Eric Danziger on the Reality Gap in Robotics and AI

Transcript

Brian Heater (00:00)

Do people who work in the industry also tend to overestimate what systems are currently capable of doing?

 

Eric Danziger (00:07)

If you were to get someone in the industry out to dinner and, and not be recording, ? they would have a probably fairly realistic estimate of where things are and kind of what's going on. I do think that a lot of people's business does depend on a same inflated expectation, right? That we are very close and that this is, all I need is another couple hundred million bucks and we'll get there. We're close, we're close. We just need this next tranche of investment. We're gonna hit all of our milestones, right? And don't worry about the fact that we haven't hit any of our previous milestones.

 

Brian Heater (00:39)

Hello, I'm Brian Heater, the managing editor of A3. I am back. We are back with another episode of Automated. This week we are going to be speaking with Eric Dazinger, the CEO of Invisible AI. If you're enjoying the show, don't forget to like and subscribe and check out the newsletter over at automated.fm. Here's Eric and also me talking to Eric. Enjoy. So you were in the army for what about Eight years, is that right?

 

Eric Danziger (01:16)

Seven and change, I guess. Yeah. ?

 

Brian Heater (01:19)

You didn't do the right out of high school thing, right? I mean, you went to college and then went into the army.

 

Eric Danziger (01:23)

Yeah, I went into college and then went into the army, but I still enlisted ? instead of going the officer route. And it was kind of a funny story actually, while was in infantry basic training, my drill sergeants pulled me aside and they were like, dude, they tried to show me the pay table. So like, look at this, look at the difference in like what could happen if you just went to officer school. And they tried to convince me to go be an officer. ? no, I don't know. It's just, felt a little bit too much. think like a real job, if that makes sense, you know, you work up a desk and do PowerPoints and, ? kind of, you know, rest of your life to do that stuff. Right. So.

 

Brian Heater (02:06)

So is, I mean, is that in a sense, is that why you joined the military was to kind of like delay the inevitable?

 

Eric Danziger (02:12)

There's a few different like kind of archetypes maybe for people that joined the military, right? Um, some people find it, uh, kind of a calling a career, right? They're going to spend their whole life, um, doing it. Uh, that, that was never exactly, uh, me, you know, I'm, very interested in technology. knew that I would wind up kind of here eventually, right? Working in tech and working in, um, kind of advanced intelligence systems and that kind of thing. So really it was more on what do you want from your life? I don't know. Without making it sound way too deep, right? think there is a lot when you're younger, right? That you just, you you want to have a sense of purpose, right? You want to feel like you're doing something important. And I feel like, you know, if you have to choose between serving in the kind of global war on terror or going to be a consultant, it wasn't a hard choice, right? For me at that age. But I do think that, that, yeah, I don't know. It's tough, right? I think that like, it's a good thing to do.

 

Brian Heater (03:09)

But you know, there was something that kept you there. seven to eight years is not an ? insignificant amount of time.

 

Eric Danziger (03:18)

One of the things I think that was kind of the most striking for me was on the limited amounts of leave that I did get going to visit friends of mine that were working and living in San Francisco or New York, having a great time, right? They were living, I guess, the traditional kind of post-college life, right? But at the end of the day, you know, I would get to wake up and, you know, go just be outside for 12 hours. And that was like my workday, you know? And I think there's something very kind of very different about that. ? and I think, you know, one of the things that people need more than an occupation, is, is that kind of, I don't know, sense of purpose, sense of kind of doing something larger than yourself. Right. and it's funny. I mean, this is, this is honestly one of the reasons that, I think I'm so drawn to things like manufacturing, right. Is that, ? manufacturing is not a job that people typically get for. You know, for the pay, although they can't pay better than other jobs, right? A lot of times the amount of work and the amount of effort that people are putting in is way beyond what they would do at another job. Um, but it is, you know, it does feel very different to feel like you're a part of making something right or doing something. Um, and not just making lattes or, you know, kind of being a security guard at a mall or something. Right. So I think there's. something to be said for having that kind of collective sense of purpose. find that to be, I don't know, I think very important. And certainly when I was in my 20s, it was something that I needed.

