Humanoid Robot Startups Are Hot. This AI Expert Cuts Through the Hype.

0:00Isn't home where we all want to be? Reba here for Realtor.com. The pros number one most trusted app. Finding a home is like dating. You're searching for the one. With over 500 ,000 new listings every month, you can find the one today. Download the Realtor.com app because you're nearly home. Make it real with Realtor.com. Pros, number one most trusted app based on August 2025 proprietary survey. Over 500 ,000 new listings every month based on average new for sale and rental listings. July 2024 to June 2025. five. Ayanna Howard is the Dean at the Ohio State University College of Engineering, a former robotics researcher at NASA's Jet Propulsion Lab, a startup founder, and an advisor to our nation's highest officials on artificial intelligence and competitiveness.

0:46She has been working on humanoid robots and the software brains that power them for longer than almost anyone else in the U.S. These people-shaped bots could assist shorthanded warehouse crews by packing goods in delivery boxes. They could help take care of the elderly or even do your dishes. But think more C-3PO than R2-D2. Don't get technical with me. That makes her an expert at the intersection of two of the most hyped areas in tech right now. An insider with a unique view of what's happening in these fields. Humanoid robots is the next shiny penny that's out there that tech folks and VCs, venture capitalists can invest in.

1:26But I think it's also this logical, for those who aren't in robotics, it seems like the next logical step around artificial intelligence and generative AI is, oh, let's think about generative AI with a physical embodiment. And the founder of at least one robotics company, Zyrobotics, which developed personalized education tech for children. But Howard is also an outsider to the current wave of startups. and an academic who isn't obligated by her cap table to hype up whatever the latest thing is in these white-hot fields. If you're connected around an AI in a virtual environment and you don't have great bandwidth, you just wait a few seconds and your answer comes.

2:10In a robotics world, if you are losing a second of information, you have a robot that falls or hurts someone. We haven't really addressed that yet. As dean of engineering, Howard oversees the education of 12 ,000 students. She's also a member of the National AI Advisory Committee, which gives the president policy recommendations for keeping American AI competitive. And she sits on the board of design software company Autodesk to name a few of the titles on her resume. All of it shows that Howard has the ears of some of the most powerful people in the U.S. making decisions about the future of AI.

2:47I like to use healthcare because I think that's the holy grail. I'm more trusting that the AI can filter and provide a customized, personalized solution or decision than one person who is, you know, in, say, Atlanta. They're going to have a very different perspective than someone who's seeing patients in California or New York or Texas. I'm going to trust the AI a little bit more. From the Wall Street Journal, I'm Tim Higgins. And I'm Christopher Mims. This is Bold Names, where you'll hear from the leaders of the bold name companies featured in the pages of the Wall Street Journal. There are more than a dozen humanoid robot startups worldwide.

3:31And tech giants like Meta and NVIDIA are pouring money into the field. Today we ask, what's the deal with all these companies? And how likely are they to succeed?

3:45Ayanna Howard, thank you for joining us today. There's a lot that we could talk about, but I want to start with why you now and why you and I started talking in the first place, which like seemingly everything else on this show comes back to Elon Musk. Must be humanoids. That's right. So suddenly humanoids are the hottest thing in tech other than AI. There are, by my count, more than a dozen startups worldwide that have racked up significant investment to work on these. Some of these startups, it's in the hundreds of millions of dollars. So as in my estimation, the world expert on humanoid robots, who's not already employed by a humanoid robot company, I have to ask, why is this happening now?

4:30What is going on? There's a couple of things. One is that humanoid robots is the next shiny penny that's out there that tech folks and VCs, venture capitals can invest in. But I think it's also this logical, for those who aren't in robotics, it seems like the next logical step around artificial intelligence and generative AI is, oh, let's think about generative AI with a physical embodiment. And so if I'm thinking about the technology lineage, it's going to be humanoids because generative AI is about human knowledge, human knowledge in a body, human body must be a humanoid. So it's just a natural progression for a non-roboticist to think that way.

