Inside iRobot: How the Roomba Sparked a Revolution in AI Robotics with Rodney Brooks

Rodney Brooks (00:00:00):
Science fiction says the robots are going to rise up. I'm not worried about that anytime soon. They're not smart enough to do that. How they annoy people and hurt people is by their stupidity, and my belief is you have to have an avenue for people to take over from their stupidity.

Eric Ries (00:00:21):
Welcome to The Eric Ries Show. Like so many revolutionary successes, artificial intelligence is an overnight success that's been decades in the making. Imagine what it must have been like to be there in the early years of the revolution, long before ChatGPT or almost any AI system that you've ever heard of. Imagine how crazy it must have seemed to make a piece of silicon think for itself. And now imagine being the guy that those pioneers thought was a little bit crazy. Rodney Brooks has changed the way we think about the relationship between humans and artificial intelligence forever. He's one of the founders of iRobot, which you probably know for the Roomba vacuum cleaner, which if you can believe it came out in 2002. We've gotten so used to robotic vacuums. We hardly even see them as artificial intelligence anymore. And yet that company has had success with robots for space exploration, mind detection, search missions, and military applications.

(00:01:17):
It was already 15 years old when it went public in 2005. A classic overnight success story, years in the making. With the Roomba, iRobot led the way in getting all of us to learn how to integrate a potentially frightening technology into our everyday lives. Now we do it without a thought. Rodney himself had been working on AI and robots for decades as PhD student and professor at Stanford, at MIT, at CMU, Rodney spent his days coding next to all this foundational AI talent. They were all aware they were building something entirely new with general intelligence, although they couldn't even begin to grasp the implication. Rodney has spent his career inventing the future, and that experience is more relevant today than ever. In our conversation, we talk about the prevailing anxiety about AI and our encroaching robot overlords. We talk about the hype around machine learning in general, how risk, difficulty, and luck have all played into his success. And he describes his firm belief that if you don't have a customer, you don't have a business. Here's my conversation with Rodney Brooks. Rodney Brooks, thank you so much for being here.

Rodney Brooks (00:02:29):
Thanks for having me, Eric.

Eric Ries (00:02:30):
Honestly, it's very rare to get to interview someone who has been a pioneer in multiple fields in AI and in robotics, and also on top of your really storied academic career. I have a Roomba in my house, and I think lots of people listening to this have used the products that you had a hand in building. So first of all, thank you for all that you've made.

Rodney Brooks (00:02:49):
I like to say that I started out as a pure mathematician and ended up as a vacuum cleaner salesman.

Eric Ries (00:02:56):
Yeah, you have to be a special kind of person to see that as an upgrade.

Rodney Brooks (00:02:59):
Yeah.

Eric Ries (00:03:00):
Tell me what it was like taking iRobot public.

Rodney Brooks (00:03:04):
It was a whirlwind trying to get the deck right for the road show to explain why what we were doing made sense. Also, to explain why we had two businesses, because we were both the vacuum cleaners and military robots, both of which after 14 failed business models had hit in 2002. And then the difference it made in the lives of all the people in the company because they had stock options and well over a hundred of them were able to buy nice houses, which was really a good feeling that it had a material impact on their lives. They had worked hard and there was a reward for them at the end of the rainbow.

Eric Ries (00:03:52):
Yeah. As an entrepreneur, you spend so much time trying to convince people that the dream is real and the vision can come to fruition, and people are so skeptical. It's so rewarding to have those moments when you make it tangible for folks and it actually works

Rodney Brooks (00:04:06):
Out. But the best thing about iRobot was when the Roomba was the topic of a sketch on Saturday Night Live, and we didn't know it was going to happen, but they didn't have to explain what a Roomba was. It was now in the culture and wow, we did something that just everyone knows about.

Eric Ries (00:04:26):
Just like you dreamed it up. I'm sure you knew that was going to happen from the beginning.

Rodney Brooks (00:04:28):
Not at all.

Eric Ries (00:04:31):
What did you notice in the company? How was the company different in the transition from being a private company to a public company? Anything that you noticed in the change in culture change in the people that wanted to work there? What did you notice?

Rodney Brooks (00:04:43):
If you go to Boston now and you talk to anyone at robotics companies, there's always people who came out of iRobot. So other people who are in the company believed, hey, I could do this. I could start this. And so it led to a whole robotics industry. Even at my current startup, I've got ex iRobot people who happen to have changed coasts. So it was that belief, "Oh, robots are real. We can do something for real." That was, I think, the biggest change.

Eric Ries (00:05:16):
This episode is brought to you by Vanta. If you're a startup founder, then product market fit is obviously your number one priority. But to land bigger customers, you also need security compliance and obtaining your SOC 2 or ISO 27001 certification can open those big doors, but they take time and energy pulling you away from building and shipping. That's where Vanta comes in. Vanta is the all-in-one compliance solution, helping startups get audit ready and build a strong security foundation quickly and painlessly. Through the platform access trusted experts to build your program, auditors to get you through audits quickly, and a marketplace for essentials like pen testing.

(00:05:54):
So whether you're closing your first deal or gearing up for growth, Vanta makes compliance easy. Join over 8,000 companies, including many Y Combinator and Techstar startups who trust Vanta, simplify compliance and get a $1,000 off at vanta.com/eric. That's V-A-N-T-A.com/eric. So let me take you back. As I understand, you grew up in Australia, you had a really interesting background. You were both a tinkerer in the shed from a young age and building mechanical devices. And I think I even read robots even back then. This would've been in the '60s.

Rodney Brooks (00:06:31):
I was building what I called computers, things that could play games, tic-tac-toe and stuff like that. I tried to build robots, but the mechanical, getting the motors connected well wasn't good. The first one that it wasn't until I was 16 or 17 that I really got a robot to work, but things that could play games or add or do other stuff, much younger.

Eric Ries (00:06:51):
But you also had a really rigorous classical math education.

Rodney Brooks (00:06:56):
I did.

Eric Ries (00:06:56):
And went deep into the mathematics. As you were saying a second ago, I wonder if at the time, did you see those as two different pursuits or did you understand that they would one day be related?

Rodney Brooks (00:07:07):
I didn't actually understand they would be related. It was the intellectual curiosity of the mathematics that I loved. And at the same time as an undergraduate, I was always in the computer center where we had a 16 kilobyte computer mainframe. And eventually I managed to get, it had four full-time operators During the week, you would punch your cards, put your card deck in, the operators would run the program.

Eric Ries (00:07:40):
A lot of people listening to this, do not know what a computer operator was, so maybe just explain.

Rodney Brooks (00:07:44):
Four people, four people whose job was for-

Eric Ries (00:07:46):
Somebody's full-time job to operate computer voice.

Rodney Brooks (00:07:49):
Full-time jobs on a 16 kilobyte computer who would take the deck of punch cards that you put in the input place and they would stack them into the machine, run the program, take the output that came on a printer and put the printed output and the cards back at the pickup place and you could get about a 24-hour turnaround time. There were so many people using this 16 kilobyte computer. By the way, it had a megabyte disc, one megabyte, but only faculty could store stuff on that. But I just lucked out. One of my professors taught numerical analysis, arranged for me to have that computer along with a friend to ourselves for 12 hours every Sunday where there were no operators.

Eric Ries (00:08:35):
Wow.

Rodney Brooks (00:08:35):
We could just do it again and again and again and work down at the basic level of the computer. And we even had our own disc cartridge of a megabyte. So it was wild.

Eric Ries (00:08:44):
Wow. Yes. So, boy, it's funny because obviously grew up in a different era, but even the computers I grew up on the kids today quite primitive, and of course are dwarfed by what everybody has in their phone now. But that I can really relate to that feeling of just pure joy and discovery when you can be direct with the machine without even really knowing what it was good for.

Rodney Brooks (00:09:09):
Yeah, that wasn't important.

Eric Ries (00:09:12):
It was not important. It's hard to explain to people now how useless of a hobby it seemed to respectable people at that time. Did you have that experience too?