 

Brian Heater (04:53)

So you were kind of like scoping out Silicon Valley at that point. You had an expectation that you were going to probably work for a startup at some point in the future, but after the military.

 

Eric Danziger (05:07)

To be honest, I wanted to work in space robotics. That's what I decided.

 

Brian Heater (05:12)

Yeah, who's among us, right?

 

Eric Danziger (05:15)

It seemed like really cool. There was some asteroid mining company. I tried to like send them an email and they totally ignored me. But

 

Brian Heater (05:22)

watched Armageddon once and you're like, right, right. That's what I want to do. I really want to really I mean, that is a way to make a difference, right? It's to mine an asteroid.

 

Eric Danziger (05:32)

Yeah, I watched Top Gun and I joined the military and then I watched Armageddon and I was like, okay, this is my next career. Yeah, exactly. So I think, look, at the end of the day, I figured, okay, I guess this is going to be the next thing that I work on. I looked up, where can I go to graduate school to work on robotics? And funny enough, I don't think I've told this to anybody at Carnegie Mellon, but I'd never even heard of Carnegie Mellon at that point. Growing up in California, it's just, Everything's about California schools, right? Um, and that's just kind of the world to live in. But I found out about Carnegie Mellon, um, and I applied to a robotics program. Um, actually this was kind of in my third deployment to Afghanistan. was in Afghanistan when I applied and then got in, came back in, I think, uh, April or March and was in, in grad school in August. So, um, it was pretty, pretty quick kind of transition out. from the military. And it was a, that was an absolutely crazy, crazy time. I mean, I've been in school for almost a decade with a bunch of people that were more or less all coming directly from undergrad. That first semester was incredibly, incredibly tough. But Coney Mellon's a great school. And I think they did a great job to kind of help us prepare to work in whatever you want to call it, automation robotics. The big thing at that time was self-driving. A lot of my friends went to go work in all of the kind of major self-driving programs. Meanwhile, I kind of was looking around trying to understand what I could do in computer vision and in manufacturing. I actually visited in Pittsburgh a ton of different factories, trying to understand some of their problems and trying to kind of understand what could be done. And then I moved to Silicon Valley and doing startup and kind of, you know, working on a whole bunch of other problems that had nothing to do with manufacturing.

 

Brian Heater (07:37)

Yeah, I mean, obviously, like as you get older, you know, six, seven years isn't isn't a huge amount of time. But I'm guessing like you still you're pretty self conscious at that point in your life that you hadn't been in school for seven years. And all of sudden, you're the guy like coming in and you're with all these like recent recent graduates.

 

Eric Danziger (07:57)

It's yeah, it was just it was just different. You know, I think it's good in the sense you have a little perspective, right? ? I think I think going to school from kind of undergraduate graduate school, you just take a lot of it for granted. And I certainly did not do that. I was I was very kind of grateful to be there. It was a very different experience of being in the military, especially being enlisted ? in the sense that I didn't have to tell anybody where I was. have to make any plans for the weekend. I could just literally do whatever I wanted. The school itself is very difficult because it requires a lot of effort. But it was something that I was kind of well used to, I guess, at that point, right? it was, ? you know, just found that I could kind of work through a lot of the problems just through kind of force of will, I guess. ?

 

Brian Heater (08:49)

Yeah, yeah, brute force. Yeah. So obviously, there's a lot of ? tech companies that are, you know, contracting that are that are working with the military. But as far as actually like being in the military and working on technology in a meaningful way, it sounds like that's not as much of a straight shot. Is that fair to say?

 

Eric Danziger (09:10)

Yeah, absolutely. The kind of interview process, for example, when I was, ? leaving the military was very, very interesting because I had, ? a fair bit of kind of job experience, if you will. Right. But it was in a field that no one really understood. ? and that they knew that they kind of cared about, right. And this may be more abstract way, ? but they just didn't have a way to kind of map it to something that they, they did. Right. So I basically got kind of two categories of job offer. One was to be kind of a leader, but in some of the like kind of blue collar areas of tech. Right. So in self-driving, for example, it's working with the vehicles themselves, right? So there's maintenance teams, there's drivers, there's a whole bunch of people that have to make that operation happen. Right. Um, or I could be an entry level engineer. Right. And those are the two options. And so very different, right. Very different levels of seniority. ? And there was no way for them to kind of map those two things together. ? And I think that's, you know, part of honestly, ultimately what kind of drove me to entrepreneurship and startups as well.