5:17shiny penny uh there's clearly a lot of hype in this space right now and probably no one's bigger at the hype than tesla's own techno king elon musk and he's recently saying that he thinks his car company uh can build 10 000 humanoid robots by the year's end a goal that he even he seems to concede they won't meet but if they get to thousands uh he'll be happy and then there's a a company called Figure, which is out recently talking about how it's going to deploy 100 ,000 robots in the next four years. So generally speaking, where is the technology currently? What are these kinds of humanoid robots capable of doing in reality in your estimation?

6:00If you think about the abilities of humanoids and what you would like them to do in theory, you'd like them to go from point A to point B, so navigation. They can do that pretty well, given that you can model the environment. Once they get to another location, you want them to be able to manipulate. So move objects, pick up things, build things, and then go back from B to A or B to C, the last mile, as an example. Now what they're really - The last feat. The last feat, at least in our human environments. The reality though is that just like with self-driving cars is that they can be very, very, very effective in limited situations, limited environments, limited tasks with humans not rampant around them.

6:50They can be positive, but that's not going to give you a billion-dollar valuation of a company. It feels like these robots are where autonomous cars were, I want to say 10 years ago, but maybe it's more like 20 or 30 years ago. Is that a fair analogy? I would say that they are at the point when all of the investment incurred, which I would say that was about 10 years ago, where startups were just popping up. They were talking about self-driving cars are going to be rampant within the next 24 months, including well-established car companies. And as roboticists, we're always like, do you remember that car that drove across America about 40 years ago?

7:38That's where we are with humanoid robotics. We've seen humanoid robotics 40, 50 years ago. Clunky, not well designed necessarily. And just like with self-driving cars, I always like to think about Hunger Games. There will be a winner at some point. We just don't know, one, who the winner is, and two, what is that application that is going to really set it apart from everyone else? Well, you're talking about kind of the important part of the startup ecosystem. The hype brings the money, right? And then the money maybe will allow somebody to make it through the end of battle. I'm curious, you kind of alluded these robots, humanoid robots are a lot like self-driving cars.

8:23Maybe driverless cars are robots in disguise, you know, more than meets the eyes. But I guess I wonder, 10 years ago when I was in a lot of these prototypes that were being used to raise money, these cars, it turns out maybe, you know, doing a prototype is easy, easier. Going into production and actually putting and deploying them in a Waymo-like fleet, a lot harder. And I guess I wonder, as you look at some of these humanoid robots out there now, are you seeing kind of some similarities in the prototyping, the hype part of it versus what is actually deployable? There is a big gap. And when I think about deployability, I also think about the cost.

9:10If I had$100 billion to build and deploy and I was giving it for free, I didn't need a customer or a client to purchase, then we have some opportunities. But that's not our reality. And so if you think about building a prototype that works in limited situation, pretty doable. Putting it now in anyone's home or in any warehouse, irrespective of where it is, that's almost impossible at this point. Just like with autonomous cars on safe streets where there's no other drivers, where there's no rain or snow, easy. As soon as you add in any dynamic, it becomes, oh, we need more data. We need more information.

9:51We need to think about how do we share information between cars, data, data, data, data, data collection. What's the business model, do you think? Are we going to have a Waymo of humanoid robots? Is this going to be a kind of a service that I pay for? Or am I going to have a robot that lives in my house and then walks my dog? So I would say this is the biggest issue is, is what is the, that the holy grail of an application? And I'm just going to give you my bias. If I think about where there are flows of need, one is in the warehouse. It's the packing, it's the sorting, it's the closing up the boxes, transporting the box to some other location.

