When will advanced robots become affordable?

This question of course assumes that advanced robots will ever become affordable, feel free to disagree.

The definition of advanced is also ambiguous. For me it means intelligence mirroring humans in most respects. I consider a robot with a human form who can perform most tasks a high school graduate can to be advanced.

When will we initially see them? I think Asimo is the closest thing we have right now, at a cost of $150,000 per month to rent, and not quite up to the standard of a high school graduate yet.

We’re nowhere close to creating a system that “mirrors human intelligence in most respects”. It’s not like we’re working on the problems of creating such an intelligence and it’s going to take a lot of time to work out the bugs, it’s that all the advances in machine intelligence that we’ve achieved over the years have been nothing like human intelligence.

We’re not likely to develop the prototypical “robot” of science fiction–that is, a mechanical man who does most of the things a human being does, but does not understand hu-mon emotions and is unable to use contractions. We’re never going to build a robot maid who can vacuum the rug and wash the dishes, instead we have self-driving vacuum cleaners and dishwashers.

Robots are terrible at the sorts of tasks a human high school graduate can be trained to do, and great at things no human being can do. So even when McDonalds is fully automated, you’re not going to see C3PO standing over a grill with a spatula. Instead you’ll see an automated assembly line that loads ingredients in at one end, and delivers bags of hamburgers and fries at the other end. You’re not going to see a robot chauffeur driving it’s owner to the mall, you’ll see a car that drives itself. You’re not going to see robot shopkeepers, instead you’ll pick stuff out of a catalog and it shows up at your house the next day.

And so on. The prototypical general-use mechanical man robot of science fiction will likely never exist for the foreseeable future, and even when robotics advances enough for them to exist they will be toys that you get just for fun rather than household helpers or workers.

Interesting stance, never is a long time and I’d be interested in hearing why you feel that way.

I believe he is correct. The reasons are fairly simple on the surface. Artificial intelligence driven machines including robots use very different strategies than people do for accomplishing tasks and that is a good thing. Special purpose robots can already do many things much better and faster than humans can because they are designed for that task in ways that use their strengths over humans (speed, accuracy, stamina, reliability etc.).

People already exist and there are plenty of us around. If you need to do a task that specifically requires something a human is good at, that is a task that a computer program or robot may not be optimal for (sales or teaching for example). In short, there is no need to build robots that replicate everything people are good at because that is an unnecessary duplication of effort even if we had any idea of how to do that in the first place (we don’t).

Instead, you can build robots that perform tasks that people aren’t especially good at by taking advantage of strategies that robots are very good at. There is a misunderstanding among the general public and even some experts that the human brain is just a biological version of a computer that is much more powerful than the computer on your desk but still mostly analogous. That isn’t true at all based on our current understanding.

The differences between today’s AI and the human brain aren’t just a matter of computing power - they are fundamentally different types of systems that each have their own strengths and weaknesses. If you scale up today’s computers 10, 100 or 1000 fold, all you will get is a machine that can run Windows extremely quickly with no other emergent properties until many other breakthroughs are made in AI theory. Many other major breakthroughs would also need to be made on the robotics side as well if you wanted to build a general purpose robot that can do some things much better than a person and many others not nearly as well but there is no need to do that when we can build special purpose robots to do some things extremely well and let people stick to what they are best at.

Try, for example, building a robotic hand/gripper than can pick up a steel bar weighing a hundred pounds and that can then turn around and pick up a chicken egg without breaking it. It’s quite difficult.

I’m not saying we’ll never have a mechanical man/robot pal. I’m saying that when we develop a computer system that has “an intelligence mirroring a human in most respects”, stuffing that system onto a chassis shaped like a human being is going to be the least useful thing you could do with it. So when we create a human-like intelligence, the only reason to make that human-like intelligence control a human-like mechanical body is for fun, to make a cool toy that’s fun to play with.

Putting a bunch of humanoid robots to work on an assembly line to replace the human workers is the sort of nonsense that will never happen. We already have nearly fully automated production lines, and none of the robotic assembly machines look anything like a mechanical man. We already have a robot that can vacuum your floor, and it doesn’t look anything like a mechanical woman who can hold a vacuum cleaner, it looks like a vacuum cleaner that runs itself.