Rodney Brooks (00:09:20):
Yeah, and some of the physics graduate students were really angry that I got the computer to myself with Toilets and you're not doing anything with it. You, you're not doing a calculation. You're just playing around. So I was very, very lucky. I've been lucky my whole life. It's been great.

Eric Ries (00:09:36):
So I want to fast-forward a little bit. You came to the US, you were at Stanford for a while, and then you found yourself at MIT, really at the dawn of the field of AI when it was the first practical AI. Anyway, leaving aside the AI of Turing and church and the more speculative thing, the real development of Lisp and symbolic logic. Set the scene for us. What was that like to be in that hotbed of new ideas?

Rodney Brooks (00:10:03):
Well, when I got to Stanford in 1977, it was a different world from anything that I'd even heard of. We had the internet and there were maybe a hundred hosts on the network. We could send email to people. We had graphics on our screens, which was unheard of. We had screens, which was almost unheard of. There was a vending machine connected to the computer, and you got a monthly bill for the stuff that you'd asked for. There were two news wires that were connected to the computer so you could browse the news as it came in. And this was all in 1977, which was just a completely different world. And it was at the Stanford AI lab. Why the AI lab? Well, the people who were trying to do AI were into doing everything that they could think of quickly. And so both at Stanford and then when I was at MIT, it was just innovation, innovation, innovation, new idea, new idea.

(00:11:00):
It was the constant flow. And I was lucky to be able to know the very founders, the people who had started AI back in 1955. John McCarthy at Stanford was on my thesis committee. Marvin Ninsky at MIT was at the lab, and I knew Newell and Simon from CMU. So I knew all those foundational people. And my big regret is I didn't ask them more piercing questions about what it was like at the start. They were just around. I got to see him, Arthur Samuel, who wrote the very first, he wrote the Checkers playing program in the '50s, and it was the first paper where the term machine learning had been used. He was a research associate at the Stanford AI lab, and he worked on the screen editor in assembly language, always wore a tie. The rest of us were hippies. It was the '70s in San Francisco. So I got there, sit next to Arthur Sandel while I was writing code. It was great.

Eric Ries (00:12:05):
Do you have any funny stories from that time of just what it was like to be around people who were... I don't know how self-conscious they were. I guess my question is how self-conscious were they about really doing something genuinely new?

Rodney Brooks (00:12:19):
Everyone knew it was new. They didn't know what impact it would have. They didn't see that. And it was about the excitement of innovating, doing things that no one had thought of doing before was what drove people. And there was certainly some characters around of all sorts of characters, a lot of misfits, both at Stanford-

Eric Ries (00:12:47):
As breakthroughs tend to attract.

Rodney Brooks (00:12:48):
Yeah, yeah. Everyone thought they were trying to build general purpose, artificial intelligence. Now we say, oh, not just AI, but AGI. But AGI was someone's marketing ploy later. Everyone was trying to build general intelligence but weren't thinking about what the implications would be if we did that because it was a little far off. We knew it was years away.

Eric Ries (00:13:12):
And the first startup you did wasn't in robotics, it was in tools for doing symbolic recently.

Rodney Brooks (00:13:16):
Yes. It was an artificial intelligence software company called Lucid, and it was building Lisp, which was the programming language used by artificial intelligence. At the time. There'd been two spin-outs from MIT, which built special purpose hardware for Lisp. And I realized that our software solution was going to be better because that special purpose machines were not going to go as fast advance as fast as a general purpose machine like the Sun workstations, which had come out of Stanford, that was Sun Stanford University Network. There were general purpose workstations, and because there were much broader market than AI, there was enough engineering going into them to make them better and better, faster and faster than the special purpose machine. So I was confident that a software solution would be a special purpose hardware solution.

Eric Ries (00:14:15):
And Lucid was venture backed.

Rodney Brooks (00:14:17):
Right. It was venture backed.

Eric Ries (00:14:19):
Venture industry was so different then though. What was it like to raise money at that time?

Rodney Brooks (00:14:22):
Well, I was the technical person by that time. I had moved back again for the second time to MIT, I had $140,000 house and I had a $100,000 sun workstation in my house, which had initially had one megabyte of ram. Later it got to two megabytes and I worked on the compiler. So I was not out raising the money, but I did learn some really strong lessons about startups, which I would be happy to share.

Eric Ries (00:14:59):
Yeah, yeah. Tell me. Because I think this is at a time now, startups are so much part of the national zeitgeist, the international zeitgeist, but doing a startup at that time was still a pretty rare activity, restricted only to the privileged few. And so there wasn't like, you didn't have the best practices and the massive corpus of knowledge about how to build startups that we rely on now, what was it like?

Rodney Brooks (00:15:20):
Not at all. And we were building Lisp because that's what we used and that's what we thought was important. And it became clear after a while, the market was not big enough and the VC said, why don't you build essentially an IDE for C programming, which was a new concept at that time. You can see the seeds of our destruction in the name we chose for our internal product. It was called Cadillac. It had everything. It was so great. And there was this other little startup called Borland. They had a horrible little thing. Horrible, horrible. They charged $100 for theirs. We charged $20,000 for ours. Guess who won

Eric Ries (00:16:11):
Their speed of iteration was a lot faster too.

Rodney Brooks (00:16:13):
Yeah. And their low price, and they could sell it to way more people who could afford $20,000 for a single programmer. It happened then, but it was a very elite programming. But the $100, that was closer to what by then the PCs were out and someone at home could buy this piece of software. So the market was just enormously bigger and in engineering teams, a hundred dollars per person to be able to really be productive in programming made sense. Whereas $20,000 per person did not make sense at all.

Eric Ries (00:16:49):
Kids these days can't imagine paying for compilers at all. So it all sounds like science fiction.

Rodney Brooks (00:16:53):
Yes, absolutely.

Eric Ries (00:16:57):
You had this interest in robotics though, all the way through. Were you like the pariah among AI and software people that you also wanted to be doing hardware and robotics at the same TIME?

Rodney Brooks (00:17:07):
Not necessarily pariah or a weirdo.

Eric Ries (00:17:08):
Weirdo perhaps. I don't want to overstate it, but yeah, it wasn't exactly a popular choice even among the misfits of AI, it wasn't a popular choice. Right?

Rodney Brooks (00:17:18):
Well, at the Stanford AI lab, there was a robotics group which I was in, which was programming robot arms and Hans Moravec had a mobile robot. And when I went to MIT, I first worked on arms there, robot arms, because MIT had had that for a while. And then when I went back on the faculty, I started a new mobile robot project, and that became my project for many years. So I was a misfit, but within the bounds of craziness, that was acceptable.

Eric Ries (00:17:52):
And what did you think was going to be possible at that time? I mean by today's standards as both the software and the hardware were quite primitive?

Rodney Brooks (00:18:01):
Well, yes they were, but also we figured out how to do a lot with almost no computation. And that was what we had to do at that time because a big computer was a hundred megahertz tower PC by that stage, which couldn't fit on the robot anyway. So I like to think that mobile robots were going to be something, and I kept looking for places they could be used. And when we started Light Robot, it wasn't a vacuum cleaner company, it was a space exploration company.

(00:18:35):
That was our initial goal. We didn't take venture capital. It was purely self-funded, but we were trying to build robots to send to the moon or Mars, and that was our goal. And small robots with small processes, because at the time, I think JPL was talking about a $12 billion mission to send a robot to Mars. And I could tell there wasn't going to be a 12 billion budget for that. It had to be much lower cost. So we worked with JPL, started a mobile robot project there, which later became Sojourner, but we still thought they were too slow. So we started a company to do it ourselves.

Eric Ries (00:19:20):
And I don't think it's that well known that iRobot was bootstrapped in the early days. Was that a reaction to having taken venture for business?

Rodney Brooks (00:19:27):
That was exactly a reaction to that. So we started in 1990. We did not take any investment. I think it was '98 was the first investment.

Eric Ries (00:19:35):
Wow. So an overnight success, eight years in the making once again.

Rodney Brooks (00:19:39):
Well, the IPO was in 2005. It was 15 year overnight success.

Eric Ries (00:19:44):
Yeah. As is typical with these things. Tell me why you felt that way. Why did you want to do it bootstrap, and how did you bootstrap?

