Ep. #1166 - The Future of Robotics: Innovating with AI

Matt DeCoursey  00:00

And we’re back back for another episode of Startup Hustle, Matt DeCoursey here to have another conversation I’m hoping helps your business grow. Speaking of businesses growing, over the course of business history, innovation, robotics, and now AI are driving so much output and scalability. So we’re going to talk about that today. And before I introduce today’s guest, today’s episode Startup Hustle is powered by FullScale.io. Hiring software developers is difficult and Full Scale can help you build a software team quickly and affordably and has the platform to help you manage that team. Go to FullScale.io to learn more. If you’re not aware, that’s my business. And we love talking to Startup Hustle listener, so click that link in the show notes. And let’s see if we can help you out. With me today, I’ve got Ben’s Benji Barash. Benji is the co founder and CEO of Roboto.ai. You can go to Roboto, oh, excuse me, of Roboto AI. You can go find more information about Benji’s company at Roboto.ai. There’s a link for that in the show notes. Once you scroll down and click it so you can have some con context about some of the complex and tricky things that we may discuss today. Straight out of Seattle, Washington. Benji, welcome to Startup Hustle.

Benji Barash  01:18

Thanks so much for having me, Matt. I’m big fan of the show. Really glad to be on.

Matt DeCoursey  01:21

I’m a big fan of robots and AI, but I can’t I don’t know as much as you do. So I’m looking forward to learning more about it. Before we get into that. How about we start with a little bit about your own backstory? And what brought you to all of this stuff?

Benji Barash  01:35

Yeah, sure. So I’m originally from London, now living in Seattle, so you might hear a bit of an accent My

Matt DeCoursey  01:41

And God Save the Queen, I think,

Benji Barash  01:43

Or the King, the Queen is dead.

Matt DeCoursey  01:47

Do we have to all change to say that now?

Benji Barash  01:49

We do we have to change all the money is.

Matt DeCoursey  01:51

People are people from Europe are always surprised when I say that. But thanks for the correction.

Benji Barash  01:57

Yes, I’m originally from London moved to Seattle, I actually moved to Seattle with Amazon. That’s who I was working for in the UK. And then obviously, Amazon is based in Seattle here they have a big headquarters. I was working for some of their robotics projects in the UK, especially Prime Air, the drone delivery project. That took me here to Seattle to do more R&D for them. And then I spent about seven years working at Amazon in their robotics divisions, and got kind of fed up with how hard it was to build robots. And me and my cofounder decided to do something about it, and started this company Roboto AI to really help robotics companies get to market faster. And the hard things along the way.

Matt DeCoursey  02:34

What is hard about building robots?

Benji Barash  02:36

Great question. They, you know, robots are just so so powerful, but they still are remarkably like, almost like naive at doing certain tasks. Sure, there was a lot of edge cases that pop up. People think that, you know, you can just strap a camera to some hardware and have it go and do some complex things. But the real world is just such a random number generator, and things always go wrong. And so companies get to the point where they think that robots are ready for production, that they can roll them out and scale up a fleet of them, they start doing that, and really that they just discovered kind of quickly that things go wrong. And that’s because they weren’t able to catch a lot of the problems during development and even verification early on in the lifecycle. And ultimately, things just don’t pan out. And they they end up spending a lot of capital on that hardware development. But it’s it’s, it’s just a tricky space to get things right.

Matt DeCoursey  03:23

So at Roboto do you build robots and AI? Or do you consult and help people make their robots and AI actually work properly?

Benji Barash  03:32

So we’re building this specific thing that we’re building is data infrastructure to help robotics companies analyze all of the data that their robots produce? And so they usually need that during development because they need to kind of build new algorithms to make the robots do the things they actually want them to do. And they also need to be able to analyze all that data to figure out like when something went wrong, why did it go wrong? Like how many times have we seen this issue before? Is it the first time we’ve seen this problem? Or is there actually something like really, really bad fundamentally with how we’ve architected the system? So our primary product is, is a data platform that these robotics companies can upload all that data to, and then get more insights really about how their robots are performing and functioning?

Matt DeCoursey  04:12

Yeah, we’ve had a lot of data discussions and episodes on this. And the one thing that I always bring up is that data is pretty worthless if it doesn’t create an actionable outcome. And that’s essentially what it sounds like you’re doing is, like, and you got to do it in a in a quick way because reviewing your data and realizing that 30% of your clients churn for a particular reason, but they’re already gone. That’s just, that’s just history. That’s a history class at that point. It’s not necessarily innovative, you haven’t done anything to move the needle. Now. You mentioned you talked about businesses and growing and scaling and all that I feel like whether, you know, regardless of how people some people love Amazon, some people might not. Amazon is mastered efficiency with and if you have if you have I’ve never done it, go to YouTube and and like, type in Amazon fulfillment robots or something that you just see these like, I mean, they are they do amazing stuff and why? Why is it important to study because I’ve used Amazon’s mentality, not necessarily, I’m not going to say I got it and studied it, but that Amazon will look at everybody that works in these places. And they’re gonna like they’re trying to reduce the number of steps that they take the number of turns they have the number of times they have to bend over, you know, different things. Because when you scale this across, like hundreds and hundreds of 1000s of employees, or hundreds of 1000s of actions, all those wasted movements turn into dollar signs. And the same thing with robots because, like, what I do know about him is that well, first off, that you’re not they’re not taking over the world and killing everyone in the factory, but they might kill everyone, they could kill everyone, and there are some people in the factory by turning the wrong way or grabbing something the wrong way. I mean, there’s a lot of stuff. And it’s, I would imagine that that some of that is probably is that some of the first problems that you have to try to help avoid is like just the general error code stuff.

