Tech Brew Ride Home - (Bonus) Bryan Salesky, CEO of Argo

Episode Date: May 23, 2020

The founder and CEO of a company we talk about all the time: Bryan Salesky of Argo AI. Bryan tells us more about the unique Silicon Valley/Detroit hybrid that Argo represents, we discuss the unique business model strategy they’re exploring, and we find out, where the self-driving space is now that Covid-19 has basically put everything on pause. Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to another week on bonus episode of the Tech Meme Ride Home. I'm Brian McCullough, and it's a big one, everyone. The founder and CEO of a company that we talk about all the time, Brian Selesky of Argo AI, joins us today. Brian tells us more about the unique Silicon Valley slash Detroit hybrid that Argo represents.

We discuss the unique business model and strategy that Argo is exploring, and we find out where the self-driving space. is now that COVID-19 has basically put everything on pause. Brian, I really need to start with the most pressing question at this moment in time, because in terms of what I've covered on this podcast recently in this space, has been, you know, basically everything I've heard is that in terms of the actual research and, you know, on the road testing and things like that, everybody in the industry has had to put that on pause. So I'm just, if I could get a sense of for you all, if that's true, is everything on pause,

what is the state of your work at the moment? Yeah, it's a good question. We've been off the road since around middle of March when it became clear that, you know, the right thing to do is to go ahead and quarantine. and we were very proactive as a company to, we always want to prioritize the health and safety of our people. And it was clear that this was, you know, the world was going to change from that point forward.

We didn't know how long, but it was going to change. So we shut down a lot of our vehicle operations that were on the road. A lot of those people ended up working from home, actually, and helping us with a number of other types of activities. And they continue to do so. So we're here towards the end of May, and we're just now starting to see vehicles be able to go back on the road in some cities, not all of them. So that was going to be my next question. For how long are you going to be on pause?

But like everything else, seemingly, you guys are starting to slowly get the gears back going. We're slowly getting back. We've been able to resume in a very limited way in D.C. And here in Pittsburgh, here in Pittsburgh, we actually were tapped on the shoulder by the Allegheny County Health Department. They needed some help transporting test samples for them to be flown out to a laboratory in California for processing. And they said, hey, we need to get to a bunch of locations in a sort of fast amount. of time, bring them all back, and they need to get on a plane by a certain time each day. And we said, sure, yeah, let's do it. And it was a great way to combine what we're doing and

also help out the community. Well, I've heard in the interim, other folks have been saying they've just been doing everything they can in simulation and such like that. But is that just a placeholder like, has this, to what degree has this pause setback timelines and things like that for you guys. I assume there's only so much that can be done in simulation that like the real meat of everything is when you're actually on real roads in the real world. Yes, simulation can only get you so far. The answer is an end. You need both. A shutdown for a couple of months is not really that big of a deal. We had, we had been operating for several years now at this point across many different cities. All that data that we had brought back were able to use in simulation and get, you know,

we're able to get a lot of value out of that. So our team has continued to be productive, the writing software, testing and simulation, testing against our both real world and simulated data. And that helps give us confidence in the software that's been written over the last, you know, eight weeks or so, six to eight weeks. I would say that you can't do that indefinitely, though.

So I am thankful that we're able to get back on the road. I think we're all anxious to see, you know, what the performance is as we put it on our test track to see kind of how it compares to what we were seeing in simulation. It'll be actually pretty informative because we've never had a shutdown this long. So this will be interesting. But so it does sound, it sounds good in the sense that it's not like you're saying, oh, we had to throw out all of our metrics for this year because of this couple months pause. But yeah, okay. No, and our, if anything, we're sticking to the plan this year where we've told the team they know what they need to hit. We reprioritize some work in order

to that lends itself better to simulation. But at the end of the day, the work's got to get done. We're going to get through it. So I've been really excited to talk to you because, in my opinion, you all are one of the more interesting stories in this space because, you know, I come from tech. I actually lived in Detroit for five years in the early 2000s, and I was the weirdo tech guy among all these car guys. And what I love about your, because the narrative has always sort of been, like, for this technology, there's the Silicon Valley approach, and then there's the Detroit approach, and like, who is going to win out in the end? And I feel like, you know, both sides have their strengths and weaknesses in terms of getting this technology out the door and, and