Rodney Brooks (00:19:51):
Well, I wanted to do bootstrap because I thought the VCs had pushed us into the C programming environment, and that wasn't what we were good at. And how did we bootstrap? Well, we started by building robots under contract for various people who wanted them as research robots. And we just said 50% upfront, you got to pay us upfront. It's the only way we can do it. And we did that for a few years and we ended up getting various research contracts with different governments, with the Japanese government, with the US government where it was pay as you go after we got there by building robots for researchers. And that's how we managed to bootstrap. But we never had a financial plan that promised we weren't going under in less than three months, I mean more than three months. That was as far out the horizon as we could ever see.

Eric Ries (00:20:50):
It probably forced you to be super disciplined about the scope of the robot since you could only make what you could pre-sell.

Rodney Brooks (00:20:57):
Yeah. One way of looking at it is disciplined. Another way of looking at it is we were really scrappy and we figure out how to do something. It wasn't exactly discipline in the sense you want a disciplined startup these days. So by the time we got our first external investment, we were 30 people, but we had six divisions doing totally different stuff.

Eric Ries (00:21:29):
I know you very famously had 14 different business models before settling on the ones that worked. Do you want to talk about some of the zanier early concepts for what the company could have been?

Rodney Brooks (00:21:40):
Yeah. Well, toys was a thing that got further us along. We actually built Mass, produced some toys with partnering with Hasbro, and that's actually where in 1997, I went to the Taiwan and sat at the feet of Masters of Low-cost manufacturing. The particular person I apprenticed under for a few weeks was building Tamagotchis. You may not remember these.

Eric Ries (00:22:10):
I'm sure I remember the Tamagotchi, but a lot of our viewers will not remember the Tamagotchi.

Rodney Brooks (00:22:13):
Tiny little tiny handheld things. And there was a character in there and a black and white LED screen, and you had to feed it and give it certain things or it would die. And the kids had them and they had to look after their creature and it would go on for weeks and weeks of life and death decisions.

Eric Ries (00:22:34):
Some of our listeners will have played Neopets growing up. And if you can imagine a physical hardware version of Neopets that you had to physically carry around with you, that's what it's like.

Rodney Brooks (00:22:43):
Yeah. So they were building them in massive numbers because it was-

Eric Ries (00:22:49):
Oh, it was such a fad. Oh my goodness.

Rodney Brooks (00:22:51):
It was a fad and different people with different versions, and this guy was building a million of them at a time. And it was a real education in how low-cost manufacture mattered. You had to push it down. You had to, I remember hearing them argue on the phone, I need a million of these plastic cases. What do you mean four pennies each? I'm not going to pay more than three and a half pennies, but it's a million and I want them in six weeks. So that was a real education to hear how you had to be aggressive with your supplies. You had to search everywhere for good supplies. And when it came time to build the Roomba, I was back. We'd already built My Real Baby, which was a baby doll, which had facial expressions, and you fed with a bottle and you tickled it. You played with it, diapers virtually got wet and you had to change them, etc.

(00:23:55):
And that was built in conjunction with Hasbro at a little place in Shenzhen, just across from Hong Kong called Jetta, J-E-T-T-A. And at that time, they had been building stuff, plastic Happy meals for McDonald's or Burger King, whoever had happy meals and things like that. And this had moved them up the chain a little bit. And that was who we got to build the Roomba's first, because now we're building electromechanical systems, but very low cost. So that was that toy experience that got us knowing how to build low cost. And meanwhile, I would go to Taiwan and Hong Kong for MIT business, and then I'd stay a couple of extra days and I'd go around to the little chip houses in Shenzhou down the next to TSMC and look at their lowest cost processor chips. We had about a 50 cent budget for the processor in the Roomba to make our target and looking at what their low cost processors were and what they could do and evaluating whether they could do enough that we could use them.

(00:25:06):
I think one of the really important things I learned there was the discipline of low cost. I mean, because I knew from before from my software days, high cost killed you, but Roomba was not the first robot vacuum cleaner on the market. Electrolux had the Trilobite, it was made in Europe, and it cost 2000 euros. And that was when a euro was worth more than a dollar. So it was way over $2,000. And we set our target, our retail target at 1 99, 99, $200. And so every decision was, does it push you over? Does it push you? Are you staying under? So that was just a real discipline and low cost manufacture, which was why it was successful, I believe.

Eric Ries (00:25:54):
Do you mind just talking about how you got from those 14 business models and the crazy stuff, the filming the robot Amara's to how did you wind up selling vacuums, and ultimately, how did that pivot lead to the company's success?

Rodney Brooks (00:26:06):
At iRobot, we were trying to find a market for what we were doing it wrong. We didn't start with the market. We started with technology, technology, technology. And we eventually started shipping Roomba's in 2002. But I actually have a prototype that we built Colin, Hellen and I built in 1992 at the company, as we had thought for a long time that a robot vacuum cleaner might make sense. And I had written in a paper at MIT in 1983 or '84 probably about how you might have little tiny robots cleaning up and then dumping the stuff in the middle of the floor and the big robot coming and getting it. And I was thinking about out of the box thinking way to change the structure of cleaning. So I'd been on mines for a long time, and then after we had seen how to do low-cost manufacturing with my real baby with Hasbro, then we said, okay, there's almost the elements we need here.

(00:27:17):
But also in the mid-'90s, we had a contract with SC Johnson to build a large cleaning robot. So we had worked on large scale cleaning robots, which in a wacky way, they were doing three things. They were sweeping, wet, mopping and burnishing in one pass, because before that, people had three passes. They did. So we built this robot that could do everything and we ended up licensing the three-in-one technology to a manual manufacturer. Tenant got rid of the robotics, but the mechanism. So we realized we could build mechanisms. We realized we could do low-cost manufacturing. We'd been thinking about vacuuming for a long time, and every time we said to anyone, we've got a robot company, people would say, well, building robot vacuum cleaners. That was just the zeitgeist. When am I going to get my robot vacuum cleaner? So we said, okay, we'll do it. And that's how we got started. That journey.

Eric Ries (00:28:24):
You just made me remember that in the Jetsons, the humanoid robot carries a hand vacuum cleaner to do the cleaning.

Rodney Brooks (00:28:31):
Exactly. Which is another thing, as you see these humanoid robots, are they going to carry a package around a warehouse walking? No, they're going to put it on a cart. That's why I build carts. Carts have been a good idea for 5,000 years. Carts are still a good idea, and we can make them intelligent, but carts are important.

Eric Ries (00:28:48):
Of course. I learned, I've learned in my travels a real reverence for manufacturing and for the skill of it, the history of it, just the incredible dance that is these global supply chains that make manufacturing and especially low-cost manufacturing possible. What surprised you about it when you first encountered it, when you first learned the discipline of it coming from a software background, from a tinkering background, how was mass production different than what you maybe would've expecting?

Rodney Brooks (00:29:15):
Well, mass production in China at that time was very manual, and it was very mom-and-pop. So this is the late '90s into the early 2000. It was before Shenzhen became what Shenzhen is now, but it was family manufacturing businesses run out of Hong Kong and not enormous, a lot of workers doing manual steps and a pretty resilient supply chain. Even then, if you needed 10,000 surface mount resistors, you could get them within 24 hours from some other place nearby. But the really surprising thing to me back when I was learning about Tamagotchi manufacture was the processor chips were being made in Taiwan at Fabulous houses who use TSMC facilities.

(00:30:18):
And it was a little tiny piece of silicon, maybe less a quarter inch square maybe. And it wasn't even bonded in a integrated circuit with leads for soldering. It would be put on the board in China and hand bonded each wire onto that. But the thing that just got me was how do you get them from Taiwan to China to Shenzhen and direct travel wasn't possible then. But even better than that, you would have this little case maybe this big with thousands and thousands of these in them, and they'd be handcuffed to someone's hand, so they couldn't be stolen. And that person would go to Hong Kong, get on a boat, go there, get to the factory, and then the handcuffs would be unlocked because it was the Wild West. Totally. The wild West. The wild east, whatever.

