those jobs and is currently the only form factor empirically shown to have that capability.
There are a plethora of tools available that can be used by humans to do a plethora of tasks. That is one big reason why humans are still in the workplace and not completely replaced by single purpose machines. A humanoid robot in principle should be able to use all these tools and be equally as multi-functional.
It also means that the human brain is capable of doing those jobs and is currently the only “operating system” empirically shown to have that capability. There’s no reason to assume it’s the shape that matters, rather than the brain.
Whenever we have these discussions, I am reminded of this particular WSJ article about the immense difficulties involved in automating the task of….straightening candlewicks. A job that would appear to be easily automated, but still isn’t. And the reasons it hasn’t been automated have nothing to do with the absence of a robot with a humanoid form, but because we don’t yet have robots with the brains and nervous systems to do these jobs. To sense when a wick is off, to have the delicacy of “touch” to fish it out of wax, etc.:
Why It’s So Hard for a Robot to Straighten a Candle Wick - WSJ
Again, nothing to do with a humanoid form. This job physically could be done by a hand (or series of hands) or any other type of gripper mounted on a typical kuka-style arm above the candle conveyor belt. But we don’t have the brains and sensory systems to make that work, so it’s still done by a human. If we had the brains and sensory systems to make it work, it would be done by a robot - and almost certainly not by a humanoid robot.
Beating a dead horse here but can’t resist a final comment. The reason for humanoid robots is not simply to straighten wicks. It is to create a single machine capable of a multiplicity of functions similar to the flexibility and adaptability of humans. There are many who believe there is great economic value in such machines, particularly if manufacturing processes become more adaptable in real time to tech advances. We know the human form can do many tasks, we have already a plethora of tools that the human form can use to do a plethora of tasks, and we have an extensive physical infrastructure already built for the human form.
Connect the dots.
Sure. All of that is true. But it doesn’t point to humanoid robots being necessary, or even especially useful, in many environments. After all, we know that robot forms can do many tasks, and we have extensive physical infrastructure built for robot forms as well, and that robots are easily integrated into many (if not most) environments.
The key is that for many (most?) jobs - especially industrial jobs - there isn’t a need for the thing doing the job to be capable of a multiplicity of functions. Many industrial jobs are like the candle-wick straightener, or the poultry processor. The company needs a person/robot to straighten the wicks or carve the chicken. They need a person/robot to do that all the time, and not do anything else. They will always need a person/robot to do that task, and not anything else. So being “capable of a multiplicity of functions” adds virtually no value to the robot in that role, while optimizing the form to the job at hand does.
It’s instructive to look at our efforts to automate driving. Cars and other vehicles are built for the human form. They are optimized for the human form. A human can drive a car, and (nearly) all driving is done by people in a humanoid form. It requires a lot of hardware (cameras, chips) for a car to be automated, which hardware can’t then be used for almost anything else. And there are literally a billion light vehicles in the world (to say nothing of other types of vehicles) with human drivers that could be replaced, compared to the very tiny tiny number of bespoke self-driving cars that will be or could be built in the next decade or so.
But no one’s pivoting to a car-driving humanoid robot. Everyone’s still anticipating that the car will have the ability to drive. Because the car will always need a driver whenever it is being used, so it makes more sense to have that driver be purpose-built into the car (the cameras are in the frame, the “muscles” to turn the wheel are in the body of the car, etc.) rather than replicate a human being manipulating the vehicle.
But that is only because you have limited yourself to a 20th century mindset. The new manufacturing paradigms given terms like “unboxed”, “industry 5.0”, “parallel manufacturing” all emphasize adaptability, flexibility, and real-time change as fundamental objectives. They are all about multi-functionality in robotics and designing robots to work together with humans in an integrated workspace. Humanoid bots are designed for this.
Even non-human intelligences agree–-from Google AI:
Industry 5.0 emphasizes combining human creativity and problem-solving with the precision and endurance of robots to enhance efficiency and enable personalized production. Humanoid robots, which can navigate and interact in dynamic human environments, are ideal partners in this model, capable of performing a wide range of tasks, including those that require more dexterity and adaptability than traditional industrial robots.
I agree since the car has but a single purpose. You want humanoid robots for multipurpose use in human centric environments.
Let’s consider home building and renovations. Homes are built for humans so they often have sharp turns, narrow doorways, steps, stairs, etc. Anyone who has helped move a mattress into a 2nd floor bedroom knows this. Home construction requires workers to do a multiplicity of tasks often in the same day, from nailing, dry walling, sanding, painting, cutting, gluing in an environment with all sorts of obstructions on the ground. Seems like a perfect environment for a multipurpose humanoid robot that can use human centric tools.