 

Brian Heater (10:28)

This is really interesting. It really struck me the first time I visited Pittsburgh is that obviously there's this like great history of manufacturing there and that and it is something is a history that the town really takes pride in. ? And this is less and less the case as there are more and more robotics startups. But it is interesting that there does seem to be or there was kind of a disconnect as far as like Maybe they're not being a ton of manufacturing, robotic startups, or that not being the focus in the same way that self-driving was out of CMU.

 

Eric Danziger (11:02)

It's tough, right? think that manufacturing is a very, very tough industry, right? And I think that self-driving's advantage was that it was a lot of it was, was really about technology demonstration, right? There wasn't a kind of customer per se building up kind of large technology demonstrations. It's really just about being a very good technologist and that's it. And I think Carnegie Mellon kind of excelled in a lot of ways.

 

Brian Heater (11:30)

That's interesting. So like in a sense, it almost benefited by how like unrealistic the timeline was, right? That you couldn't do this overnight.

 

Eric Danziger (11:38)

Kind of, right? think that the challenge, right? So being in the Bay Area, right? And being right next to Stanford and kind of knowing a lot of what Carnegie Mellon is trying to do, right? They have a lot of great entrepreneurship programs, right? And they're really trying to stay in Pennsylvania, is trying very hard, right? Everyone has been trying to foster an environment there. There's just a huge amount of natural gravity here in the Bay Area right, which makes it very, very easy for you to learn about startups, to kind of, you know, interact with either new startups or venture capital or whatever it may be, right? It's much, much harder, I think, in Pittsburgh. And again, they're doing some incredible work to make it easier, right? This is something that they are very, very focused on. They understand it's very important. But I think that, you know, in an area where it's really just about building kind of, if you will, the best robot, right? Then they can compete with, with anyone, right? They are the best robotics program in the world. What you need to do is kind of build and understand commercialization and all that kind of stuff, right? I think it's very tough.

 

And I don't think it's, ? and that part, be honest, I don't think it is a fault necessarily of quantum email. I think just robotics in general is very tough, right? And I think manufacturing especially is very tough And I think that any kind of group of people that I've talked to in manufacturing technology, it's something that we all kind of have in common, right? Is that this takes longer. This is harder than if you just start a simple software startup, right? And so anything you want to do in a physical world, anything you want to do with robotics, it's very, very difficult, right? And so I think, you there will be a time in the very near future where those companies are very prevalent. I think we all feel the reason that we work in robotics, the reason we study robotics is because we feel that this is very important, that this is one of the most important areas of anything that may come out of the AI boom that we're in. But yeah, it's really, really tough. Ordered to take over the city for it to really kind of replace what what steel was and it's in its psyche, so to speak.

 

We need many of these companies to succeed and we need them to become very, very large. I think the only kind of example I guess that I can see of that ? are things like ? like Dell and Intuit maybe in Austin, right? And Amazon and Microsoft in Seattle, right? Amazon and Microsoft. Obviously much, much larger and so you needed a much bigger kind of tech ecosystem, think naturally just by virtue of their size. Uh, but I think without kind of breakouts like that, it's very difficult for you to kind of create that. If that makes sense. Right. And then of course, in the bigger, you have whatever, probably three dozen companies like that. Right. And so it's, know, it's different. The natural gravity is very different, you know, and I think that. It's tough when you kind of have to consider where you want to start a company, where you want to live, right? You, ? all these factors kind of come into play. ? and so you really are working from behind as, as any, anyone who isn't the Bay area, basically.

 

Brian Heater (15:12)

Yeah, and as far as the kind of the big self-driving boom, I was talking some about this recently and I don't want to put too fine a point on this. The parallel between what we're seeing now with humanoids and what we were seeing then with self-driving cars. But I mean, you go where the investment is, right? You know, you go a lot of a lot of people do where the money is going and where the interest is, like knowing full well that that's where the innovation is going to start to happen. It may not manifest in exactly the way that you expect it to, but, you know, that's where a lot of these companies ultimately end up being built out of.