Eric Ries (00:31:17):
Yeah, yeah. Well, and people, I've learned this too, that people don't appreciate how much hand labor goes into even the simplest appliance and just the amount of human labor. I mean, obviously this is connected in even at that time, a lot of issues with human rights and how people were being treated. What was it like being, this is the dawn of that new age of globalization and the decisions that consumers are making about whether to play the Tamagotchi or buy this Tamagotchi or that, or whether to buy the Roomba or some other product are affecting the lives of so many people who are sitting there mastering a craft that is totally invisible, totally hidden from us in the West. How'd you feel about that? Being in the middle of that?

Rodney Brooks (00:31:57):
I watched the change of economic status of people in China, because when I started there, economics were really bad. People would come from far away, work for two or three years, build up reserves, send them back to their family, and then they'd leave. So there was a constant churn of workers. And over the twenty-some years I spent manufacturing and seeing manufacturing in China, it went to the point where now people were getting educations and didn't want those manual jobs, and it became much more automated. So I watched the transformation. I think Americans don't understand what, especially the eastern part of China is like with the number of hundreds of high-speed railroad systems, hundreds of much faster than any train in the U.S. And they went from zero in around 2016 to hundreds of lines. Now it's a scale of modernization that makes much of the US look pretty backward and way behind.

(00:33:07):
That's not to say that parts of the Western China are still not very low, they're still low income, but Eastern China has been totally transformed by this manufacturing boom. And for a long time, I thought that China would keep staying on top because they had built such a flexible supply chain that it was hard to think about competing elsewhere. And then over the last six years, those supply chains have started to get built in other countries, Malaysia, certainly Mexico. And so it had its moment of a few decades where China was the place, and now it's a more global facility. There was always sewing, making shirts and stuff in Bangladesh where you didn't need capital equipment to build the stuff. But now it's going in India. It has become a high-tech manufacturing powerhouse in a way which wasn't there before.

Eric Ries (00:34:10):
Yeah. Part of the common narrative when people critique globalization is they say, this is a race to the bottom, and capital is fleeing whatever jurisdictions have higher standards to go to places with lower standards. And obviously in garment manufacturing, we for many years had accusations of sweatshops and child labor and stuff like that. I'm curious, you've seen the flip side of it, of the ways in which this consumer preferences in rich countries wind up elevating the economic status of a whole generation of people in a low wage country. I'm just curious, how do you feel? It's obviously very much a mixed bag. How do you feel about it and how do you feel like consumers who are buying these products should feel about their relationship with the people that make them?

Rodney Brooks (00:34:54):
Yeah, it depends on the products and the standards of the companies that do it remotely. Some are more exploitative than others. On the other hand, other countries leapfrogged the US in terms of phone ownership and payment systems, which it is taken a while to get to the US compared to a lot of what used to be called Third World countries. And my previous startup to this one, Rethink Robotics, was originally called Heartland Robotics. And the idea there was, well, let's see what tools we can give to make manufacturing more viable in the United States. And I had spent six years as an advisor under the Global Technology Innovation Council to John Deere. And so I been to John Deere factories all over the country, and that was in the 2006 to 2012 timeframe. Fantastic factories, really great factories. So manufacturing was alive and well in the US, but it wasn't for more consumer products.

(00:36:08):
And I was thinking, how can we bring it back to the US because at the same time, I was seeing labor shortages in China, because when you get wealthier, you don't want those jobs. I used to say to audiences in the US who here wants their kid to grow up to work in a factory? And they'd be, oh, not my kid, but the poor people's kids. Yeah, somebody should do it, right? So how to change work. And that was the thought behind Heartland Robotics, which the customers then told me. So I had to come to Boston to hear about the Heartland. They really took that name as an insult, and I didn't mean it that way. So we had to change the name.

Eric Ries (00:36:54):
Well, I think we first met in the context when you were working on Rethink, thanks to our friends at GE, where we were both doing some things with, and I remember giving just an unbelievable demo about those. Those were industrial robots that could be used in a factory, not in a cage, in the presence of humans. And they were force sensitive, not in the Star Wars sense, but in the sense that they had tremendous awareness of the forces that they were exerting. And I think you maybe showed us a video or a demo for us of putting your head in the path of one of these industrial robots. It was terrifying. And so you had no fear because you knew that it would be able to sense your presence and stop. Yeah. Talk a little bit about Rethink Robotics and why you made that attempt.

Rodney Brooks (00:37:40):
Well, I made that attempt because I thought that we could improve the amount of manufacturing in the US. But I also realized, and this is a theme that I had subconsciously before, but now I have consciously, and I can give you some examples where other companies have gone wrong in this recently theme, is if you want a robot that's working along with people, it better not take away their agency.

(00:38:16):
So let me give you some examples of robots recently taking away people's agency, and they're both from San Francisco, which is where I live, and I have taken lots and lots of rides in cruise vehicles. And there is not 10 minutes out of my house where I don't see a Waymo vehicle driving around travels. But the cruise vehicles, which are no longer allowed to operate anywhere in the country, were really annoying the San Francisco Fire Department because they didn't have much training on fire situations, and so they didn't know what to do. So they'd be going along the street, there'd be a fire engine there fighting a fire in a house, and they'd just go into the middle of where the firefighters were. Firefighters telling at it. A person, any person would, oh, oh, I better not go there.

Eric Ries (00:39:09):
They would know what to do. Yeah.

Rodney Brooks (00:39:11):
And then the cars got confused. They'd stop. Sometimes they'd stop on the fire hose and now cut off the water, and the firefighters were just, they'd lost their agency. They had no recourse. So the head of the fire department in San Francisco said, it's okay to go at one with a sledgehammer before it gets close to stop it. There was another famous case a few months ago with a Waymo vehicle on Chinese New Year. The street was covered with people. No human drive would try to go down that street, but the Waymo vehicle had instructions go down the street. So it just crept along, crept along, and all the revelers were there, and they got annoyed at that car. They couldn't stop it. So some kid grabbed a skateboard and smashed the windscreen, and then people threw fireworks in and soon the whole car was on fire.

(00:40:06):
So there's some examples of people getting really annoyed by robots in this space that they don't have control over. And the Roomba had a handle. You could pick it up and move it if it got stuck on cords, the Rethink Robotics robots, you could grab them and that would stop them. You had control and you could reprogram them just by moving them around, showing them what to do differently. And so I think that's an important thing for acceptance of robots, that the people in the environment still have control or recourse because robots are all dumb at heart and they're going to do things which are not right. And the people around need to have some recourse and need to be able to take charge.

Eric Ries (00:40:55):
Yeah, it's interesting as you're talking about that, I'm thinking about even I as a consumer, certainly if I am working with a robot or I have a robot in my house, there's this real element of trust because in order for the promise of agency to work, the person has to believe that if they interact with this robot, it's going to yield to the interaction and be safe to do that. I wonder if you thought about consciously, like how do, and obviously humans, we've had a hundred years loss of fiction, of robots, of uprisings and the terror of robots, and there is something really scary about the loss of agency that robots can represent. I'm curious if you've thought about it. How do you build companies that are trustworthy, such that people are willing to engage to find out that the robot in fact behaves the way it ought to?

Rodney Brooks (00:41:43):
Yeah. So there's two things that you brought up, science fiction. And science fiction says the robots are going to rise up. I'm not worried about that anytime soon. They're not smart enough to do that. How they annoy people and hurt people is by their stupidity. And my belief is you have to have an avenue for people to take over from their stupidity. So you have to have some... As a company, you have to have some trust in the end users or the people around the robots. You can't see them all as enemies. I've got to close my system so no one can get through and use the robot in and it changes what the robot's doing. And so you have to really rethink how humans and robots going to interact, because if it was a person out there working with those people, you'd have the normal social obligations, which smooth that way.

(00:42:43):
And so even if you are in charge of workers going and doing road construction or something, you don't tell them, just be brutally ignoring all people. You tell people who are trying to cross the road, don't go and drive at them in your vehicle. There's no one thinks to tell their human workers to be. And so you have to say, okay, how do I make it so my robots is never an to other people for the acceptance. And some people, as they build a company, they think there's those users out there, and then there's me and my product, and this is what I'm doing and this is how it's effective. But you have to think about the environment in which it's embedded and that other people have needs and desires. And if you just stomp on them willy-nilly, it's not going to end well. And if you have human workers, they never stomp on them that nearly will because we're socially wide, not to be total at all times.