Benji Barash  06:13

Yeah, totally. I mean, you hit on a really interesting issue as well, which is like the safety and reliability of these kinds of systems. And really, anywhere, there’s a robot that might be in a kind of what we would call an unstructured environment. So humans are the people walking around, maybe there’s other things going on, they just have to be ultimately super safe, and super liable. If anything goes wrong, you know, you can imagine that potentially huge liability for a company like that if, you know, unfortunately, some some injury happened or occurred. So that’s, that’s definitely an area where these companies struggle pretty quickly. And a lot of robots today have actually been successful in mostly structured environments where you know, they’re behind a cage of some kind, and no one can really get to them. And that’s because the risk is lower because there’s less happening. But as soon as you bring robots into a warehouse, or maybe into the real world, maybe drone delivery, or self driving cars, the stakes just go way up because, you know, stuff goes wrong, because humans are there. And if and if a robot does the wrong thing, or you know, break suddenly in the self-driving car, someone could get hurt.

Matt DeCoursey  07:11

I’ve done it. I have two Tesla’s that are self-driving, and I love the technology, but 9, 99.9% of the time, it drives better than I might. And honestly, it does. I mean, because it follows the rules. But then then there’s that one out of 1000 times where it does something weird. And you’re just like, what? Yeah, you know, and it’s funny because I remember not the good parts. Yeah, no, like, there was one I just, you know, if you’ve ever self-drive, the Tesla’s self-driving technology, like you look at it from a user perspective, first off, one of the things when it comes to robots is people are like, Well, what, what was different about it? I’m like, well, first, you got to learn to trust it. You have to, because it’s coming up on another car. And when you first do it, you’re like, please stop, please stop. And it’s gonna stop. It’s gonna stop. But you please stop. Probably the same thing goes with like, any robots and all that. So you know, one thing I should have asked for at the beginning. What is a robot?

Benji Barash  08:08

Yeah, that’s a great question. So I think generally speaking, a lot of people think of robots as the sort of C-P3O or R2D2. They’re not, they’re definitely not. Most robots don’t look humanoid. Really, a robot is actually just anything that’s got some kind of sensor strapped to it. So it can perceive the world, either a camera or a lidar or a radar, something like that. And then it usually does some kind of movement in the world as well actuates. So in response to its environment, it will do something it’ll move a thing it’ll you know, move forward or, you know, enact some action that’s useful. And that’s really all a robot is and most robots actually, yeah, they certainly don’t look humanoid. They might actually just look like, you know, some metal with some arms coming out of it, or even a set of in Colorado. Those are all robots as well.

Matt DeCoursey  08:52

Yeah. And a robot kit is that simple arm that that lifts itself up, it drills a hole, the thing moves down the line, and it does on the next one, that’s a robot. So speaking of robots, and AI decided to ask the world’s most popular AI GPT what a robot was, while you were mentioning that says it’s a mechanical or virtual device designed to perform tasks automatically or autonomously. It’s typically programmable and can be controlled by computer programs or algorithms. Robots come in various forms ranging, ranging from industrial to manufacturing to humanoid. And then it’s got a lot more it does mention they’re almost always equipped with sensors actuators, control systems. Now, here’s the thing, that stuff’s complex.

Benji Barash  09:35

Yeah, it is. And there’s so many ways that those systems fail. It’s kind of incredible. I mean, just in the robotics career I’ve had, like, you know, we used to have issues flying drones and a bird would poop on the cameras and suddenly the drone can’t see anymore. And you know, it’s same with with different types of weather conditions and self-driving cars. You know, those things

Matt DeCoursey  09:53

That happened to my car. I have like a mud or something like, got up and it hit like one of the side cameras and, like, here and I had to stop and clean it off. Because like, every mile I was getting this really annoying alert there was just, like, but part of that is that CYA mentality cover your ass, because Tesla doesn’t want you to hit something on the left side because the camera can’t see it. It’s the backend of that how much how much of what you guys do is falls under that CYA.

Benji Barash  10:25

It’s definitely a lot of it, I mean that almost the point you’re hitting on is that there’s this really long-tail of stuff that can go wrong with the system. And honestly, like the successful companies that have deployed these sorts of robots in production already like Tesla. They just have enormous teams of people that are building this kind of data infrastructure to find these edge cases and find problems, even in simulation, or in real drives that they do in test environments. They want to find these problems a long time before any of their customers do because, of course, as you said, it completely ruins the experience. You know, you’re, you’re driving along one minute, you’re like, Wow, this is the future, and then it does the stupid thing. And you’re like, okay, yeah, this is nowhere near as capable as I am as a human being. And so that’s exactly what we help companies with our data infrastructure makes it possible for them to find some of these edge cases during development. And that’s much, much cheaper to fix and triage than it would have been if they ended up finding them in production and had a customer call them up and be like, hey, the robot you sold me did some really stupid shit.

Matt DeCoursey  11:19

Yeah. So where does the AI component come in? With what with the problems that you’re helping solve?