I mean, that’s a lot of jargon - and perhaps in some situations those terms might be applicable. But if you’re making candles, or processing chickens, or picking apples, or most types of industrial tasks that are currently done by human hands, those concepts don’t really have as much relevance.
And you can make a multipurpose robot that doesn’t have a humanoid form. A rectangular cabinet on wheels or on a quadrapedal base with four kuka arms could do a ton of different things.
But the car is the tool, in this case - not the analog to the human. To use your construction example, consider the robot that’s walking around using a power drill and a hammer and a saw. All those tools are still single-purpose tools. You have the multi-purpose robot using them, rather than making a robotic drill that can drill by itself (like the giant welding robots in a car plant are basically just huge welders that do the welding themselves). The idea is that the robot serves as the analog to the human - a user that can take up a tool, use it for the given job, and then move to a different tool to do a different job.
But we’re not going to do that with cars. We’re not planning on having a robot that can drive the car and then get out of the car and do other things. The robot is locked into the car - it’s a single-purpose robot that can only drive, rather than a multi-purpose robot that can drive the car to the store and then get out and do the shopping. Rather than make a robot that can use a tool, we make a single-purpose robot that is the tool.
And we’re doing that even though, quite frankly, driving is actually a pretty good use case for a humanoid robot. The environment (interior of a car) is optimized for human form, it can’t be navigated by wheels or a quadrapedal body, manipulating the environment doesn’t require too much haptic sensitivity or fine dextrous movements with a hand, and there’s literally 1.5 billion cars already on the road that could be operated by a humanoid driver. It’s a better candidate in many ways than all those industrial jobs that we’ve been talking about on this thread. But we’re not taking that approach.
BTW, I completely agree that there are some situations where a multi-purpose robot would be vastly superior than a robot optimized for a single task. You mentioned construction, and I mentioned a home “valet” robot upthread. Jobs like general janitorial services and the like often require a person to do lots and lots of very different tasks. But those are different kinds of jobs than typical industrial jobs, which far more often involve humans doing specialized and repetive work rather than moving between many very different jobs during a working day.
Exactly! Every year AI is getting better and robotics cheaper. Why buy the single purpose machine when the multipurpose robot can do more and be more cost effective?
Humanoid construction robot finishes tasks at site A. Company needs it at site B, 10 miles away. Humanoid construction robot takes the company self-driving car, which it can use because it has a humanoid shape, and uses that car as a transportation tool to get to where it is needed.
Humanoid robot uses human tool.
Because in many cases I don’t need the multipurpose robot to do anything other than the one thing I need done, and it’s less cost effective if the robot has the tradeoffs that come with being able to things other than the thing I need done.
If I need a robot to straighten candle wicks, I don’t need it to have legs. I don’t need to pay for it to have legs. I don’t need it to have to expend energy/battery time or compute power on balancing on legs. I don’t need it to have hands to pick up different tools, since it will just be using the same tool all the time every day. I don’t need it to be slower because it uses hand + tool rather than integrated tool. Etc.
It’s the same reason I have a power drill/screwdriver instead of just using my Leatherman multi-tool for all my screwdriving needs. The former is a specialized tool, and does vastly fewer things - but it does the things it does better than the multi-purpose tool.
But why does the company have a self-driving car? Why does the company have two robots - the humanoid robot and the car robot - instead of just having the humanoid robot drive a standard car?
The reason is the very one we’re talking about. It’s better to have the single-purpose robot (the self-driving car) than to try to have an “everything robot” use an ordinary tool. Make the tool into a specialized robot that does only one thing but does it vastly better than a multi-purpose robot can.
Once they get the engineering figured out, using mechanical legs won’t be much more expensive than using wheels. The mechanical stuff isn’t that expensive once you can mass produce the components. This isn’t the 20th century. Technological progress is happening and costs are dropping.
Wow, you do get hung up on your analogies. Forget the self-driving car. The humanoid construction robot calls Uber instead or takes a bus, which again it can do because of its humanoid shape.
I’m all for making single use tools better or smarter because doing so increases the functionality (and therefore value) of the entities using that tool, whether human or humanoid robot.
Seems unlikely. Legs = more parts, and more parts equals more expense…and more importantly, more maintenance. If you look at Amazon’s experience (one of the biggest purchasers of robots around), they’re constantly on a quest to make their robots as mechanically simple as possible in order to reduce their maintenance and down time.