Eric Ries (00:43:51):
So I want to pick up on attention that you're talking about, which I guess I've had this latent sense in robotics and nearly in manufacturing consumer products. On the one hand, you have the imperative for low cost manufacturing cost is such a critical component. And the story of the 4 cents versus three and a half cents component. I mean, I've seen that play out over and over again, but I mean this is an idea going all the way back to Deming, that low cost is also sometimes intention with trustworthiness and intention in particular with quality, if you go to the absolute lowest cost at all times, if you squeeze your suppliers and your workers to the maximum degree, you wind up producing unreliable products, products that cannot be trusted, they're not safe because they're defective because you run risks. I'm curious how as a company builder, have you tried to reconcile these competing imperatives?

Rodney Brooks (00:44:40):
Well, there's two things there. There's always, if you have a physical device, there are standards which vary from locale to locale that you have to meet. There's regulation. And regulation is good for safety of people around. But I always say to my teams in a sense, even though we have to meet those regulations, they're not important to me. We got to go way beyond them and be better and safer in a way that we feel anyone in our family is safe around this thing. That if we came close to violating one of those standards, it's bad. And even if we just got all those standards, we could still build an unsafe product if we wanted to, but we're not going to, we're going to build a safe product. And that happens to be my attitude. I can't say everyone might have that attitude, but it served me well by building a brand that people don't feel intimidated by.

Eric Ries (00:45:49):
How do you institutionalize that intention?

Rodney Brooks (00:45:54):
The first thing, and I had a little struggle with some people in this current company who are no longer with us, when we first started testing that in my current company, testing the robots, they wanted to put up safety walls around the robots and no one was allowed in, etc. I said, no, from day one, these robots are going to be amongst people. He said, "But that's not safe." I said, "We'll make it safe."

Eric Ries (00:46:21):
Safe by design rather than by exception.

Rodney Brooks (00:46:23):
Yes. And that has completely changed how people make the robots work because they're always surrounded by moving robots at work. They're everywhere. They're just there. So they notice when, oh gee, that timer did this thing, which isn't quite right. Why is that? And it's just continuous improvement. Continuous improvement, because they're living it. They're surrounded by these robots.

Eric Ries (00:46:51):
So tell us about the new company. This is after Rethink. This is a new company since then, right?

Rodney Brooks (00:46:58):
Yes, yes.

Eric Ries (00:46:59):
Yeah. So tell us about it.

Rodney Brooks (00:47:00):
Robust AI was originally founded to build a general operating system for making equipment into robots. And we were thinking initially about doing it in the construction industry because there aren't many robots in construction. It's a environment full of people. This world changes every day. So we thought we can make smart robots, which will figure out when to avoid people, how to do stuff and ease some of the labor shortages we have in construction. That was mid 2019. Then along came COVID and we thought, like everyone else, what can we do to help?

(00:47:40):
And before, you might remember in the early days of COVID, it was thought that COVID was transmitted on surfaces. So everyone was wiping down surfaces, everyone was cleaning their hands. And so we started building some robots and building technology where we could use ultraviolet to disinfect surfaces. So that the idea was that because we didn't know how long COVID was going to be around, but in a hospital or even in a workplace or any restaurant or cafeteria, there'd just be robots going around and getting rid of the stuff. We've got some patents, everything's great. And then, oh no, it's in the air. It's not on the surface. But so if anything comes along with stuff is on the surfaces, we're ready. You're ready, you're ready.

Eric Ries (00:48:26):
Yeah, exactly. You're ready for it.

Rodney Brooks (00:48:28):
So then we decided, okay, what are we really going to do now? And then we saw this other labor shortage, which had been exacerbated by COVID.

Eric Ries (00:48:41):
Oh yeah. It was much worse.

Rodney Brooks (00:48:42):
Warehouses, because now we changed how we bought stuff, and there was just so much volume going through individual fulfillment warehouses that there was terrible labor shortages. So that's what led us to that. We were affected by COVID negatively, somewhat, positively, and then even more so it's a COVID story in a sense. Three different influences from COVID. So we've been working hard on this. We have robots in various places. Earlier this year, DHL announced a big partnership with us for us to supply robots to them. And we are working as fast as we can. We are, once again, I'm building robots. I have a new production line, just set up building robots at lowish volumes. By next year, we'll be in partnership with some contract manufacturers at higher volumes. And then probably in 2026, we'll have really high volumes as the demand for these robots looks like it's only going to grow.

Eric Ries (00:49:51):
This episode is brought to you by Runway. You ever feel like you're the only one who gets your company's numbers, like really gets it? Explaining it to other people can feel like a drag. It's unbelievable to me how many founders, their team doesn't understand how the business work. Well, that's where Runway comes in. It's not just for you, it's for your whole team. By integrating with hundreds of tools to pull in your data, having everything in one place makes it much easier for the whole team to understand how their individual decisions impact the overall business. This is how you build metrics you can believe and you can trust. It's all drag and drop. Everything updates instantly. You get real-time investor reports, and everyone can play with the model. When people understand the model, they make smarter decisions. If you want to see how Runway can give your team the clarity they need, head to runway.com/eric for a demo, and your first two months free.

Rodney Brooks (00:50:47):
Person might ask, why did I start here at another company?

Eric Ries (00:50:51):
I was going to get to that. Yeah.

Rodney Brooks (00:50:53):
After Rethink, I had moved to San Francisco because all three of my kids live in the Bay Area. Welcome, welcome. And I got lonely and I thought, oh, if I start a company, I'll have people to talk to. And then COVID happened soon after we started, so that didn't work. But I like being right at the coalface, doing the low-level stuff and just doing it because I make lots of pronouncements about how long technologies will take to mature, but I'm not doing it in a vacuum there trying to make technologies mature every day. So I feel it's a legitimate position for me to take on how quickly things will come along. Anyway, so where we got to after some false starts, some driven by COVID is that we are building robots for warehouses and factories that move stuff around.

(00:51:53):
So in warehouses, just in the US for instance, there are I think 700,000 floor associate jobs unfilled. There's a tremendous need for more automation. The big Amazon places have lots of automation, but most of the stuff that you touch, you buy has gone through unautomated warehouses. So we're trying to make life better for the floor associates so they can be more productive and have a better life. Frankly, with our robots, these robots move stuff around. They interact with people in various ways. They're aware of people that may be a picker who's picking stuff from shelves, putting them in the robot, and it's helping them lower their cognitive load by saying which of the 24 bins that goes in by lighting up a light, they just put it there and then they're on the next one.

(00:52:45):
Or they're going down an aisle looking for some eight-digit long number of boxes of stuff, and the robot goes to the right one and is sitting there waiting for them, oh, it must be right here next to this robot. It's got to look up and down. So it makes it easier. And then when stuff is on what would have been a manual cart before, they just say we're done and it goes off and drives 400 feet or 500 feet. We've been looking at these warehouses and some of the floor associates walk 32,000 steps a day, and I'm sure all listeners to this know what it's like to walk 10,000 steps a day. Sometimes you do 15, rarely you do 20. But can you imagine doing 32,000 steps every day that wears you out?

Eric Ries (00:53:32):
Well, prescribed steps too.

Rodney Brooks (00:53:34):
Sorry?

Eric Ries (00:53:34):
Prescribed steps. You have to go to the exact place. You're told that over and over and over and over again.

Rodney Brooks (00:53:39):
Yeah. So where we can take steps out, it makes the job much more pleasant and where we can take away cognitive load so that not every action is a struggle. What do I do now or how do I do it? It becomes a learned mechanical skill and people get in their flow much better. So we'd spend a lot of time with hundreds and hundreds of workers interviewing them before we started building the robot. So there's that. The other innovation is the robots have a handlebar on them. What's a handlebar good for? Well, if there's a handlebar there, maybe I should grab it. Okay, so zero training, you just put the affordance of a handlebar, like on a shopping cart. Everyone pretty much knows how to grab a shopping cart. When you grab our robots by the handlebar, they go into essentially, we used to call it block of ice mode where it's you can push them in any direction and it's just like they're floating.