Benji Barash  11:26

Yes, so we get involved with companies where they already start collecting a lot of data, they’ve put that data maybe in the cloud somewhere, or they actually just have it on stored in different devices. The data that these robots produce is enormous, like, you can imagine a self-driving car is collecting and you know, like, 10, 15 minutes, it’s collecting hundreds of gigabytes of data. So as soon as you need to review and analyze what happened in just one drive, you have to get 100 gigabytes worth of data, maybe terabytes of data just for a single drive. And we use AI to almost identify the interesting parts of that data. So you don’t necessarily need to go through all of it just to find kind of where the the problems necessarily occurred. And then we also use AI to retrieve from the database of video and imagery that you might have all the other similar cases that might have happened as well. So for example, a lot of the time these problems happen. People are like, Wow, I can’t believe how crazy this edge case, a robot ran into us. I bet we’ve never had that problem before. But then you actually you go back and look at the data, if you can, and you discover that you probably almost had that that issue hundreds, maybe thousands of times you came really close to it not being a problem. And you were kind of lucky, or there’s other cases, and just this one time, it kind of tipped over the edge.

Matt DeCoursey  12:40

That’s the interesting part of the common use of so many forms of AI is looking for the correlating factors that come in. Now, I don’t have an I’m going to give a very inaccurate number here. But I was reading a book about data science a year or so ago. Yeah, that’s what I do with my spare time people. So I’ve written this book, well not wanted to be able to communicate about it that I was reading it. I actually didn’t even read a book, dude. I was listening to the audible, all right, which trying the same thing, but it was talking about it was using a phone company like AT and T or whoever, and they are collecting petabytes of data, like, every hour, and then they have a requirement to store it for years upon that. And you’re looking at, like, the nature of that there is not a human that can consume that. And then with that, we’re also biased. Totally, yeah, in a lot of ways, like and you know, and that, that that ability to connect all those dots. I mean, there’s a genius somewhere that can do it. Or come close, but not I mean, it’s I don’t know, I had I had someone telling me. I asked someone I was like, well, they were you doing some I’m not gonna try to give away who this was, but I don’t want to put someone on blast. But they were they were saying that they were doing stuff. And I was like, Well, you know, you should just be using ChatGPT for that, it’s probably gonna speed it up. They’re like, I don’t trust the answers. And I literally said, so you think you’re smarter than the world’s most sophisticated generative AI? And they didn’t even reply. That mean it was, yeah, they were kind of, like, because saying yes to that would be a remarkably egotistical, and probably I mean, 100% wrong, but okay, so. So with that, that GPT search, we have industrial robots service robots, mobile robots, humanoid robots, collaborative robots are those are Oh, those are cobots?

Benji Barash  14:35

Cobots. Yeah, that’s right.

Matt DeCoursey  14:36

Wow. Like what, which of those are the biggest meat on your bone?

Benji Barash  14:41

That’s interesting. Probably somewhere between like industrial and agricultural. We’re seeing the most interest that’s usually because the stakes are pretty high. But really, we’re pretty all serving in that in those spaces. The thing that’s really cool about robots they always they look completely different, like an agricultural tractor versus a drone versus self-driving car. completely different things with complete different jobs. Yeah, the data that they produce is remarkably similar, because they all have cameras, lidar ,and radars and propellers and actuators and control systems. And all that data looks looks very similar. So this the same system on our side can help process and analyze all of that data for them. So we generally work with a lot of different types of robotics companies that you mentioned.

Matt DeCoursey  15:20

The agriculture side of things is I find to be really interesting one I’m so I’m located in Kansas, where there is a lot of agriculture. I also just bought a farm. Now I’m not farming anything other than memories for my family there, but I’m around and I see a lot of that going on. I have my whole life. And, and you know, and I know a lot of people that have been in that space, and it’s a challenge and you know, people look at things like inflation and the cost of goods and stuff like that. You know, the robe, the robotization, of of agriculture could make a big dent in that one we’re not really producing the number of farmers isn’t really trickling down the way it is, and a lot of that stuff, you know, okay, so I’ve only got a little 20-acre farm, but I did use the tractor that I that came with it to mow five acres. And I will tell you right now, I went in my house covered and dirt and gross shit. And the first thing I Googled was robotic lawn mowers. I did because I’m like, God, this is just like this is felt like me because it’s go down this line straight, turn around, and then do it again. But a lot of that stuff, but here’s the thing, if you got to pay a person to do it, or have the different kinds of equipment, it can be tough. And like another thing too, that I thought was really interesting is as the the lot of that so there’s a big push to turn it electric. Why? Because when you’re in the middle of frickin nowhere hauling diesel fuel around is no joke, man thing. Yeah. Yeah. So like some of that, like the electric side of it. Now, there’s a green nature to it as well. But I mean, refueling is like a real thing. And you know, some and then the cost of it, the availability of it, the sustainability of it, and the environmental impact of it. Just, you know, theoretically, if you think about it, you probably don’t want diesel fumes all over your food, either. Absolutely. or collecting in the field. So yeah, yeah, good luck with that. But folks, there’s a thing, this is what’s going on around you. It’s not all about, like, warrior drones. I mean, most of the things that that businesses and entrepreneurs are trying to solve, just like it’s hard to get people to do some of these jobs that are highly repetitive, that boring and mind-numbing.

Benji Barash  17:38

Who wants to do that stuff like it?

Matt DeCoursey  17:39

No one. A lot of a lot of them create a sense create that well, you know, like carpal tunnel syndrome, or these overuse things and Okay, so if you’ve ever read the book, I think it was in Freakonomics the book but not the movie. They talk about strawberry farming. And strawberry farming is really brutal. Because strawberries need to grow in this little tight, they usually grow in a little mound, they have to often have to be handpicked. And, and there’s, and so you get a lot of back problems. It’s like, I mean, it’s been if you’re a strawberry farmer or work at one, you’ve got a five-year window, but that’s true. Like you have like a five year window before you can’t physically really do it. Because it’s grueling. That’s where, that’s where a robot would be a better choice.