For my candle wick straightening robot, I don’t need it to move around at all. I’ve got an assembly line, and so I can just stick the robot on a fixed point and have it work. I don’t need it to have a hand (an incredibly complicated piece of machinery) to hold a tool - I’m far better off if the tool is just part of the robot. Just like we’re not going to replace all those huge welding robots with humanoid robots holding hand-held welders - the single-purpose robot is just going to be more efficient.
But that analogy is the point. The reason we’re building self driving vehicles (like a self-driving Uber or a self-driving bus) is because a robot that is optimized for a single purpose (like the robots that are incorporated into the Uber car or the bus) are expected to be better at doing that job than a humanoid robot using a dumb version of the tool (a humanoid robot driving an ordinary Corolla or a standard city bus). By making a specialized version of the robot, it has access to more cameras and more integrated control of the vehicle rather than just using two “eyes” and their hands to move a steering wheel and their feet to move the pedals. The car/bus will always need a driver, so there’s absolutely no benefit to having the robot driving those vehicles be in a humanoid form sitting in a chair behind a steering wheel - instead, the robot is built as a single-purpose device and incorporated into the vehicle itself.
If more parts increases functionality, then it becomes cost/benefit calculation. Do you have plans to switch from your expensive multifunction cell phone to a much cheaper flip phone?
It’s your point, not mine. You are arguing with your own straw man. My point is that the humanoid construction robot can use the self-driving car as a transportation tool so that it can reach a location to do all kinds of other tasks. The car is a tool. Making it self-driving makes it easier for humanoid robots to get to where they have to go. Any tool or tool improvement that helps humans usually also increases the functionality of humanoid robots.
Nope - because I derive benefit from the fact that my phone is multi-function.
If I only used my phone to make phone calls, though, I wouldn’t have a smart phone.
Since most industrial jobs fall into the bucket of “one job that always has to be done, all the time,” the equipment being used for that job doesn’t benefit from being able to do anything else.
But why is the robot driving the car single-purpose? Again, that’s the point. Why are there two robots (one driving the car and one riding in the car) rather than a single robot? Or even if you have two robots, why is the robot that is driving the car a specialized single purpose robot, rather than a second humanoid robot that can do something else if it’s not actively driving the car (such as cleaning the car)?
It’s an excellent illustration of a circumstance where using a single-purpose robot might be much better than having a humanoid robot try to do the same job in the same way a human being would. Right now, Ubers are driven by people. An environment made for humanoids, used by humanoids. But no one’s trying to make a humanoid robot to drive the Ubers - instead, we’re making single-purpose specialized non-humanoid robots integrated into a car. That way these special robots have many more cameras (not just two “eyes”), integrated control systems into the body of the car (not just feet and hands on pedals and wheels), and can get information directly from the car itself (not just eyes watching a dashboard). It’s an example of how even the company that’s in the business of making humanoid robots (Tesla) recognizes that a single-purpose robot can be much better, even for an application that today is almost entirely done by humans.
Kind of flogging a deceased equine, but you are correct. Having a multipurpose robot to do only one or two things is a waste. But even if you had a multipurpose robot, it wouldn’t need to be humanoid. If we could perfect a robotic hand comparable to a human hand, you could put that on any form you wish. Appendages tuck away when not needed, and deploy when needed. You could have four hands on such a machine, if you wanted.
The versatility is not in the humanoid form, but in the manipulators and any AI onboard. You don’t need Bicentennial Man.
One possibility, approaching certainty in the long run, is that houses and stores and other place of life and work will increasingly be designed and built to make it much much easier to use robotic and dim-witted AI’s. The dance between existing reality and robotics is absurdly difficult.
It’s almost like this thread doesn’t see the very empirical evidence that specialization is all around us
Probably because of the notion of tradeoffs: it’s very unlikely that optimizing along one dimension also optimizes among several other dimensions.
Yep - there’s a definite tradeoff between specialization and versatility. If we are able to use AI to solve some of the problems that have prevented automation of various jobs, it is likely that AI will end up being deployed in both types of use cases: better specialized robots and better (or new) versatile robots.
What’s likely, though, is that jobs that are currently done by a human doing specialized work (like our candle wick straightener or poultry processor) are more likely to be occupied by specialized robots. Jobs that involve one person doing lots of versatile tasks (like a grocery store clerk, who might have to restock and clean up spills and answer questions about where goods are located and so on) are more likely to be done by more versatile robots. And jobs that involve environments that are difficult for robots to traverse (jobsites with steps and stairs, or that are especially cramped or old) are better candidates for humanoid robots.