(00:54:43):
Even if there's 200 kilograms of stuff on 400 pounds of stuff, it's feather-light. So you don't have to be a big football player to move these carts around anymore. You can be a small person, and when you want to move it manually, you can. When it's operating as usual, you often don't have to, but you've got the option. So if one is stuck in your way, you just grab it with one hand and move it out of the way so you're not stopped in your task by some stupid robot that thinks it's the boss or that its stupidity is more important than your work. And by putting a handlebar there, you don't have to train anyone. You see a handlebar, you know what to do. You make it look like a handlebar.

Eric Ries (00:55:30):
Yeah. I'll go back to your theme of human agency with regard to the robot. And you don't need any special training tricks. There's no code you have to enter. You just grab it, grab it. If it's no,

Rodney Brooks (00:55:41):
That requires a trust on the part of us as product developers and on the part of managers in the warehouse that their people are not going to go crazy and do something weird with this thing. We have other safety stuff in so that if you've grabbed the cart, there is no way in the world you can push it into a person because it's got computer vision and it just won't let you push it into a person. You want to move it aside, fine, but it won't let you ram anyone with it.

(00:56:12):
We don't expect workers to do that in the general case because they're there to work and they're not party revelers out on a Chinese New Year who they know they're part of the workflow. That's what they're there for because they're going to get paid if they work, and they're not going to get paid if they do crazy stuff. But we still make it safe even if there was a crazy worker. But you just have to trust that the workers will not be so offended by the robots, but will treat them as tools that they can use and makes their life easier.

Eric Ries (00:56:54):
Well, that's exactly what I was going to ask about because I think one of the things I learned from lean the literature on lean manufacturing, going back to Taiichi Ohno and Shigeo and everything, is that many important humane ideas in that system. But one of which is that it's both ethically wrong, but also very strategically mistaken to ask workers to invest in improvement if that improvement is going to be at their expense, to train their own replacement effectively, to make themselves obsolete, that you should basically never do that. Even if there are savings to be had. Those savings are fundamentally short-term. And if people are incentivized to sabotage the effectiveness of your intervention, they certainly will do that. So I'm curious, as you had those interviews with the people who really are the and customer for this robot or the people that it works with, what were those conversations like and have you encountered that issue of workers being afraid that these robots are ultimately there to replace?

Rodney Brooks (00:57:50):
I'll start with the last one. Because our robots are just mobile robots and they're not doing the picking whether hand goes in and grabs something and why are we not doing that because it's so hard. I don't know how to do it. Even though there's some companies trying to do it, and even Amazon and the highly automated systems still, it's bringing stuff to a human picker so the people can see that they're not being replaced as pickers, but their life just got a whole lot easier. They don't have to walk so much. Leila Takayama, who's our VP of Human-Robot interaction, she went and asked lots of people questions like, what's your favorite tool? And the favorite tool of a warehouse worker for fulfillment is their knife that they use to slice open a carton because it always works. It's ready at hand, it's on their hip. They don't have to go search for it. They have one on their hip at all times, and it always works, and it's way easier to get the cartons open if you have that.

Eric Ries (00:58:52):
Oh, on the affordances of a well-designed knife. Yeah, a box cutter they're called.

Rodney Brooks (00:58:57):
Yeah, box cutter. It just works. The second favorite one is the, which many of them aspire to is the forklift, because now you drive around and you can lift really heavy stuff, but you're not hurting yourself. So making things easier on people's backs, easier to pick stuff up, where to put it, what height to put it, and that there is no, oh, now I have to open this box. How do I open the box? What do I do? Having all the tools ready at hand for what the task is they have to do is what makes people happy. An unreliable tool really annoys them. If they don't know where this tool is going to work this time.

Eric Ries (00:59:43):
That's a pain can relate. So looking back on, on the course of your career, you seem to have a knack for starting really, really hard companies where the degree of difficulty is really high and where you combine the worst aspects of technical risk. These are often products where there's real question about whether the thing even can be built and perform at the level necessary to achieve human symbiosis, if you will. And also on top of that, plus the market risk, whether customers will buy it, whether the price point is right, whether you have the right business model. So the highest degree of difficulty startups that you can do,

Rodney Brooks (01:00:24):
I beg to differ.

Eric Ries (01:00:26):
Oh, yeah? Tell me.

Rodney Brooks (01:00:27):
I think there's a lot of people doing even higher risk things that will likely fail where they're assuming that some technology that hasn't been developed yet is going to provide a magic bullet and they're investing in the magic bullet and they're getting massive amounts of money from big players to invest in that magic bullet and it's not known whether the magic bullet can do things, although it may look as though my companies have everything wrong with them or everything going against them. I feel like we only try to solve a problem where we can see how existing technologies can be put together. There's certainly some risk in how you put them together and whether you can do it at the right price. But there have been lab demos 10 years ago of all the pieces we need, and if we can put those lab demos together, then we can get something to work.

(01:01:24):
So I break this rule sometimes, but I say to all my engineers, if you can't point to a demo of this from 30 years ago, it's probably too not mature enough yet. We relax that. We're in deep learning all the time, and really that was 2012, 12 years old. We use deep learning. Deep learning has got to the point where we can say, we want to do this task deep learning, this visual thing, and we can know, yes, we can do that, or we don't know whether that's going to work. If it's, or we don't know whether that's going to work. That's not going to go into our design of the product. Yeah, that's well enough. Understood. We can say that we need a training set that looks like this and then it will work. So that's what I try to do. Pick technologies that we can, which are newish, but that can say with assuredness, yes, this piece can be built and we can make it at this cost. Then getting that all together and getting everyone on board is the long hard labor.

Eric Ries (01:02:37):
Well, that actually, so you bring up a really important distinction that people who have not been in entrepreneurship for a long time often don't understand because there's a distinction between how risky something is and how difficult it is. And I think what's really interesting in your answer, and really you see this throughout all the stories that you've been telling, is that you're not afraid to tackle very difficult problems, but you don't perceive, and obviously we all know entrepreneurship is risky, but to you using, by being prudent about the different dimensions of this that we've been talking about, I won't recapitulate all of them. Fundamentally, you are de-risking the enterprise without necessarily making the problem that's to be solved any easier. So you were talking about the distinction between risk and difficulty.

Rodney Brooks (01:03:25):
Yeah. Another thing around that, which I spend a lot of time with my teams is as we're figuring out any particular solution, if I tell them, if you say the customer can just, then you've lost already.

Eric Ries (01:03:41):
Give me an example. Give an example. I think I know what you mean.

Rodney Brooks (01:03:45):
The customer can just put a tags all around their warehouse and then our robot will be able to know where it's, they can just do that and someone's saying, the customer can just put recharging floors across their whole warehouse, or the customer can just go through these menu options. The floor associates choose that.

Eric Ries (01:04:06):
Sure. Exactly. I was thinking that the customer can just train their associates to get out of the way. What's the problem?

Rodney Brooks (01:04:10):
Yeah, yeah. Which they don't want to do. I'm a person, don't devalue me. So the customer can just, is no validation or nothing. They can, but then my engineers get confused sometimes and they say, we know you don't like April tags everywhere, but can we put an April tag on all our robots so they know which robot is nearby? And they say, of course. And they say, we can, you said no April tags. No. I said, no April tags the customer has to put on.

Eric Ries (01:04:48):
No, we are allowed to take it on. We can make our life more difficult.

Rodney Brooks (01:04:51):
Sure. Yeah. We can slap an April tag on every robot. That's not a barrier to adoption for the end customer. It's just something we have to do. So it's always what do we do to make the friction less for the end customer? Here's some friction. Then you can have our glory, our fabulous robots, if you will, jump through these hoops.

Eric Ries (01:05:15):
Yeah. If you humiliate in these seven ways, we'll be glad to give you access. I think all of us can relate to that as consumers of arrogant tech companies from time to time.

Rodney Brooks (01:05:24):
Yeah. Yes.