10:37That is really around the warehouse. That is one area. If you can figure that out, unfortunately, that also means loss of jobs. But you now have more efficiency. Ideally, that efficiency and lower costs will transfer to the consumer. But we, of course, know that that's not going to happen. But that's one area. The other, I truly believe, is with respect to health care. One of the things that we know is that in most countries, including in the United States, there is not sufficient labor to address the older adults that want to live in their home. For example, the baby boomers, once they retire, have some resources.

11:17There are not enough individuals that are going to be working that want to go into the home to do the caretaking that's necessary for them to stay in their home. So that's an area that one, there's going to be a huge need. Two, we don't have the personnel in terms of people to fill it. And three, there are resources in terms of actually paying for that if it's at a reasonable cost. I really think that that's one of the holy grail applications that these robotic platforms can social impact, affordable, and address a great need that is coming across the world. Yeah. And just a footnote, what you said recently, I was speaking with the chief executive of Aptronic, which is an Austin, Texas based humanoid robot company, which spun out of a lab at the University of Texas at Austin.

12:10They've announced a deal with GXO, which is a big logistics company, which does things like uses humanoid robots in a Spanx distribution warehouse in Atlanta, Georgia. But he said his long-term goal is he wants to help people age in place. He was inspired by seeing family members take care of his own grandfather, and he thinks that's where we can ultimately go. Of course, he also said that the unique enabler now is he's like, look, we're trying to build the best robot body. We've been waiting for the right, quote unquote, brain to put in it. And the enabler is artificial intelligence. It's these new transformer architectures.

12:54Is there a there there? I mean, is the new software actually going to enable more capabilities for these robots? It will enable more capability. And that's this element of the natural progression from generative AI, large language models, is this element of how do we use data in a more efficient way? How do we share data to associate human input, human desires to actions, motor actions? And what does this look like? I think that there is some positive trajectory in that space. And just so you know, when you look at the use of transformer models in these applications, the videos are amazing, but it's still a really, really limited application.

13:40I will say one of the things that a lot of people aren't thinking about or considering is that, so it's one thing, if you're connected around an AI in a virtual environment and you don't have great bandwidth, you just wait a few seconds and your answer comes. In a robotics world, if you are losing a second of information, you have a robot that falls or hurts someone. And so I think that disconnect of do we have the infrastructure for doing this in orders of less than milliseconds, nanoseconds, and not losing any connectivity, we haven't really addressed that yet. Ayanna Howard just described the capability gap between the AI models we have today and what robot bodies need to operate safely.

14:30But she's not just speaking as an academic. For the last few years, she's been a member of the National AI Advisor Committee and helped advise former President Joe Biden on the subject, especially on bias. But now, how might AI policy change under a second Trump administration? When I think about bias, I think about it as all types of bias. It could be around ideology. It could be around gender, race. Bias is bias, and bias could be detrimental to Americans. And so I think when we think about extracting that one source of bias, the question is, is it an overarching term for all types of bias? I'm not quite sure.

15:15Stay with us.

15:24Well, speaking of AI, you've been on the National AI Advisory Committee, I think, since 2022. Your term is supposed to run through April of this year, but the White House just took down the official page on that committee. Is it still operating? Have you heard anything? What's going on? We are still, until we are notified otherwise, still acting, still meeting. With the new administration, the old AI executive order was rescinded. There's a new one. I'll just give you a line from it. To maintain this leadership in AI, we must develop AI systems that are free from ideological bias or engineered social agendas.

16:07And, you know, you've spoken with our colleagues before at the Journal about bias and AI. What do you make of that wording and what you interpret as the intent behind it? When I think about bias, I think about it as all types of bias. It could be around ideology. It could be around gender, race. Bias is bias, and bias could be detrimental to Americans. And so I think when we think about extracting that one source of bias, the question is, is it an overarching term for all types of bias? I'm not quite sure. I'd have to actually read the entire piece to find out. But I think what I've seen, though, in AI in general is that it doesn't always reflect the different populations, our different mindsets, and it doesn't give us a choice to query it based on our own viewpoints and say, hey, I think that framing is incorrect based on my lived experience.