Benji Barash  18:26

So, and these industries, they just have massive labor shortages right now. Right? I think, I think robots really are going to be helpful in industries where there’s labor shortages, like agriculture and construction is another big one. There’s so many construction jobs and think about welding, I think the average age of a welder in the US is probably 50. Yeah, that aging out and no one is going into that.

Matt DeCoursey  18:47

I use that example a lot. It’s, I’ve compared it because software developers are the same thing. They’re, why don’t why does Full Scale and once again, if you want to build a team of software experts go to FullScale.io of the people the platform and the process to help you with that. There aren’t enough people to do it in the United States. So like all of our employees are in the Philippines. So we create, you know, we’re creating, giving you vetted people. And that’s what it’s come down to in a lot of cases. Now, that’s easy to do because you can ship code around the world with a push of a button. You can’t ship corn that way or build a car and do a lot of stuff like that. But everyone’s complaining about inflation, but fuel rates are increasing, hourly rate rates are increasing. And that all trickles down to your price at the grocery store. And all of it so, like, this is important stuff. And this is innovation that should change the shape of markets. Absolutely. Yeah, yeah, I met it’s really slow to come out and do like I tried to buy electric farm equipment and stuff like that I had an impossible time finding it. Most of the manufacturers have a product they say they make but when you get into it, they don’t. That’s right. Yeah. have like a small batch. They’ve put out at quite honestly an amount of probably an unaffordable price. And they’re trying to get up to production speeds with that, what can what can change that not just for agriculture, but for all of it?

Benji Barash  20:13

Yeah, honestly, that’s, that’s really, that’s really kind of why we why we exist now. Because people often ask me like, where are the robots? Like, do we not have the technology are all around you? Yeah, they’re all around you. But it’s

Matt DeCoursey  20:24

my floor this morning.

Benji Barash  20:28

Yeah, it’s the honestly just really hard to really help to build and get right only I would say, in the last, like five years, have the right kind of cameras even existed that are kind of robust enough and easy to integrate, and kind of weather proof for this kind of application. So and of course, with the advances in AI themselves, like, you know, you can actually build a pretty good computer vision algorithm. Now that’s pretty reliable at detecting, you know, that that weed that you need to zap on your on your, you know, lawn maybe or your giant field, or maybe that rock that you need to pick, for example. So a lot of the computer vision and AI applications have gotten a lot better now. And the actual sensor hardware has gotten a lot better. So we’re seeing this kind of exciting convergence, where a lot more types of robots are kind of coming to coming together. Now. I was in Israel a few weeks ago, and I, I saw a couple of cool companies there. And one was even building robots that can climb skyscrapers, and they clean the windows on skyscraper. So there’s this amazing, amazing stuff.

Matt DeCoursey  21:24

And by the way, the humans that I knew a guy that used to do that, really. He made like, hundreds of dollars an hour for doing it because, one, there aren’t many people that are lined up for that job. And two, it dangerous. Hell yeah. Yeah. I mean, he did really, really well with that. Yeah, that was a very fearless dude. I’m afraid to be on a too tall of a ladder. So that wasn’t the yeah, that wasn’t, uh, do you need?

Benji Barash  21:51

I think you’re super tall, right? Do you need a ladder?

Matt DeCoursey  21:53

I am quite tall. But. And you know, that’s a common mistake that people make, because I’ve heard, they’re like, you’re kind of scared of heights, but you’re really tall. And that’s not how that works. I’m not inherent. I’m not like afraid to be at the top. We’re in the top row of an arena watching a basketball game. It’s when you, like, the first time I realized I was scared of heights was actually I walked down a glass staircase on a on a cruise ship. And I felt like my stomach fell all the way to the bottom. I was like, Whoa, I felt really weird. And that’s kind of realized that I would get queasy on some of that. So yeah, anyway. Yeah, then being tall doesn’t solve that. Alright, so you talk about I think you made a good point. So I want to back up. So you mentioned Open CV or Computer Vision. We’ve had done some episodes about that. And this is why some of this innovation has been a challenge because until recently, not a lot of people were working with Open CV or CV, computer vision. So what is that computer see things in 2d? And in order to to create depth and perception and field and you look at it. So we actually did an episode a long time ago, there’s a guy in Kansas City, that is one of the top open CV engineers in the US. And what did he used to do that he invented the app that you use, that helps you pick the color of paint at the hardware store. And the level of complexity in that is shocking, shocking. So you say I’m looking at your background, and I can see that you have like a beige colored wall, but because of the way the light shines on it, and there’s a corner up there, like that is actually very from the the optical sense of what your brain does, is highly complex. And then so that’s been out for 30 years. And you know, it’s been out there for a long time. But you’re 100% right, the the ability to have an onboard processor processor that can make those decisions. That’s what blows my mind about the Tesla and my wife who’s tired of hearing it at this point. I’m like, Jill, do you know how to programmatically and processor wise, what’s really required to make this thing not drive us off a cliff? And she’s like, I don’t care. I’m like, but it’s so amazing. She’s like, I’m sure it is. But it’s true, right? Like, that’s the thing is if you can’t add up the data and process it. I mean, I don’t know. Yeah. So what So what is the Alright, so with that great inventors use concepts like Moore’s law, and all of that to invent what they want in 10 years, not what they want tomorrow. They’re thinking about what they got in 10 years. Because if they’re waiting for NOC for nine years to start that someone else is probably nine years ahead of them. Are we truly in this golden age of a lot of this stuff finally, like turning around and being realized, and like that should accelerate tons of stuff, right?