Eric Ries (01:05:25):
Talk about, I think maybe you haven't said this explicitly. I wonder if the difficulty of these startups, because you're tackling some of our society's like most fundamental problems, we're talking about labor shortages and safety in the iRobot story. There's not just vacuums know life-saving IED detectors and all kinds of stuff that are solving previously considered to be intractable problems. I'm curious if you've ever experienced any advantage to having that level of ambition? Do you feel like you can attract a higher level of talent? Do you feel like consumers and customers and employees are drawn to that vision precisely because of its difficulty?

Rodney Brooks (01:06:06):
Because of the difficulty and because they can see that maybe they can do something and deliver product, make a difference. So over the last six months, I've hired a whole bunch of people whose projects were canceled at big companies in Silicon Valley where they were essentially doing advanced research for years, but they knew that the stuff they were doing, they couldn't talk about outside the company and they knew it wasn't going to get deployed. I know it's a half. And now they come to us and we're say, yeah, we're deploying now. Are you ready? And it's exciting. Now they know that what they do is going to be there somewhere. They can talk about it. They can say, I worked on that. I made that happen. So we've just got a fantastic bunch of people over the last few months from tech layoffs in the Bay Area.

Eric Ries (01:06:57):
Yeah. People always tell me that they need to go join a big company because joining a startup is too risky. And I've had even people who are in the midst of being laid off or having their project reorg out of existence or they're complaining to me about the massive politics of it, explain to me that they still can't join a startup because it's too risky. And it's like, well, where is the risk really? Here you have a chance to bet on yourself.

Rodney Brooks (01:07:17):
Exactly.

Eric Ries (01:07:19):
It's only risky if you don't have a self-confidence.

Rodney Brooks (01:07:21):
You be right there in the decision-making and see whether it's not relying on someone who's being paid 10, 20 million a year financially engineering you out of existence because we're a startup. You're scrappy. If that didn't work, try this. If that didn't work, try that. Get the money this way. Get the money that way. Look for a different a model of delivering the product to people. If you didn't get it right the first time.

Eric Ries (01:07:53):
You were talking about having to have line of sight to the utility of a technology, have it be old and tried. And you talked about obliquely, but I think you've done this explicitly in your writing, especially that you've built a discipline of trying to predict the developments of technologies to know when things are going to happen. And I think I've heard you say that there are these patterns where a technology that it's probably, there's a paper that's been written about it today. So which paper? Today's paper will be the breakthrough. Think about the Transformers paper or attention is all you need or something like that. The long lag between when those breakthroughs are first written about and when they become wide-scale, commercialized. We have a tendency to treat that as an unpredictable chaotic system. But you've said that there are patterns in this that allow you to then anticipate what technologies will be ready for prime time and when. Can you talk about your process of doing that?

Rodney Brooks (01:08:46):
Yeah. Well, it's looking at, because I've been around so long, I've seen so many cycles of this and so many, this is the wonder solution. Yes, it may be the wonder solution, but let's look at what ancillary things need to change for that to be adopted. So I wrote a thing that lot's easy and what's hard, and I said, reusable rockets and electric cars are easy, hyperloop is hard. And why is one or why are those too easy? And Hyperloop is hard. Electric cars are easy because we've had electric cars before, and if you're building an electric car, you can reuse the wheels, the tires, the windscreen wipers, the regulatory

Eric Ries (01:09:33):
Things. Yeah. It's a massive supply chain.

Rodney Brooks (01:09:35):
The supply chain, how people buy it, how the financing works, et cetera. Reusable rockets not easy, but DCX did 10 vertical landings of a booster in the '90s at White Sands Missile range. It had been done before. Every powered landing of an astronaut was on a vertical landing on a rocket on the moon, no less. So the grid fins that you see, they've been on every Soyuz launch since the mid-'60s. They're on all sorts of technologies. There's all these technologies. And besides that, there's the whole insurance industry for launches. There's launch sites, there's ways of buying and selling launches and capacity hyperloop. On the other hand, you're going to be in a cylinder going at 600 miles an hour underground or above ground. How does it stop?

(01:10:30):
How do passengers get on, get off? How do they feel about being sealed inside that thing? How do you keep something so precise for hundreds of miles in length? There's just technical problem after technical problem, business problem. There's no way to finance these things. There's no insurance industry for these things. So it's going to take a long, long time. It's not just going to happen in two or three years. And so I look at what you have to do, what's already there, what can be reused, how much of a shift it is from where we are today and how similar it is to existing businesses to predict how long it's going to take.

Eric Ries (01:11:09):
Yeah. A lot of our so-called genius entrepreneurs don't understand the extent to which they rest on the shoulders of giants. And so therefore they-

Rodney Brooks (01:11:16):
Oh, absolutely.

Eric Ries (01:11:17):
... understand the nature of their contribution, which not to take anything away from that contribution, but when the ego gets its hands on it.

Rodney Brooks (01:11:23):
Yeah. The and the ego also leads to any entrepreneur who succeeds has had some luck. And often people as ego won't allow them to say, I was lucky. I've been so lucky my whole life. Luck, luck, luck. And some entrepreneurs are better at producing. And one way you produce luck is by taking risks. Then you get a chance for luck to come in. But it doesn't mean everything will be successful. And I've had unsuccessful companies too.

Eric Ries (01:11:56):
Join the club.

Rodney Brooks (01:11:57):
Even though there were great things and I love them.

Eric Ries (01:12:01):
Yeah. Talk a little bit about when people are talking about your reflections on the one magic trick and idea of the breakthrough will solve of everything, I'm sure almost every person listening to this has in mind the transformer architecture, the large language model, ChatGPT, just what are your reflections on the current, I mean, we haven't seen a hype cycle like this in a little while. It's pretty remarkable full on platform war, the big incumbents plus the startups. It's a very dynamic and interesting situation. How do you feel about the level of hype that machine learning and the new AI has attracted?

Rodney Brooks (01:12:33):
Well, machine learning has been around for a long time, and deep learning has now been around a dozen years. So we know how that works. And this is the latest iteration of using deep learning. I think people are over hyping it because they get enthralled by the positive examples they see. And underplay how important is to keep some of the negative examples out and think, oh, we'll solve that. We'll solve that. But I think often people don't understand that the basic attention is all you need architecture and what it is doing. So it'll change our technology forever in terms of language capabilities. It is a mind-blowing that the language works so well. It's unbelievable with that architecture that is total surprises. It makes us have to rethink a lot of things. But it doesn't mean that it is ready for prime time and just going to solve everything. And it certainly doesn't mean that it's the right technology now to attach to a robot, and that's somehow going to make magic happen. So let me give you an example. I was thinking about this prepping in my head for this.

(01:13:51):
In World War I, we had aerial combat for the first time, and it was intense. They needed to make their planes better and better and better. If you'd give them flat panel displays, it wouldn't have helped them a whole lot. What they needed was more power to weight ratio of the engines would to be better, structures to be better. And thermal and breathing management for pilots, those four things were still critical 25 years later in World War II. But those additional stuff like radar, I think, but still, and they even had a display in those planes, but a flat-panel display wouldn't have made the difference. Now, you cannot have a fighter without a flat-panel display or a car or any airplane or anything. They're everywhere. But the other technology, they didn't help at that time. There were other fundamental problems. And in robotics, there's still other fundamental problems that the promise of regenerative AI doesn't, or even reinforcement learning doesn't necessarily help. Right now.

Eric Ries (01:14:59):
How does it feel... Like thinking about the whole arc of your career getting started on a 16 kilobyte mainframe and the scrappiness that was required in the early experiments, both in AI and in robotics, and now any kid for 20 bucks a month or a few cents a token can rent a machine learning capability that even a few years ago, no government on earth for any amount of money could have had. Now everybody's got access to it and hardly even know what to do with it. I think about people who are starting out in their careers now or starting out thinking about trying to build a startup with that. How does it feel to you that they have these vast capabilities at their disposal, and what advice would you give them to put those to good use?