 

Eric Danziger (15:53)

tough, right? I think that if you look at self-driving, example, Waymo and Tesla being two major players even back then in this space are still major players in this space because they have a bigger business basically that supports them, right? And without that, no one would have been able to put the amount of money that they've put into those programs for that amount of time. And I think that fundamentally is the challenge is that Robotics, mean, the reason I studied robotics reason anybody studies robotics is because you believe that it's possible, right? Like, you know, go into a field thinking as a network. I think that the challenge is that, you know, where we are today, we have, you know, some amazing examples of locomotion, right? Of moving around in the world. But, you know, I think if we can't rely on This, kind of new form of AI that's become prevalent in the last, you know, three to five years to write documents correctly for us or to, you know, guess what we want from code correctly, right? Both of which are much more constrained problems in some ways than robotics, right? Um, it's hard to imagine that you'll have fully working, fully capable, dispatchable robot to go do a job. Right. Um, and it's tough. think that, you know, everyone in robotics knows that demos are easy. Right. Demos are the lifeblood of robotics in many ways.

 

That's how funding happens. That's how everything happens. Right. And yet, you know, it's really hard to kind of bring it to the real world. ? One of the examples that I, I, that really stuck with me is ? from Boston Dynamics, right. And again, Boston Dynamics, they're a team that has been working on this for a very long time, but a lot of it wound up after the kind of DOD and government funding by being a part of another business, right. Being a part of Google, being part of Hyundai. Right. have the funding to kind of continue to work on this stuff, even though it's very hard and it takes a long time. It will be a while until they can pay their own way. But there was a great article about one of their big dogs, right? It was doing a mock patrol with a bunch of Marines. They put a bunch of equipment on it. And of course, you know, halfway through it, like Nick, the hydraulic line or something and fell over. Right. So this thing's like bleeding out. It's hydraulic fluid, all of the weight of this machine, plus all of the weight of the Stuff that they had put on the machine is now the responsibility of these Marines. In addition to the fact that this thing probably cost millions of dollars and they can't just leave it. And that is kind of the problem in robotics in my mind in a nutshell. A demonstration is great. But anytime you want to actually get something out into the world, we see this with electric vehicles. Electric vehicles work. There's no disputing whether or not they work. I've owned them. The challenge is that, you know, We have a new infrastructure problem now that has to be solved. Right. And it is being solved, but that's a relatively simple infrastructure problem. Right. It's literally just like, where can I charge this car and how far can I go and all this kind of stuff. if you imagine kind of what it would take for business to switch over to using robots from people, right. That's a whole different infrastructure problem that you have to be able to solve at the same time as the robotics problems. Right. And There's a whole kind of ecosystem of things that need to come together basically for this to be useful. We don't know quite frankly what that is. We have some limited deployments of robots. There are companies like Locus that have done an incredible job of deploying robots in warehouses. ? And there are robots that obviously do work every single day and have for probably 60, 70 years in factories all around the world. So we know some of how to do these deployments. ? But the idea that you're going to have a fully packaged kind of person replacement that just drops in and does useful work. think people kind of get caught up in the demo and they forget how hard it is, right? Just like self-driving. You can go on the highway. You can take people around a kind of neighborhood that you've mapped 100 times. But I haven't heard of any plans for anyone to go to Toronto, for example, right? That's not on the roadmap yet. We're still trying to get, you know, the three or four cities that everybody's kind of figured out to work. One of the most interesting, I think, things about this technology is that we can both see it so clearly, like self-driving cars or robots. mean, these are things that we've talked about for a hundred years, right? Uh, probably longer, guess, but we know like what we want to do. It's just really hard and it's going to take longer than we expect basically.

 

Brian Heater (20:34)

Yeah, and it's always like the last few millimeters that proves to be the most difficult part of it. a lot of it, I think, ends up weighing the consequences, right? mean, if chat GPT makes an error in your school essay, not a big deal. If a Waymo hits a pedestrian, that's a much bigger problem.