Benji Barash  24:45

It is already Yeah, I think that’s exactly what we’re seeing. Now. There are so many more robotics startups that are popping up just because like it was it was cost-prohibitive 10 years ago to start some of these products because you’d start doing your bill of materials for the for the robot you’re going to put together and be like, yeah, I mean, when, for $15,000 per camera on one device, like, No one’s even gonna buy the device for $1,000 that just the economics didn’t work out with some of these cameras, and now they’re becoming more commoditized. Ultimately, it’s much easier.

Matt DeCoursey  25:14

Is the cameras are the linchpin in that. And now, is that like the main thing, or is it the processors, processors and chips?

Benji Barash  25:21

Absolutely. Those two Yeah, it’s really the advent of GPUs and embedded GPUs that can process all that camera data on board those robots and knock it back to some kind of, you know, cloud to do all the heavy lifting. Right. It’s the commodity

Matt DeCoursey  25:36

Onboard processing and the ability to actually do it. Yeah, you talk about like a camera. So I did I do. You look at repetitive things like, so, we’re going to use iRobot for an example. Because I think that’s something a lot of people have in their home. I’ve had one for eight years, the very first one I had frickin hated it, dude, it just kind of went wherever it wanted to go. Like, if you ever had that old Roomba. It’s just bounced around like a pinball, you know, and like, you’re sitting there looking at that pile of dirt. And you’re gonna come on. And I spent more time watching that frickin robot try to find that dirt then I would have spent actually just vacuuming the floor through. It’s kind of like if you watch the office there was looking for that screensaver where the square went exactly in the corner, you’re just like, Come on, get the dirt, dirt that. Now on the flip side of that the newer versions, they have a camera up front, they have a light, like they identify like it stops. And you know, the here’s the funniest thing. So you mentioned like, whatever it is that was stopping for Do you know the number one thing that had to build it for? It was for poop. Wow. Because if your dog poops on the floor, and then your your, then your robot vacuum comes by and drives over it, it will then it will then glaze the rest of your carpet, either with visible poop or some form of it. It’s like the top thing. And I hadn’t thought about that until an article came out about it about a year ago. And I was like, whoa, and it had a bunch of videos on IT people. Yeah, and that while that’s the thing that was the main driver was not getting the robot to drive through poop. But But with that, I’ve seen this you talking about this big evolution like those robots are really smart. Now they map the house, you can you can set them up in zones it knows to detect when Hey, I picked up a little something extra here. Maybe I should back up and go over it again. Yeah, and all of those. And that’s part of that evolution with these products too, as well. One being able to trust that it isn’t going to smear poop all over. Right, right? Yeah. How do you run you’re in the PR department at iRobot. And you’re like, God, we got another YouTube video this thing? Just you know, I don’t know. But but but that I have noticed that. And I didn’t know if that was just their R&D. But it makes a lot more sense that

Benji Barash  27:59

They’re still selling the naive ones that just like bump around. And they are, they are remarkably kind of, you know, silly, sometimes. But well,

Matt DeCoursey  28:08

We’ll get the utility, though. So the next thing was is you could have an effective one. So I have the one that has the eyeball on it. It does actually like in their app, I can look at it and I can train it to it’ll show me the obstacles that it found and it stops and will drive around and avoid it. It does little it’s the little things too. Like the old ones. They just went everywhere. The new ones actually do straight lines in your carpet. Yeah, that’s right. You know, and so some of that, but I’ve been really impressed with that. And your ability to like program at using the existing technology like Wi Fi to map things out. I mean, it’s pretty impressive overall, compared to where it was just five years ago,

Benji Barash  28:43

10 years ago, it’s really the last five years, it’s yes,

Matt DeCoursey  28:47

10 years ago, what eight to 10 years ago was that first one that just kind of bounced around and, and you know, they asked you to like they want to collect data. They’re using the collective, you know, network of all these things that you can say yes or no, but it learns to that that’s part of that open model that’s gotta be using open like computer vision.

Benji Barash  29:08

Absolutely. If it’s got a camera, it’s using computer. Yeah, for sure. An open CV is the largest popular open source library to do computer vision. And

Matt DeCoursey  29:17

It’s so few people know anything about it.

Benji Barash  29:19

Yeah, you know, it’s not, it’s not the easiest thing to get started with, I would say computer vision. And actually, even over the last five years itself, computer vision has changed a lot like the over the last 30 years. It was it was more geometric. So like computer algorithms, were looking at stuff in the world. And that’s okay, that’s probably a face, you know, I kind of show that to face. But these days, computer vision is mostly deep learning oriented and deep learning. I’m not sure how familiar you are with it, but it just, you know, you just end up pumping a lot of data through an algorithm and that algorithm kind of learns from labelled examples of things and comes up with its own interpretation of why something is something and that makes it kind of tricky to wrangle sometimes and make sure it’s going to do the Right and reliable thing because there’s a lot of non determinism in there. And it might do the right thing. 99.9% of the time, like you said in your Tesla, but then that point 00 1%

Matt DeCoursey  30:09

Tell someone the one time out of thousands. Did you ever watch HBO’s Silicon Valley show? Sure, sure. Okay, so deep learning was needed when it was hot dog or not hot dog. Yeah, and I’ll let you use your imagination as far as what could be not be a hot dog. Yeah, but yeah, that was that. Yeah. So that was the app they built, it was great at identifying things shaped like a hot dog, maybe not identifying whether it really was.