Rodney Brooks (01:15:40):
If you don't have a customer you don't have a business. So you've got to build something that a customer wants. Some customer, maybe it's a government, maybe it's the old lady down the street, maybe it's the fire department, maybe it's the military. You got to have a customer. If you don't have a customer and you're not solving a problem for a customer, it doesn't matter how good your thing is and how whiz bang it is, and this is what I just see again and again. I got a whiz bang, I got a whiz bang, but look at my whiz bang. Yeah, where's your customers? You got to focus on the customer.

Eric Ries (01:16:14):
We have really cool machine learning demos, but not a lot of deployed ones.

Rodney Brooks (01:16:17):
You got to figure out how to deploy it. Scale too. You got one customer, it doesn't matter. You have to figure out how is this going to be used in lots of places.

Eric Ries (01:16:27):
That's great. Okay. Can we do a lightning round? Do you mind?

Rodney Brooks (01:16:30):
Sure.

Eric Ries (01:16:30):
Okay. That's my favorite part in preparing for these interviews is just-

Rodney Brooks (01:16:35):
And how quickly am I supposed to answer?

Eric Ries (01:16:37):
As quickly as... Well, I have some quotes of things that I've heard you say over the years just to react to. So you can give me your one word answer or you can go as deep as you like. Fair?

Rodney Brooks (01:16:46):
Yeah. Yep.

Eric Ries (01:16:47):
Okay. This is just a great quote. You said, "Complexity may dazzle the mind, but simplicity fuels innovation."

Rodney Brooks (01:16:54):
Absolutely. The simplest solution you can have that solves the problem the best it is. Yeah. Why? Because then you can build a lot of them. You can evaluate whether they work or not. It's probably going to be a higher margin product. So it's good. Yeah.

Eric Ries (01:17:15):
You said this one that really spoke to my heart. You said, when elders doubted, I pressed forward for the path to progress is often paved with skepticism.

Rodney Brooks (01:17:23):
Yes. And now I'm the elder.

Eric Ries (01:17:25):
How does that feel?

Rodney Brooks (01:17:28):
Yeah, it's a little uncomfortable. I loved being the wild young man, and I was the wild young man for a long time. But I told my engineering managers, I hired some wild horses. We need some wild horses in here. Some people who aren't going to accept the status quo and are going to try some crazy thing. And you can't have too many of them in the engineering organization, but need a few, the ones who had the spice and "Ah, yeah, we could do it that way." Oh, so important.

Eric Ries (01:18:06):
Any advice for people who are trying to endure in the face of skepticism?

Rodney Brooks (01:18:10):
You may be wrong, you may be wrong. That's the cruelty of it. Just because the skepticism doesn't make you right. On average, the skepticism is well-placed.

Eric Ries (01:18:27):
Yeah. They laughed at Newton. They laughed at Copernicus, but they also laughed at Bozo the Clown, right?

Rodney Brooks (01:18:31):
Yeah. All sorts of things. So you just don't know. You just don't know.

Eric Ries (01:18:37):
This is one that I feel like many of my friends do in high scale machine learning probably should check in with every once in a while you said the world is its own best model. And the trick is to sense it appropriately and often enough.

Rodney Brooks (01:18:50):
Yeah. Because if you're trying to build... Traditional AI certainly was building a three-dimensional model of the world planning in that three-dimensional model and then moving or acting. And by the time you got to do that, the world was different from what it would've been when you built the model. So the world is what's there, and if you can sense the critical stuff quickly, you've always got the latest update on what you should be doing. So I stand by that. Absolutely.

Eric Ries (01:19:20):
You've also made the analogy of that to insects. You've talked about how insects can navigate with such tiny brains, and you got a great turn of phrase here, that elegance lies not in grand algorithms, but in refined feedback loops.

Rodney Brooks (01:19:33):
Yeah. That was where I really got my start in robotics by thinking about that. And I was able to make robots 40 years ago that could do stuff that other people could not do because I got those right refined feedback loops, and I still do that. So in our current robots for warehouses, we have world models. We use visual slam. We're probably the best visual slam company there is in producing accurate world models. But as the robots moving around, it's getting feedback loops on immediate sensing at the same time and doing both things. So all the robots in the warehouse are building a digital twin of the warehouse at all times. But as they move, things are different. Unexpected that weren't detected until right now. And we react to those, and that's how we get a smooth, viable system working, which is what people do by the way. They know where the packing station is, but they also are looking as they're walking.

Eric Ries (01:20:36):
In case it moved. Yeah, exactly.

Rodney Brooks (01:20:37):
In case something changed. Yeah. In case the pallet on the ground, which there often is in a warehouse.

Eric Ries (01:20:42):
For sure. For sure. I feel like you're one of the people best placed in the world to explain why self-driving cars are so difficult. And I know somebody asked you once about why they were important. Why is it important to have self-driving cars? And you said it was a simple one-word answer, "Economics." So both why is it important and why are they so difficult?

Rodney Brooks (01:21:03):
Let me talk about why it's so difficult. And I think this was a certain piece of hubris, which led to a conceit. And the hubris was we can get the cars to drive themselves and not change anything in the world. And so that we know capital cost in infrastructure every other time we've changed how our transport system works. We've done something to the infrastructure. We've changed infrastructure. Back in the turn of last century, we went from mud roads in cities to paved roads so the cars could travel.

(01:21:41):
I think we would've been way better off if we'd just spent a little bit on infrastructure and put sensor networks on all our roads, which were stable, collected data from multiple cars and multiple sensors at once, and fed them at high low latency messaging to the cars. We would be in a lot better off position to be solving the long tail of situations. So there is a long tail because everything's outdoors. Things change in the warehouse. That long tail isn't there because the warehouse operators have gotten rid of it, just so the people can work more efficiently. So there's a long tail and the promise of not having to change anything, let us not put just a little bit of infrastructure in, which would've made enormous amount of difference.

Eric Ries (01:22:35):
All right. Last one. And this one again, it's such an elegant turn of phrase, but it's an observation. It could be talking about science, you could be talking about AI, it could be talking about robotics, entrepreneurship, but also about life itself. And I just thought, curious what you had in mind here. "In the silence of observation. Amidst the buzzing of insects, insights emerge like whispers in the wind."

Rodney Brooks (01:22:57):
Did I say that? It's really poetic. Wow, isn't it?

Eric Ries (01:23:00):
Exactly. Thought you might react that way. Just such a great turn of phrase and you rattle these off without even realizing.

Rodney Brooks (01:23:05):
Well, yeah. I think I was referring to when I was first married and living or staying in southern Thailand with my wife's family where no one else spoke English. And I was just sitting watching insects as I was about to start my mobile robot lab and watching and thinking and taking notes. And yes, it was sometimes clearing your head from everything else and not having certainly no email, but there wasn't any talking to me. No one talked to me. So that silence was great.

Eric Ries (01:23:41):
Taiichi Ohno would've been really proud. So if there's somebody listening to this right now who has a hardware startup or an idea to change the world with robotics or AI or something really difficult, any parting words of advice for them?

Rodney Brooks (01:23:56):
It's going to be a hard road to get to build real hardware. Jensen Huang, CEO of NVIDIA said, "If I'd known how hard it was going to be, I wouldn't have done this. It can take a long time. It's not going to be an overnight success. It's going to be years and years and with many twists and turns, and some days it's going to feel really, really bad. But then when I see my new robot coming off the production line and just working, yes, it's such a great feeling." So I live for those highs, but they come really sometimes.

Eric Ries (01:24:37):
Yeah. Well, I so know the feeling. Yeah. The many years wandering in the desert in between those highs can be very difficult.

Rodney Brooks (01:24:46):
Yeah.

Eric Ries (01:24:47):
Rodney Brooks, thank you so much for taking the time and for all you've contributed.

Rodney Brooks (01:24:52):
Thanks so much for a really good conversation. I really enjoyed it.

Eric Ries (01:24:55):
All right. Thanks again. You've been listening to The Eric Ries Show. The Eric Ries Show is produced by Jordan Bornstein and Kiki Garthwein. Researched by Tom White and Melanie Rehaq. Visual Design by Reform collective title, theme by DP Music. I'm your host, Eric Ries. Thanks for listening and watching. See you next time.