 

Eric Danziger (20:56)

Waymo's are incredibly safe, right? You feel very safe being in one. It is a big deal, right? And it is taking those things into account is very, very tough. Again, there's a lot of work to be done, right? I think, I guess one way to kind of continue the analogy in my mind is, you know, try to imagine at what point the majority, let's say of United States will be able to hire a robo taxi, right? ? or at least like maybe the, all the places you can easily get an Uber today, right? Could you get a completely autonomous vehicle? Like what year do we think that would happen? Remembering that we got really excited about this in like 2016, right? had demonstrations of this probably from the eighties, right? So, we could, we could do this for quite a while. ? and I think if you kind of take that and then extend that to this Humanoids project, right? We can decide if we think it's harder or easier because it's not going at 80 miles an hour, right? And so there might be less risk. ? but if we think that, you know, Humanoids today are kind of where self-driving was in 2016, and we think it's going to take another decade, two decades or something for this to get fully rolled out in self-driving, right? I think that's in my mind, a pretty solid analogy of when we think Humanoids will be kind of capable enough to do a lot of work.

 

Brian Heater (22:19)

Having kind of sorry, ? pun, partially intended, but having a front row seat specifically to the world of self driving, both through CMU and you actually spent some time as startup as well. It sounds like that made you much more of a pragmatist because you've been through one of these big hype cycles.

 

Eric Danziger (22:35)

Honestly, going back to the military too, right? There's a dozen examples that we had in every single deployment of, of kind of really cool whiz bank technology. But then stopped working at maybe the most important time for it to work. Right. And then the like thing that you have to do instead of that whiz bank technology is actually more dangerous if you didn't have it at all. Right. And you'd plan from the beginning to not have it. The reality is like any time that the stakes are high. You really do wind up being very conservative, I think, in your kind of application technology. do feel like self-driving, you know, it felt, I think it felt for a brief moment, very, very easy until you start to really kind of weigh the risks and, you know, try and solve the problem completely and not just, you know, to the 80 % level. ? I think that that did fundamentally make. me want to do something just kind of one step behind the actuation. I actually do find what we do to be very similar in a lot of ways to robotics. we take a very robotics approach at this company. ?

 

We do things in kind of a global frame. do a lot of frame transformations, right? We kind of combine data from all these different sources in 3D to make it a little bit more easier to work with, I should say, right? Whereas a lot of computer vision companies operate purely in the kind of frame of the image and trying to do everything with deep learning, right? So I think that we kind of imagine our devices as really in a lot of ways the kind of head and vision system of a robot. It just doesn't actuate. just doesn't move. That part of it is the part that in my mind, we need to figure out first, right? And if we can get the kind of vision system and the understanding of the world to the point where, ? we, we, can do it flawlessly. We can do it fast, right? We can have everything that we need kind of there at the edge, ready to, to kind of pass off to a planner or something that's going to make decisions on behalf of the robot. ? then. I think we'll be able to do significantly better. But the reality is computer vision itself is quite a bit behind the language systems that we have. And that a lot of computer vision research is now becoming language research. And they're just trying to tack language models onto image models and ? get some understanding and interact with everything in text and all these kinds of things. So we'll see where this winds up going Ultimately, we still have, I think, quite a bit of work on the vision side to figure out and understand what's happening in the world, right? A video that you or I could kind of describe effortlessly ? still trips up a lot of these systems, right? And so we need to get there. And then I think once we get there, ? that in my mind is the gate to start talking about the robot.

 

Brian Heater (25:31)

That's really interesting. I mean, obviously you're building the company is building this product that can currently like be deployed in a work setting, but you're really at the same time, you feel like you're also building the infrastructure for physical AI at the same time.

 

Eric Danziger (25:50)