Benji Barash  30:38

And that’s where these these algorithms fail. Because, you know, it can, it can sometimes just be the bias in the data that you use to train them can really screw them up. You know, the, obviously the famous examples, of course, were companies that have been using deep learning and computer vision for like facial recognition, they might have only trained on a database of of white people. And then as soon as a person comes along, or an Asian person comes along, they actually harder to detect. And there are huge issues with that from an ethical perspective and a reliability perspective. So

Matt DeCoursey  31:06

There was a series called Better Off Ted. And I remember watching that, and then they it was just this weird workplaces kind of comedy. And they, they this company made all these weird things, but they always use their own products, but they had built a racist water, water fountain. Okay, like, so it was meant to, like turn on or like the sinks do, but it was only trained, it only recommended they had trained it and only using white subjects. Like it wasn’t recognizing other people. And it was just a play on that. But it’s true. It’s true. So some of that’s Yeah, and it has to

Benji Barash  31:40

do as well, because people just use datasets that already exist in those datasets already have a lot of bias in and they then build their systems with them. And it can be the same with any application that you can imagine. Let’s say you’re building a robot that you know, kills weeds in your in your giant farms. And if it’s killing weeds, and you only you only gave it data of a certain kind of weed, and it never knew about another kind of weed, you know

Matt DeCoursey  32:01

it’s gonna skip it or it’s gonna kill your flowerbed.

Benji Barash  32:04

Yeah, maybe it’ll stop killing your flowers. Yeah, absolutely. So

Matt DeCoursey  32:06

I’ve been I’ve been really amazed, like, so at the property I just bought I had the how have a landscape guy that’s coming in. I mean, there’s a ton of maintenance and whatever, you know, he pulls out his phone and uses an app and takes a picture of a tree or a leaf or any plant and then it tells him what it is. Yep. And I’m like, Wow. You know, like, I mean, that’s pretty cool. You know, and these, but that’s, that’s probably an open CV. CV kind of thing. Yeah, yeah. So, okay. Um, yeah, it’s interesting because these are, the number one answer was industrial robots. I don’t think a lot of people understand how prevalent that is. Is it fair to say that any company or any business that manufactures at scale is highly robotic?

Benji Barash  32:56

Absolutely, especially, you know, you can see some really cool videos online of like the automotive industry, like cars being assembled and cars being painted. It’s all just like these arms flying around doing stuff. It’s 100%. Pretty remarkable, actually. So a lot of that stuff is now automated, and, and incredible, like, production efficiency as a result.

Matt DeCoursey  33:13

So you mentioned being from London recently traveling to Israel. I’m assuming that you’ve traveled through Asia at some point. If you have ever been to the airport in Seoul, Korea. Let the robot city man, they literally have these giant robots that are like human size, that are rolling around and you they they’re there for you, you can go up and talk to them. They recognize what language you’re speaking, you can ask it questions, it has a monitor on the front of it. You know, and they’re really neat and interesting, you know, and like, and they’re, and they’re smart. And they’re, and they’re interesting. Now, one of the things that, so this is a service robot, they also have the world’s biggest Roombas there, basically. I saw I have pictures of them. Because I sent one to my wife. I was like, I think we need this one. But it’s like, because we have a little mop that like takes forever to mop the kitchen, but it eventually gets there. But this thing is like three-feet wide. Oh, wow. And probably three-feet tall. And it’s mopping the floors there and doing doing a great job. What is that? What is the next level of service robots that we might see? Or AI-driven? That maybe isn’t? I mean, is it the humanoid stuff that or is it something different?

Benji Barash  34:31

Yeah, I think you know, you touched on it a bit before, it’s probably the cobot space. I think that’s gonna be really interesting. There have been a lot of cobot companies over the last few years. And you know, they haven’t. It’s just hard to make something so general, so useful sometimes. But now with AI and especially things like GPT and these large language models more than ever before, like humans can can say some stuff and a computer can kind of understand what the human wants and that’s

Benji Barash  34:55

or know that it needs to go learn it

Benji Barash  34:57

Needs to go learn and maybe Absolutely and that’s been an like This is just in the last like year, of course, that these abilities have now been impossible. So I think actually, in a lot of, you know, a lot of maybe kind of engineering environments where, you know, you might need a robot to do something for you, or maybe as a, as a surgeon, you know, you might say, Can you pass me the scapula or whatever it is a scalpel, maybe, possibly the scalpel, it’ll actually pass the right instrument to you as a surgeon. And I think those kinds of applications now where you almost have a kind of a help a helpful, assistive robot that makes you better at doing your job is a lot more possible because we’ll be able to speak to them with with natural language. And we’ve not been able to do that before.

Matt DeCoursey  35:38

Well, I think that’s something that probably could be on this list, if we talked about the future is definitely medical robots. You know, my sister’s, an anesthesiologist and her husband is a doctor as well. And, you know, but And here’s the thing is like, as humans, we naively tell ourselves that we’re better at a lot of stuff than we are. But this that if you can get that precision, you know, they use robots in surgery a lot to that a lot more than you’d think for incisions, stitching up the incisions, and a lot of different things. And I think that you’ll definitely see that down the road, you also look at a level of singularity that will occur. So nanotechnology, meaning the stuff you really can’t see, according to Moore’s law, you will have the computing you’ll will, within the next 15 years, we’ll be able to fit the computing power of an iPhone into something the size of a red blood cell. And think about that people. That’s crazy, if you think about that, considering that I’m 48 years old, and I grew up with rotary telephones, and black and white TV for a couple of years. On my daughter might have a six year old and eight year old my daughter asked me she’s like, Dad, when you were a kid was TV, black and white. Oh, well, that came right after asking if we even had TV. Oh, but also asking if I had a record player. And I had to answer yes to that. I was probably singing singularity is the bonding between machine and man. And you know, Ray Kurzweil is a guy that started out as a musical keyboard inventor and has become, are you familiar with Kurzweil? For sure, yeah. And he’s, you know, he’s been a voice on the singularity and talks about a lot of this stuff. At some point, they’re going to just inject you with nanobots, and they’re going to live in your body. And you know, having a father who had multiple heart surgeries before he passed, like these are things that will get in there and unclog an artery over time, or do a lot of different things. And that has meant that’s there’s AI there. That’s a robot

Benji Barash  37:43

That just many robots basically. Yeah.