But a lot of discussion of humanoid robots seems to talk about their use in factories. Those kinds of industrial jobs and spaces, though, are the least likely to be especially well-suited for humanoid robots. Many of the jobs in them are specialized and the environments are already designed for non-humanoid robots to do well in (no steps or stairs). You’re less likely to see a humanoid robot standing by the assembly line, and more likely to see the AI-powered automation physically incorporated into the line itself like we see here:
High Speed Robotic Appliance Drum Manufacturing Line
The versatility of humanoid robots might be useful in retail shops (like a grocery store or NYC bodega), household tasks, or outdoor jobs like construction or package delivery.
I think one reason that discussions of humanoid robots seem to get situated in imagining factory jobs is partially inertia - that’s where most robots are now, so it seems natural to think that’s where humanoid robots would go, even though those are the environments and jobs that (on average) least need a humanoid form.
But I think it’s also a bit of…wishcasting, maybe? Or wanting to see a clear path to adoption? These are going to be expensive devices at first, and will require some non-trivial amount of “on the job training” as part of the initial adoption path. Industrial users seem better suited to doing that than, say, your local Kwik-E-Mart. But that runs into the reality that a humanoid robot brings useful advantages to the Kwik-E-Mart job that it doesn’t have in an industrial milieu.
But the opposite of specialization is not universality, but rather generalization. Seems like, in all domains the ideal is some balance between specialization and generalization such that enough related functionalities are covered to be efficient, without becoming overly complex. With humanoid robots, there is the potential of solving locomotion in all human spaces once while providing specialized hands or whatever for specific functionality.
Nah. People just think “Wow, wouldn’t it be cool if a robot could do everything a person could do!”
The reality it that it will be a very long time before they can do everything because there are so many simultaneous things we’re good at, and our meat bags are well designed to operate all of them simultaneously.
Not that you can’t (eventually) bring those all together in a mimed polyalloy (Terminator 2) or, more likely, a string of metal bits and screws and cotter pins and relays and solenoids, but that’s got to be farther away than any reasonably time horizon. Oh, and couple it with a brain with a quadrillion synapses all focused on the task at hand.
We are really good at narrowing a task down to its simplest components so a robot can do it, but scaling it up to the entire suite all at once, not so much. Look how hard it’s been to teach a car to drive, a relatively simple task for humans. It even took a decade for Roomba to stop bouncing like a pinball and “make decisions” about obstacles and drive in a straight line.
Eventually? Sure. Not soon, at least not in the kind of “universal capability” that’s being talked about.
So what? What matters is the total package. Humans aren’t the fastest runners or swimmers and are not the strongest animal. Yet here we are on top of the food chain.
There was a time when people wore a watch and carried multiple specialized, single-function machines like a camera, music player, book reader, calculator, gps device. Now they just carry a single small rectangular multi-functional machine and just add apps for more function. That is the future. Heck, in many cases that is the present for those not wearing 20th century blinders.
That is the direction manufacturing is going. We have entered an era where software is where the value lies and hardware is just a commodity. We are fast developing the technology to create multifunctional robots that are just as efficient and cost effective as single-function specialty machines but are far more flexible and adaptable.
To put it another way, software is becoming more intelligent and the way one best uses that intelligence is to increase the number of functions that software can control. That’s why multifunctional robots are hip these days. Flexibility and adaptability is the mantra for future manufacturing. https://www.msn.com/en-in/money/news/manufacturing-is-moving-from-static-to-hyper-flexible-automation/ar-AA1Rwk4o
You are obsessed with your analogy. I guess that is a trait of good lawyers. It’s like asking why didn’t the US air force in WWII use jets. The technology isn’t available so it is currently not an option. Perhaps you are not aware, but folks often make phone calls using their car speakers and microphone. They do this by bringing their own phone, and then connecting it to the car. Two separate machines acting as one when needed.
The same would probably have been done if humanoid robots were available with human-level intelligence contemporaneously with computerized cars. A humanoid robot gets into the drivers seat, plugs into a USB port, and road trip!
The poultry processor reminded me of the fact that people used to physically pluck the feathers from a chicken carcass but now they use a specialized machine - that not only doesn’t look like a human, but doesn’t pluck feathers like one either. Clearly, the humanoid form was not optimized to pluck feathers.
