What’s The Reality Behind The Humanoid Robot Hype?

Humanoid robots are all over social media, doing everything from dancing to serving drinks. But are they really going to show up in our lives?

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:02Flora Lichtman?

  Hey, I'm Flora Lichtman and you're listening to Science Friday.

• 0:12Flora Lichtman?

  That is the sound of a 150-pound robot breakdancing. Pretty well, actually. I'm sure you've seen these clips of humanoid robots dancing and doing cartwheels, putting clothes in a washing machine, serving drinks. And you've probably also seen headlines from tech CEOs telling us to prepare for the forthcoming humanoid army that is going to make all of our lives totally different and better. But what is real? Where are we with this technology? And are these humanoid robots really ready to take washing the dishes off our plates or to work beside us in warehouses? Here with a perspective is tech journalist James Vincent, who became an expert on this subject.

• 0:54Flora Lichtman?

  He toured humanoid robot factories, talked to CEOs, and rubbed shoulders with humanoids for a feature story he wrote for Harper's Magazine. James, welcome to Science Friday.

• 1:05Speaker 01

  Wonderful to be here. Thank you, Flora.

• 1:07Flora Lichtman?

  Okay, we've all seen these videos. First of all, should we take them at face value?

• 1:12Speaker 01

  No. I mean, the short answer there is definitely not. Not all the time. A lot of these videos that these companies are putting out, they are selectively edited or they're selectively presented. And the great phrase that I heard used by Nvidia's Director of Robotics was the blind gymnast phenomenon. It refers to the fact that you can see a robot backflipping perfectly, doing parkour, doing all these fantastic moves. But in a sense, it doesn't know what it's doing. It can't adapt to the environment it's in. It's like a blind gymnast. So even when you see these humanoid robots creating these sort of fantastic spectacles, that might be the limitation of their abilities. The other thing is that the robot might also be teleoperated,

• 1:56Speaker 01

  which means that it might be being controlled remotely by a human using a VR headset and a couple of controllers. So yes, there has been phenomenal progress in humanoid robotics. And we've definitely seen breakthroughs that we haven't had for 10 years. But a lot of what these companies are doing is trying to attract investment. And that means that they do show you only the highlights reel.

• 2:17Flora Lichtman?

  Hmm. It does feel like humanoids are having a moment. Why now?

• 2:23Speaker 01

  I think the reason we're seeing all this progress now is there's a lot of hope being transferred from the world of AI. They're calling it the chat GPT moment. The basic hope is that the same trajectory that chat GPT took, where you got this huge step up in quality and progress from large language models, will apply to robotics. Now, the idea there is that what helped large language models really accelerate in the quality of their output was that they had a lot more data put into them. They had a lot more training data and they had a lot more computing power. And that allowed them to see a lot more patterns within the text they are analyzing. Robotics companies are hoping that there'll be a similar effect happening with robots. And what they're doing to help with that

• 3:06Speaker 01

  is they're trying to put robots in homes so they can collect lots of data about how these robots might operate in novel or unforeseen environments. And then they can use that to try and have a chat GPT moment. So they get all this data, they plug it all in, and suddenly the robot knows how to do the dishes. Whether or not that's going to pan out exactly as they hope, we don't know yet.

• 3:26Flora Lichtman?

  Well, that's what I wanted to know, because it it seems like hugely different in terms of the amount of training data you could collect from robots in the house versus all of text of all time, right? Like, do we actually have the data sets at the scale that we need for physicality or for behavior?

• 3:47Speaker 01

  I spoke to a lot of roboticists, I spoke to a lot of engineers, some, you know, inside these companies, some outside them. And it's one of those situations where there really is no clear answer. Some people say, yeah, the mechanics of scale will work in robotics as they did in text. They're just going to need more data and more sort of finely grained data involving things like, you know, touch sensor data, lots of GPS stuff, all of this extra additional information streams you'd expect for the physical world. And some of them say, look, these things are just not transferable in the same way. And they're really, really skeptical that there is going to be this chat GPT moment. You get companies like One X saying, we're going to put a robot butler inside your home in the next couple of years. Those robot butlers, they're going to be controlled by humans.

• 4:31Speaker 01

  Once those robots are in there collecting the data, suddenly they'll be able to train AI to do the work for them. But we really don't know if that's going to happen.

• 4:39Flora Lichtman?

  Hmm. What about robots in factories?

• 4:42Speaker 01

  That's a little bit more of a reasonable proposition. So the thing with robots in homes is not only is the form of work much more varied in terms of what the robots are being expected to do, but there's a lot of issues to do with privacy and safety. You know, these robots are not light, you know? If one of them topples over, if one of them has a little bit of an unforeseen technical error and it falls and say squashes a pet or something like this, then there's going to be all sorts of issues and people are going to say, we can't have these in the home. The easier thing with putting robots in factories is the type of work that people are currently training them to do is a little bit more constrained. It's mostly about moving things from A to B. It's doing logistics works, taking pallets and boxes of goods from one place to another.

• 5:26Speaker 01

  So safety is less of an issue, privacy is less of an issue, and the complexity of the work is less of an issue. So if these humanoids take off anywhere first, it will be in warehouses.

• 5:36Flora Lichtman?

  I think people might wonder, why do you need a humanoid to do that work? Like, why not just build a smarter, better conveyor belt?

• 5:43Speaker 01

  Yeah, yeah. Again, so this is one of these questions where there's lots of different answers to it. The one that robotics companies themselves cite most often is the world as we know it, the environment as we know it, is built for the human form factor, as it were. You know, we have handles, we have levers, we have stairs, these are built for hands, they're built for feet. So rather than adopting the environment to the robot, we adopt the robot to the environment. So let's make them human. If you're working in a warehouse, you're doing things like you're squeezing down shelves and tight passageways and you need to lift heavy objects from low to high or high to low. And actually, the human form factor is pretty good for that. Because of the way our bodies work, we're very good at like balancing things out.