17:14And the question is, is this going to be a limited definition of bias or not? That I don't know. Yeah, you've said in the past that one of the unique things about AI is that it can be built so that it's actually less biased than humans, which, of course, flies in the face of, you know, most of the coverage I read about AI, which is like, you know, it's bias means it's denying people health care or unjustly extending their prison sentences. But you seem like a glass half full person when it comes to AI bias. Could you expand on that? Yes. It's because when it's done right, it can bring in all different viewpoints around any type of whatever it is.

17:57And I like to use healthcare because I think that's the holy grail. So imagine an AI system that has learned about correct treatments based on gender. The mass of information, but you think about all with this link of gender. socioeconomics, age, are you childbearing or not? Do you have kids? Dot, dot, dot, dot, dot. I'm more trusting that the AI can filter and provide a customized, personalized solution or decision than one person who is in, say, Atlanta. They're going to have a very different perspective than someone who's seeing patients in California or New York or Texas. I'm going to trust the AI a little bit more.

18:42It could be less bias. I think the idea or the discussion around bias and AI really has become more of a popular topic because we see these chat models who have had some interesting hallucinations. And I think the makers of these chat models are really aiming to be accurate, right? I mean, they don't want to be embarrassed. But I think in the current political kind of climate, the real debate over kind of what is fact and the perception of fact is kind of hanging over some of this. And how does AI deal with that? I mean, when is a fact a fact and when is kind of it more than that? I would say this is a problem with our field of AI is because it's still a very human defined perspective of what is fact versus not fact, because it really is based on your lived reality.

19:39So I can say it's raining outside. The question is, is it raining in Columbus? Is it raining in California? Like, no, it's raining outside. So that's my fact, but it may not be your fact if you're somewhere where it's nice and sunny and 90 degrees. AI is not able to distinguish between, you know, is this your reality or is this someone else's reality? And that's a problem. The other thing is, is that typical people tend to trust what AI is doing. The fact that people call it hallucinations, what that basically means is at some point they realize that the AI has gone crazy, but they didn't realize it like the first incorrect output or the second incorrect output, which if you go back, you're like, oh, yeah, that was wrong.

20:21And that was wrong. And that was wrong. It's only when it becomes entirely literally like, oh my gosh, this makes absolutely no sense. Do people then say, oh, that's a hallucination? But no, it was actually a mistake. And the mistake happened somewhere down the lineage of question and answers. And what happens is people tend to follow the thread of thought of the AI without questioning it until it violates their fundamental beliefs around whatever it is. That's a problem. One of the things in driverless cars that's been out there, the companies behind it have been very concerned that if there's a crash and somebody was hurt or died, that this would turn people off to the technology more broadly.

21:04Whereas if I run into you, maybe you'd understand it because we're all human and we understand that sometimes car crashes occur, right? But I wonder, as we move in kind of this other field of AI, are these hallucinations or these kind of misstatements, is it going to turn people off? So, unfortunately, it doesn't turn off too many people. And why I say that is that what I say is not always what my actions dictate. I could say, oh, I'm angry at social media. It's horrible, horrible. But are you going on social media and still posting and tweeting? Yes. If I'm in a self-driving car and you hear all the news and so-and-so crashed, I didn't see a drop in people going out and putting it in self-driving mode.

21:56They may have talked about like, oh, this is really bad. And click, oh, this is really nice. This is very convenient. And so our actions dictate that we are okay with AI, even if our words are much more reactive in that space. Yeah, for the record, I don't fall into the groups that you just named because I don't have a self-driving car and I just quit social media for the 500th time in my life. While we're talking about the things that make people... Well, no, actually, this is interesting. So in our studies, we do these studies around trust and things like that. When we ask persons at initial like, would you use it or would you not?