It's funny that you say that because I think, yes, we are also building the infrastructure, right? But I think it's ? infrastructure to me is more the devices themselves, right? And the fact that we can kind of completely map in 3D in real time ? a factory. The way that we encourage our customers to use our systems, right? They're very dense. They're throughout their entire assembly area, right? And that means that we can kind of help them actually understand holistically what happened and for us to be able to kind of combine that with some amount of robotics would actually be fairly easy, right? And I think that all of the things that would be incredibly hard to imagine kind of a collaborative robot doing, understanding who they're supposed to work with and what part of them work there in and what they're supposed to do next, right? All of those things that we could feed to robot. I think from an infrastructure perspective, I think we are setting that up pretty well. But I do think that But I guess part of, for me, what this is is the future kind of understanding components, right? So the actual work that we're doing on our devices to understand the world, I think that is also setting the stage for this next generation of robotics. It's going to be more important, I guess I should say, right? Is that the end, know, someone's going to build a robot body. They're going to have their motors and a battery, right? They might have. some kind of additional components that are in there, right? Some sort of special end effectors, right? Or they'll make some special hands. But ultimately there's something that's gonna be controlling all of this, governing all of this, that understands the environment that it's in, right? What's happening? What's gonna happen next? And I think that's the piece that more than anyone else will be set up to build, right? Just because this is at that point going to be an environment that we've operated in for quite a long time, right? Having seen factories, having seen… tons and tons of different places. And, you know, one assembly shop in one automotive facility is something like 2 % of the video that YouTube uploads every hour. Right. That's a statistic that I've learned recently and I absolutely love. That means if we get, you know, 50 of these put together, right, we will have of just engineering and manufacturing video, right, as much data created every hour as YouTube create. And so, you know, having a data set like that, having a kind of a unique ability to see the real world in a domain that actually matters, Not Instagram videos or TikTok videos, right? Is, I think, very, very important. So I think in both ways, right, I see this really as a kind of a stepping stone to this world where we are eventually able to do real robotics and have it actually, you know, take on a chunk of work.

 

Brian Heater (28:43)

What do you mean when you say real robotics?

 

Eric Danziger (28:47)

I don't know. That's yeah, that might get me in trouble, guess. I think robotics is one of those funny things, right? I think the old joke is like, it's only a robot if it doesn't work. There's obviously systems, mechatronic systems that operate everywhere every day, right? They're running conveyors. They're, know, something packages between different, you know, different belts.

 

Brian Heater (29:09)

You're talking about like autonomy. Is that kind of what you're getting at?

 

Eric Danziger (29:12)

I guess I think the idea that we all have, ? of this end point, right. Is you have, you know, you're kind of manufacturing C3PO, right? Where you can like, just tell it what to do and it goes over and it does something. You have the ability to not have to kind of babysit the system, right? You don't do a ton of extra work to, to kind of set up everything specifically so that you can do it. Right. The story of kind of a lot of the robotic systems today is that the majority of the work actually is the surroundings of the robot, right? It's like, how do you install the robot so that it can touch everything? What end effector does it get? Right? How are you programming the robot? Right? How do you program all the other systems that interact with the robot? Right? The actual robot itself is just a common object that you install and you feed the program. Right? And again, it's a huge amount of work, right? There's some companies that make those systems. Obviously they're very difficult. There'd be a lot more companies that make those. I think that, you When you think about what it would take to install a robot in a specific place, it's everything around it that takes all this time. Right. And so if you could just direct a robot to go do something without having to change anything about the environment, or even have it change its own environment, right. Like one of the first steps you're going to want to do is kind of go through the lean consulting kind of, you know, how do I change this process to be more efficient so that the robot can do the more efficient process and not be stuck doing less efficient process. Right. But if the robot itself can go and kind of look at the layout and say, actually, these two machines should be closer to each other. And I'm to move this, this thing here. And then it starts work. I mean, that would be the kind of Holy grail, right? It's kind of almost a self-organizing factory at that point. And all we have to do is say like, here's the widget, just help me figure out how to make this and then make like 10,000 of them and make sure that quality is high. And then the system goes off. I think that's kind of, you know, this is where we want to get to. ? think it's just, yeah, there's a lot of steps between here there.

 

Brian Heater (31:12)

Yeah, I mean, I think that most people would probably most people who know anything about the industry would agree that like the average person probably overestimates what AI and robotics is capable of. Right. I mean, you you see the it write a story or, you know, you see these humanoid robots walk and all of a sudden, like you project a lot of things on them. But to a certain extent. And I ask this, like in the context of you saying that computer vision is behind. Like obviously computer vision is at a point where like, you know, some navigation is able to happen. You know, you mentioned Locus, like robots can hang out in a warehouse, navigate around. Do people who work in the industry, do they also tend to overestimate what systems are currently capable of doing?