Matt DeCoursey  37:45

Yeah, I mean, a wild to think that now, we might be able to do that in 15 years. And unfortunately, hopefully, AI can do some things to speed up the approval process on some of these things. And much like you’re doing figure out where the errors are, you know, because on some of that seven week thing, if you think about that, and this is where so I had, I think one of the very first episodes we did about this was actually with a blockchain inventor, and was talking about like all that. So you look at all the wearables, your eye, your eye watch, and a lot of that stuff. These things are actually getting smart. And the AI and the sharing of the data and being able to link amongst each other a lot easier, is doing a lot of things like to notify someone, Hey, you, you suddenly have an irregular heartbeat. This is a problem. Go to a doctor, where right now, so much of medical is responsive. It’s like after an event occurs, like you’re having a heart attack. Now hurry up and get to the doctor. You have a much, much, much, much, much better outcome if you’re able to get there before that event is fully wide open.

Benji Barash  38:52

Yeah, same. Same with cancer. I mean, my mom very sad that she had ovarian cancer and they couldn’t detect it until it was stage four, you know, and it was just ridiculous. Like, why couldn’t they have found that earlier? And why couldn’t have been maybe some nanobots floating around in her bloodstream that alerted immediately, you know, when there was when there was a detection of that. So yeah, and

Matt DeCoursey  39:11

then you look at robots soon. You mentioned things like tumors or other things. I mean, it’s we’re 15 to 20 years into things that have actual like lasers that go subdermal and like hit a little spot. So my aunt had had traces of brain cancer and the wet and they and she refers to it as like they were playing Space Invaders because they’re just sapping little spots that were problematic but the thing was is those would they were guaranteed to get much bigger but the level of precision and accuracy that they did that out loud you’re kind of sitting there until you’ve experienced it was like wow, but yeah, you could you talk about a shortage of medical professionals or whatever these are. These are very, very good uses of robots, what’s not what else is not because I think medical probably should have or could have been on this list. And maybe it falls under one of these subcategories. But what what else wasn’t on here? Like what? What else is?

Benji Barash  40:08

Good question? I mean, logistics for sure. And mobility. I mean, we’ve talked a bit about cell phones,

Matt DeCoursey  40:12

Mobile robots, were on there. Okay, talking about capable of moving and navigating in their environment. So you got to obviously get a whole subset of that. I guess the agriculture stuff could even fall under that. So self-driving cars. Yeah. The

Benji Barash  40:25

Categories get broader and broader as you Yeah. Cool. That’s right.

Matt DeCoursey  40:28

Yeah. And I guess the medical stuff probably could have fallen under Well, that’s not human. That almost could be a co bot. Yep. In some ways. Yeah. Yeah. That’s pretty interesting. All right. So while we’re here, I would be remiss to ask because there is so much money flowing into robotics and AI. So you guys raise some capital at one point in 2022. Did you feel that that was remarkably easier than it might have been in the past?

Benji Barash  40:57

That’s a good question. No, I think last year and this year have been pretty weird times to raise capital for startups. The markets that are a little strange right now, we were we were a little nervous about being able to raise the seed funding for a startup. But I think we’ve got quite lucky at the end of last year, it was kind of quiet in the in the markets. And then this generative AI hype kind of came along. And now there’s two years before that happened. We raised just before that. So that was kind of good timing.

Matt DeCoursey  41:21

It was trending in that direction. But I think the GPT thing just made. Yeah, blew it up, for sure.

Benji Barash  41:27

But even now, I mean, there’s a lot of investment in these generative AI startups. But investors are also still a little more concerned about kind of who the real winners there are going to be. There’s a lot of noise in that industry right now. And if we had raised, you know, if we had, well, if we did existed the year before, 2020, 2021, were pretty, were pretty good times for raising capital. And then we were pretty lucky in 2020 to 2023. Now is a little, a little more peculiar. If you die in your name, it’s maybe going to be challenging, but even if you do, the investors are pretty suspicious about you know, what’s going on.

Matt DeCoursey  41:59

Currently, the current stats that I saw that came out that covered, you know, six to 12 months I want, especially the last six months. So the GPT thing really exploded. A lot of the interest and venture capital and why am I bringing that up at the end of the episode, if this is something that you’re thinking about doing, I think that you’ve got, well look number, the numbers of the numbers. And right now the amount of money flowing into AI-related startups, just in that category may actually outclass all the other categories added together. You also mentioned like the longtail earlier, so the long tail is, is if you’re looking at a graph, and then it’s really high on the left side, and then it’s got this really longtail of like singular instances. That’s the longtail that’s often referred to in marketing. The longtail marketing is a big thing because it’s easier to go out there and pick up those ones that end up the popular example with that was when you used to actually buy music off of iTunes, all of the ones that you’re showing you all of the things that got purchased one time, every day was this longtail, but those ones all added up to be a higher number every day than two plus. Right. And that’s pretty impressive when you think about it. So my point with that is is take a look at where AI or robotics are the things that are trending might tie-in well to your business or your expansion or your scalability if you’re trying to raise capital because these are these are trendy. Thanks. You know, yeah. The ever changing the opinion of the VC.