22:35It's actually about, depending on which demographic, it's about an 80 to 20 to a 60-40. So basically, you're in the 20 % that you're never going to be convinced. I can be convinced. I can be convinced. But I'm a skeptic and a hater by nature. But whether or not people trust AI models, or if the skeptics and haters can be flipped, the technology is developing at a rapid pace, especially after a Chinese company's model, DeepSeek, turned the AI industry on its head. Where does that leave American AI developers? If there had been resources or there had been data server farms for people just to play around with, you would have had a student in their garage finding out something saying, oh, I tried this.

23:24And then the company's like, oh, that's great. Come work for us. But we didn't do that in this space. That's next.

23:54obviously not too long ago uh deep seek happened took a trillion dollars uh of value off of uh mostly u.s tech stocks in a day um and there was something i thought was really prescient you said in summer of 2024 you told the senate joint economic committee that you were concerned that if the U.S. didn't stay competitive, we might one day wake up to a Sputnik moment in AI. And here we are a half a year later. You have everyone from Senator Chuck Schumer to investor Mark Andreessen saying that Deep Seek was, quote unquote, a Sputnik moment in AI. Well, first off, are you Nostradamus? Do you feel like we're actually in a Sputnik moment?

24:43I feel like we are around it. Not quite at the level of Sputnik, but pretty close. But one of these things is it wasn't rocket science. This has always happened around technology. And the reason for that was that the academics, the ones that tinker around and try new ways, were being cut off from access because of the cost of computes. that that was it whereas if there had been resources or there had been you know data server farms for people just to play around with you would have had the student in their garage finding out something saying oh i tried this and then a company's like oh that's great come work for us but we didn't do that in this space one of the the storylines out of deep seek is this kind of idea that they didn't have the same kind of resources as an open ai or these other u.s tech companies.

25:39So on one hand, I think I hear you saying that not enough in the U.S. is being spent to allow these up-and-coming engineers and coders to learn how to play with this stuff. On the other hand, there's an argument that is certainly becoming popular that DeepSeek was successful because it didn't have resources. It had to think more cleverly to kind of compete. How do you kind of square that? So when you're hungry and have to think of alternative ways of doing things, you become more creative and more innovative. It's all relative, right? So even, you know, like they're saying it was less than training an open AI model, but there's not a lot of universities that could just devote, you know, five to$10 million.

26:26It just doesn't happen, especially for public institutions. So one, it's all relative. But one of the reasons why in the US startups do so well is because they start off, they don't have a ton of resources, just enough to, say, get a prototype that then allows them to get a CD or some other fund. It allows them to be really, really hungry. That's just the way innovation sometimes works. i'd love to um end on a note of of kind of cautious optimism because i think that's your forte um what is going to be that waymo moment that i have when i visit san francisco and i'm like there goes a car with no driver like when can we expect to have that moment um and might might be different if you work in an amazon warehouse for example When is C-3PO real?

27:20Yeah. Or R2-D2. Yeah. So I will say when you go into an airport, maybe it's the hospital, maybe it's City Hall, and you go there and not only is the receptionist a robot, but it's also going to be everywhere that you go. You go to the cafeteria and there is another robot. You go to the restroom and there's a robot that's cleaning the floors. Maybe not humanoid, but cleaning the floors. when you see more than one robot and we become accustomed to it. That's when we know robots have arrived. So it sounds like 2030, 2030, five years. I can see it in some spaces. Yes. Awesome. Well, Dr. Howard, it's such a pleasure to speak with you.

28:05I just really value your insights as both an insider and an outsider. And you know, it's been a pleasure. Thank you. And that's bold names for this week. Michael LaValle and Jessica Fenton are our sound designers. Jessica also wrote our theme music. Our producer is Danny Lewis. We got help this week from Catherine Millsop, Scott Salloway, and Philana Patterson. For even more, check out our columns on wsj.com. I'm Christopher Mims. And I'm Tim Higgins. Thanks for listening.

28:48Thank you.