 

Eric Danziger (32:03)

If you were to get someone in the industry out to dinner and not be recording, ? they would have a probably fairly realistic estimate of where things are and kind of what's going on. I do think that a lot of people's business does depend on, let's say an inflated expectation, right? That we are very close and that this is, know, all I need is another couple hundred million bucks and we'll get there. Right. ? and again, self-driving operated in very much the same way, right. Which is like, we're close, we're close. We just need this like next tranche of investment. We're going to hit all of our milestones, right. And don't worry about the fact that we haven't hit any of our previous milestone. ? and so I think it's tough, right. I think that, that at the end of the day, ? I think that for the most part, everybody has a pretty realistic estimate who works in the industry. think understands everything, right. They have a pretty realistic estimate of where we are, how much is left. What is. possible today. A lot of the kind of robotics use cases that I've heard are frankly robots walking around, right? Because that is what they can do today, right? Locomotion and kind of working on a very terrain. It's a difficult problem, right? It's not trivial. They've solved it, but that is kind of the extent of what kind of they're going to be trying to do with these systems outside of like very carefully calibrated demos. And I think the biggest problem and the thing that drives all of this, right, is that there's always that just fraction of a chance that we're all wrong. Right. And that like, we are very close actually. Right. And then it did only take this one little tweak and all of a sudden everything works. Like you kind of know that it's not true, right? Like you're sure like the sun is going to come up tomorrow. Like all of these things, you know, we, we know with pretty high probability that this is, we have a ton more work out of us. Um, but think it's tough for people to kind of let go of that idea that maybe it's very close.

 

Brian Heater (34:01)

Yeah, I think maybe also what you're getting at it and you you're you know, you're from California. I'm from the Bay Area myself. So like I've experienced this to a certain extent through much of my life. I worked at TechCrunch for nine years. Like I get it. You've raised several, several rounds. You raised a series eight, I think, in 2022 that obviously. You're selling yourself to investors, right? I mean, that's that's what the pitch deck is you're selling yourself, ? not just what you are now, but what you're what you're going to become what your what your potential is. So how do you how do you square that? And how do you walk that line between being exciting enough to draw in investors without over promising?

 

Eric Danziger (34:51)

Yeah. you know, I think, I think that is something that we have definitely been trying to do. ? I think that, one of the things that's kind of interesting about, about all businesses, right. Self-driving went through this, ? and, and, know, we, we are going through it is at some point it's not about the technology anymore, right. And it is about your kind of business milestones, right. And we're fortunate that we work with some very, very large companies, we're pushing pretty aggressively to kind of expand and install this technology. But it's, you know, that's, is another thing that's very, very difficult about manufacturing technology companies in general, and robotics and kind of advanced technology companies in particular. We've seen some really amazing counter examples and kind of AI coding and some of the like language based systems that are out there, right? They've grown incredibly fast. But for the most part, people that are trying to make these really complicated systems work in the real world. ? It's just, you know, it takes a lot of work and it takes a lot of work for you to get this generalizable enough for kind of your variety of customers, right? To hit the accuracy benchmarks or whatever benchmarks your customers want, right?

 

I think robotics companies see this a lot, right? Where, you know, they need to hit a certain, let's say jobs per hour before they can, you know. really expand with the customer because they're competing against what their customer status quo is. And if they can't hit that jobs per hour, they're not going to expand. And so they of want to gate it by all of these benchmarks. So at some point, that is kind of what you are judged against. And it's getting bad. I think if you can kind of get into an area that expands very quickly, which is what we think we're in, Um, then you are able to kind of match kind of traditional startup targets, right? But I think this is the, the, yeah, again, the biggest challenge I've seen in manufacturing technology and robotics companies is that you, from a software perspective, the software company's growth is supposed to be two, three, five, whatever some large increase over the last year. And if you're a manufacturing technology company that's growing 20 % year over year, that actually might be for that market. Fantastic. but it's not as good of an investment as a company that's growing five X year over year. Right. And so, ? that's kind of, think the biggest, the biggest challenge. So you have to kind of, you know, make sure that you are representing the technology and the business milestones in a realistic way. I think that, you know, people that are still in the technology fundraising portion, ? an easier time promising a new demo, right. And I think that was the kind of early years of self-driving. And then at some point, investors were like, wait, how long have we been doing this? When are you guys going to actually start getting paid customers? they're like, oh. And that's when kind of the whole thing wound up drying up. It kind of split. So it's tough. think that that kind of experience, I think in my mind, the kind of shorthand I give it is like five years. the current robotics startups, they started two or three years ago, right? They have about two or three years left. And then investors are going to start saying, okay, but we put all this money into it. Like where's the business coming value? And when is this coming back? So I think until that point, we can sell technology and we can sell really well polished demonstration videos. I think a friend of mine in robotics. call some of these companies more of a movie industry, right? Cause they spend more of their time making films than they do working kind of robotic systems. But that to me is kind of going to be the big shift, right? So whenever it is 27, 28, 29, right? That's the point where everything has to be working and kind of generating revenue. Or it's going to kind of starting to get choked off a little bit. And we'll be in the same place we are with self-driving, right? We're the people that have another business that can kind of keep that thing going. We'll survive. You know, I would guess Boston Dynamics and Hyundai will still be there two or three years. I would guess that ? anyone else that can kind of attach themselves to a very, very large cash producing business will still be there. And everyone else is going to be where the self-driving companies are.