• 6:28Speaker 01

  Now, the other argument, and I think this is persuasive, although a lot of companies wouldn't necessarily agree to this. Is that there is a mystique to building a human robot and that attracts people on a personal level and it attracts investment on a wider level as well.

• 6:44Flora Lichtman?

  The psychology of it.

• 6:45Speaker 01

  The psychology. Yeah, I don't want to sort of psychoanalyze these people too much. I don't feel that's my place. Twist my arm, I'll give it a go. There is definitely an attraction for these people to build something grand and mysterious. And, you know, I spent some of this piece that I wrote for Harper's talking about the the history, as it were, of recreating the human form. And you can go back to ancient Egypt and you can see how like the first, and we use this term very loosely, the first automatons were sort of statues of gods that could be operated by string. And so there's something, I think, in the human psychology that wants us to recreate our own form for the same reason that, you know, we imagine the gods in our form. We, you know, we want to perpetuate our image throughout the universe.

• 7:29Speaker 01

  I think that's maybe a subconscious part of why these companies and people are doing what they're doing. But I think also a human form is just very impressive. The same reason your listeners may have seen videos of these robots and gone, wow, gosh, what's going on there? That's quite scary. An investor might see that and go, wow, gosh, I should put my money in that. So I think that's another reason that they pick humans as a form factor.

• 7:53Flora Lichtman?

  Where are we in terms of deployment? Are these robots out there in the world?

• 7:57Speaker 01

  They're out there in very, very small numbers. To give you some perspective on this, there's about half a million industrial robots installed each year in factories across the globe. And when I say industrial robots, I mean those big robotic arms you see in factories lifting up cars, doing spot welds and paint jobs. In the US, in terms of US companies, I would say there's maybe a dozen or so humanoids in deployment in different factories. It's hard to get a good figure on this because the companies themselves don't talk about it that much, but I would say it's on that order of magnitude, certainly.

• 8:32Flora Lichtman?

  But that's so different from what we're hearing from people like Elon Musk and with his Optimus robot and, you know, the New York Times got ahold of documents from Amazon with a plan showing that they had a plan to replace half a million warehouse staff with robots in the coming decade. I mean, we seem very far away from that.

• 8:51Speaker 01

  I think in a lot of these reports, there's a lot of elision between different types of robots and different types of deployments. So for example, with the New York Times story about Amazon, a lot of the robots, the vast, vast, vast majority of those robots are going to be what are called AMR, autonomous mobile robots. These are the ones that look like little coffee tables on wheels and they pick up shelves and they move them around them. These are not humanoids, they are nothing near it. And those are robots that have been under development for decades and they've taken a long time to get to be as useful as they are. Then when we talk about Tesla and Optimus, Musk may make claims about how many robots he's using, but you know, these are trial deployments. These are not necessarily doing useful work. I spoke to one of the leaders in this space, Agility Robotics, and other called Apptronic.

• 9:37Speaker 01

  They are deploying a very small number, as I say, you know, something like four or five at a time in these warehouses and in these factories. And they're not necessarily doing the most useful work. Again, these companies are very opaque about it and it's in their interest and it's in the interest of their clients to boast about how fantastically futuristic and advanced this technology is, but it is still under trial. It's getting better, it's on its way, but these are not coming in to take anyone's job this year, next year. It's going to be years out from that.

• 10:08Flora Lichtman?

  James, you rubbed shoulders with some of these humanoids. What was it like to be in the room with them? You know, what was your response to them?

• 10:16Speaker 01

  I mean, my response is sort of gleeful. Like, I get why people are excited about these. You know, I'm a tech guy, I'm a nerd. I love looking at this stuff. Looking at it up close is incredible to me. I tell you what, okay, so the, one of the first times I saw one of these robots was, it was an Apollo unit by this company, Apptronic, and I was giving it a shove test. Because you know, you see people kicking and shoving these robots to see how their stability is like, well, I was like, I want to try this. And in a way, it felt to me like a way of seeing what's real. How do you see what's real in life? You give it a kick, you see if it kicks you back. So.

• 10:52Flora Lichtman?

  Okay, maybe, sure.

• 10:54Speaker 01

  I tried to kick it and they said, no, you can't do that. And they gave me like a, they gave me like a broom handle with a bit of foam taped on the end. They said, give it a poke with this instead. And I thought, okay, we're doing this the scientific way. I got it. Anyway, and I'm, I'm shoving it about and I'm trying to knock it down and I'm being very cautious with it. And then they're like, no, no, you can, you can give it a harder shove than that. And I give it a real, oh, thrust right in the chest and it sort of staggers backwards and it, it just, it looks so human. It loses its balance and then it swings its arms forward as if it's like reaching out to me for help and it regains its balance and then trots and stands right in front of me in the middle of the room. And honestly, I'm just sort of awestruck by it. I think, my God, I mean, I've been, I've been writing about this stuff for years and they have gotten so much better than they were before.

• 11:39Speaker 01

  They're a long way off yet, but like, I just think it's fascinating stuff. I get why people are obsessed with it. I get why people are so willing to believe that they're about to, um, you know, take over all the hard work from us because they really are fantastic creations.

• 11:52Flora Lichtman?

  James Vincent is a journalist and author. His piece in Harper's is titled, "Kicking Robots: Humanoids and the Tech Industry Hype." James, thanks for joining us today.

• 12:02Speaker 01

  Thanks so much for having me.

• 12:03Flora Lichtman?

  This episode was produced by Dee Peterschmidt. Thank you all for listening and we'll see you next time. I'm Flora Lichtman.