Benji Barash  43:38

That’s right. Yeah, I think you know, that the amazing thing about this latest wave of technology is that, like, most businesses can leverage it. Like, I would struggle to think of a startup right now that like could not leverage like GPT and LLM somehow. You know, it’s like being given some superpowers, basically.

Matt DeCoursey  43:54

It’s a muse in your pocket. I was even to you know, I worked in the music industry for a while so I have a lot of friends that are professional musicians and they’ll ask me some questions like what’s what’s what’s this AI stuff? And I’m like, and I asked GPT to write me a country song that was based on the theme that I wanted my that I felt like my ex-girlfriend’s heart I wanted my beer to be as cold as my ex-girlfriend’s heart. And dude, it wrote a hit, like, right there I like read it to my buddy he’s like what? Wow, he’s like, I was like it did that like it did that and 15 seconds.

Benji Barash  44:30

You should record it. It could be or it could be

Matt DeCoursey  44:32

You can happen if you want I mean, I don’t have any sound not talented in that regard. But But yeah, I mean, but that but that’s the point like you can use it for so many things. I think that the main thing with the generative stuff, this is you know, like it is a muse it can give you inspiration, quit looking at the results that you get from that and going well, only 90% of this is good. Congratulations throughout the 10% that wasn’t because it was free anyway, or 20 bucks a month or whatever. You’re looking for the nuggets in there, you’re looking for the gold, not the dirt that it’s buried. And so I think that’s something to bring up. All right. So once again with me today, Benji Barash, who is the cofounder and CEO at Roboto AI. Go to or go to Roboto.ai. I get it right, eventually, right? You’ll learn more about it. Today’s episode of startups is brought to you by FullScale.io, building your software team does not have to be difficult. We talked about Full Scale earlier, I’m not going to extend this ad read for that. With that here at the end of the show, whenever we get a founder as as whoever having a conversation with another founder, I do the founders freestyle, which gives us an opportunity to say any closing remarks. I don’t know if you there’s something we left out anybody you want to thank anything that stood out during the show, like, what are your closing arguments, sir?

Benji Barash  45:58

Yeah, really? Well, I just think, you know, as we’ve said, robots are the future, you know. They’re going to disrupt a lot of the industries that exist today, for good reason. Because we have a lot of labor shortages. And we have an aging population in some of these domains. As we said, the whole thing about getting robots into production and making them successful. And what we want them to do are all these edge cases they run into, and that’s really what our company is helping, is helping other robotics companies with. We’re, we’re helping them find these edge cases, store all of that data, analyze all of that data that they produce, and then ultimately get their robots into production faster. So yeah, it’s been really interesting to talk about that.

Matt DeCoursey  46:36

I mean, this is the future folks. I mean, that’s this time I lead my freestyle with that, like, this is the solution to so many problems. I think people get upset, they’re like, ooh, robots are gonna take the jobs get a different, they’re gonna create different jobs, right? They’re doing the jobs that people don’t want to do, or that we and then those jobs usually have to be overpaid. All this is a trickle down effect. You talk about like the cost of of raising corn, which corn turns into so many different things that you don’t identify as corn, right? But if that cost goes up, the other costs go up with it, and all of that, and that and that’s where robots become a scalable solution. They’re everywhere, man. They’re everywhere. I think people just because they don’t have Rosie from the Jetsons. Like, there’s my age again. But that’s the that’s the whole thing is so it’s not it’s not necessarily coming as see C-P3O or R2D2. To but that’s common, too.

Benji Barash  47:35

Yeah, that that is coming to. Those those will be cobots. C-P3O was a was a pretty badass cobot, I would say.

Matt DeCoursey  47:41

Oh, yeah. Yeah, here’s the here’s the ultimate polyglot. You know, I think was he speak like 6 billion languages or something crazy like that. But, but yeah, so it’s common. And I think the main thing when you talk about innovating with AI is you are 100% right. There’s really no business that can’t benefit from the use of it. So if you’re not trying to get out there and figure it out, like, go ask ChatGPT how ChatGPT can help your business. Define your business as much as you can, talk about the features, advantages and benefits of your product and the biggest problems you need to solve all of that. And then see what it says, and you’re gonna be surprised. You’re gonna be surprised. Don’t just ask it, what are the most popular dogs in the world? That’s not a good use of it. But it house and problem solving skills that are pretty advanced.

Benji Barash  48:34

You talked about the singularity, but imagine a version of ChatGPT where it’s coming up with better business ideas itself. It then has an army of cobots that go and build those businesses for it. And then it’s just printing cash.

Matt DeCoursey  48:45

I mean, that’s yeah, that’s where some of its going. I mean, a lot of it too, is, like, you talked about, some people have been asking me a lot. They’re, like, are you worried that AI is going to replace the software developer? No, not really. The AI currently makes software developers better because it picks up a lot of errors and, and speeds up the writing of code and some of that. It’s still probably a long way away from truly having the innovative ideas and execution that go with it because that’s still a human thing. But it’s out there, man. You know, thanks for joining me today. I need to go make sure that my robot hasn’t run over any why we’ve been doing this because I’ll have a lot of work to do this afternoon. So I’ll catch up with you down the road, man.

Benji Barash  49:31

Really nice to meet you. Thanks so much for having me on.