 

Brian Heater (39:21)

Invisible AI was founded 2018. Do I have that right?

 

Eric Danziger (39:24)

Late 2018, really early 2019, but yeah.

 

Brian Heater (39:27)

How much did the pandemic speed up your own timeline for, you know, in terms of like manufacturing, but also obviously like the supply chain shortages.

 

Eric Danziger (39:37)

I don't know that the pandemic necessarily sped anything up, but it did, I think, show everybody the value maybe of having, of kind of interacting over video, right? Of using video, right? I think people got more comfortable with these tools, made it easier in some ways for us to kind of interact with our customers because everyone is well conditioned now to Teams meetings and all this kind of stuff instead of having to fly out every single time and do a meeting. The reality is that we are in a pretty dynamic environment, I would say overall. Right. think the kind of geopolitical environment that we're in now as its own pluses and minuses, right, the kind of insta-

 

Brian Heater (40:14)

The is a really nice and diplomatic way of putting it. 

 

Eric Danziger (40:18)

It changes, right? A lot. Yeah. So, so I think, uh, you know, I think there's a lot of advantage, right. And being. Flexible and helping reduce costs. Right. I think that's kind of a big focus for us. so, um, at the end of the day that we are in a really good place. We are something that people need, right? This is a time when you want to have really, really strong visibility into how your processes work into how to reduce costs and to, um, on a kind of improved efficiency across the board, right? COVID was kind of, in a lot of ways, a big shock and kind of woke up the system. But I think that the other forces that are out there kind of reshaping global trade and reshaping where we manufacture things are ultimately going to be bigger, right? But it's, yeah, I this is kind of one of the most interesting times from way to work in manufacturing and manufacturing technology. And, you know, we have… AI on one side, right? We have kind of redrawing kind of our entire kind of idea of global trade and what it means and where things should be built and how things should be built. ? you know, we have this kind of looming specter of changes in climate changes and how we power things, right? We have just a million different things, all that are out there, all that are trying to reshape kind of how we do the basics of building stuff. So, I don't know. It's going to be wild. It's going to be very interesting. 10, 20 years, trying to figure out kind of what the right mix of anything is. Um, and then of course, you know, in that entire time we'll just be getting better and better robotics and better and better automation. And, and then who knows, right? It just, doesn't seem like things are going to, uh, become less kind of dynamic anytime soon, I guess is what I'd say.

 

Brian Heater (42:12)

Yeah, so I'll put you on the spot to end things because we had this whole big conversation about predictions and self-driving cars and humanoids and everything else. let's say you said 10 to 20. So let's say 15 years from now, do you think that manufacturing will be much more meaningfully decentralized versus where it is right now? To build the things that build the things. Yeah. And yeah, things will be things will be incredibly different, but also exactly the same.

 

Eric, well, thank you so much for taking the time.

 

Thank you so much to Eric for taking the time to speak with us. Really enjoyed that conversation. Thanks to you, the viewer and or listener for viewing and or listening to this episode. If you like the show, please like and subscribe. And don't forget to check out our sister sibling newsletter ? over at automated.fm. And we will catch you next week for another episode of Automated.