The basic idea of the 50s-style mechanical man is that a robot like that could just slip into our already existing human infrastructure, literally taking the place of a human being and using our existing tools and machines and buildings and infrastruce in the same way existing human beings did. And the idea was, it would be a lot easier to build a general-purpose robotic brain and humanoid body that could operate any existing human tool than it would be to completely rebuild everything we own from scratch.

Except it turns out that building a robotic brain that’s pretty much like a hu-mon brain only without the contractions and emotions is really really hard. It’s much easier to build a system that can navigate around a human living room and cover everything than it is to build a robot brain that can figure out how to operate a human vacuum cleaner. It’s a lot easier to build a little cart on wheels that scoots around than it is to build a bipedal walking robot. It’s a lot easier to to put self-powered wheels on a vacuum cleaner than it is to build a robot hand that can grasp and manipulate a vacuum cleaner designed for human beings.

And so, if your desire is to have clean floors and you don’t want to vacuum them yourself, we already have a robot that can do that for you, and they’re not that expensive. But it has nothing like a brain that is similar to a hu-mon brain. It has a very simple suite of sensors that process input in a very simple way, compared to the complex system of vision and hearing and touch and proprioception that a human has. It has a dedicated built-in vacuuming system rather than something that is supposed to clamp on to an existing vacuum and use it.

And our ability to make effective single purpose machines of this sort is growing rapidly, while our ability to make hu-mon brain equivalent systems is going nowhere, and our ability to make robust bipedal walkers is going nearly nowhere, our ability to make robotic hands that work the same way and have the same flexibility as a human hand is going nearly nowhere.

We’re going to have cars that drive themselves long before we have mechanical men who can sit in the driver’s seat of a conventional car and drive it. We already have planes that fly themselves, are we going to replace those with mechanical men that sit in the pilot’s seat of conventional planes? For every application you can imagine a mechanical man taking the literal place of a human being and accomplishing pretty much the same task in the same way that a human being would, I can imagine a simpler and cheaper way to accomplish the same task in a better way by purpose-built automated systems.

To take one example, we often saw in 50s-style science fiction a robot cook, who would walk into the kitchen, pull out the ingredients, and whip up the food. Of course it would always be Mom who pushed the “make dinner” button on the robot, some things never change amirite? Except we already mostly have the technology to deliver the sort of meals that this robot was supposed to create. Why buy all the raw ingredients and stock your own kitchen and have the robot prepare it all when you can just call the take out place and have them deliver the food in 30 minutes? Or buy the meals already assembled and just microwave them when you’re hungry? It might seem to you to not be the same thing, but it really is. The goal is: I’m hungry, I want dinner, but I don’t want to make it myself. There are plenty of ways to accomplish this goal, but building a humanoid robot to stand in my kitchen and make everything is just about the least efficient method imaginable.

Same with washing the dishes. We already have machines that wash dishes, they’re in just about every home. We also have cheap methods for never really having to wash dishes in the first place, they’re called paper plates, or eating prepared food from a restaurant. If eating off of paper plates doesn’t seem much like a humanoid robot with a scrub-brush in its hand, that’s exactly correct. But throwing away your dishes instead of washing them is an already existing cheap solution to the problem of not wanting to wash dishes, and that humanoid robot is again the least efficient method imaginable.

If you want a robot pal then a humanoid robot with a humanlike brain makes sense. This is what I meant when I said such a system would be a toy. But if we can build a robot pal then we’re already living in a world where non-humanoid automated systems and embedded robotics accomplish almost all work that formerly required human beings.

I disagree; you assume that using robots to replace efforts that humans are good at is undesirable. I’m not sure this is true and given the uncertainties, I’d lean towards a personal opinion that it isn’t.

There is great utility beyond fun and pleasure in having a humanoid robot. There’s a competition going onfor the best emergency response robot design. You’ll notice that all of the contestants have humanoid forms, as mirroring our form allows robots to interact in our environment the same way we do. In doing so, they can also become general-purpose machines in a way that your examples cannot. Perhaps in the future, the 1950s idea of generalization will become more valuable again.

But note that these humanoid robots also don’t have anything like a humanoid control system. And they don’t actually work, at present.

The notion that a humanoid body is the best way to navigate a landscape designed for use by human beings seems specious to me. Why would they be bipedal, for one thing?

I will grant you that a humanoid body seems like a more tractable problem than a humanlike brain. At least the humanoid body just requires throwing a ton of engineers and hundreds of billions of dollars at the problem.