ultimately doing something that will have an impact on the real world. But I feel like your story is really like kind of a best of both world story. And you've done the stints in the valley. You've been at Waymo. You know, you're in Pittsburgh and Detroit. And so I'm curious on your take on the different cultures, at least when it comes to tackling this problem, this engineering problem. Sure. I certainly did my tour of duty out west. I've also, you know, done quite a bit here in Pittsburgh and Detroit. And certainly there are differences, the weather chief among them. But beyond that, I think that what I've actually found is a lot of common ground in that a lot of the folks that work for us are very mission oriented. They're mission driven. They're driven by the idea that there's a

better way to operate vehicles that addresses the safety issues around human-based driving. And I think especially in the challenging miles that there are to drive. And I've found that folks in the Valley versus here, it's no difference. If you're mission-oriented and you're doing and you want to be part of this, there's really no difference. Certainly there is a little bit of, I think people come from different backgrounds a little bit. I find a lot of folks that are in Pittsburgh or Detroit. I think they've got a little more of maybe an appreciation for the auto industry

and kind of what it means to take, you know, to manufacture really big, complex things at a heck of a scale. I mean, a lot of folks don't understand that, you know, modern auto manufacturing, That's taking 30,000 parts and putting it together in an awfully complex system. There's quite a bit that goes into that, right? I think folks that are like in and around that industry probably just appreciated maybe a little bit more just because they've seen it, they understand it, they've had family that. Can I interrupt you? Yeah, yeah.

There was recently this wired piece that was partially a profile of you, interview of you. And like there's an anecdote in there about, I can't remember what it was, but like some sort of demo car where engineers are trying to throw a bunch of like electronics in the trunk but then like some car guy comes along and is like you know by the way a key component of the the crash performance of that particular model is in is in the spare tire in the trunk yeah this is like you're your quote with something like we'll see uh you know software engineers wouldn't have known that yeah that's right i mean it's and this is probably not so much the case now where they're taking spare tires out of cars, right?

They put the fix-a-flat stuff or whatever in there now. But at the time for that particular make-model car, the spare tire did have, you know, does add to the ability to absorb energy in the rear crashworthiness of the car. And it's one of those things that if you're just sort of trying to prove a concept, that's maybe fine. But when it comes to shipping a product,

You have to think through those things. That's right. Well, so, I mean, we're kind of being a little joking about this. But in a larger sense, you know, and I was joking about being a tech guy in Detroit 20 years ago. But then there's also been, I've been telling for years my Silicon Valley people, like, you know, it's not just an intellectual problem that you're trying to solve. And, you know, automakers as an industry have a centuries worth of unique sets of skills that can get this technology to reality, right? So am I right in my sense of Argo as that like you're trying to marry at least from not just a cultural, but even from a strategic perspective, like both sides of that to solve this problem? Yeah, look, what's fascinating about robotics in general,

And what really got me into it is how it really requires teamwork and people from all different walks of life and disciplines to put these things together and to think through all the different important attributes and what the right way, you know, to find the right solution. I find that it's some of the most fascinating and rewarding design sessions that you have is when you've got a mechanical engineer, an electrical engineer, a software engineer, a technician, a mechanic. When you put all of those people with all of their extensive experiences and knowledge-based together, you can really come up with some very neat solutions. This sort of multidisciplinary thinking is hard to find in other companies just because a lot of times

you don't need all of that. Right? Sometimes if it's just if you're building a pure software product, you just don't get exposure. But what's cool about this industry is you need all the above. So another thing that I respect about y'all and your model is that, you know, you've been more open about questioning. Like you're saying this is so new.

We don't know necessarily at the beginning what the business model is for this. And then late last year, you started to come out and started to answer that question, which for you guys, and I'm going to simplify here, is essentially to use like a tech analogy, is you're going to be the platform. You're going to charge per mile. And you're going to allow other people to create whatever business models on top of that as they can. Is that, am I oversimplifier? Is that generally what you guys have settled on at this point?

No, I think that's right. And certainly there's a lot of different ways in which that can be done. It doesn't even necessarily need to be per mile. But I think the idea that there's unit economics around this where you don't need to own this really expensive asset that sits 90-some percent of the time in a driveway or in a parking lot, that there's a different way to still get personal mobility, without owning a vehicle, I think is super powerful

and is where you're going to see some of the first applications and deployments of self-driving cars. And you're happy with your partners so far, Ford and Volkswagen as, you know, 100-year-old and more automakers. What you just described is not owning an expensive asset. That's like a fundamental change

in their business model for a century. you're happy with their willingness to explore these alternatives that this might entail. Oh, absolutely. I mean, they've been wonderful. I couldn't ask for better partners. And I think that the thing folks need to understand is that they don't, by and large, it's challenging for any automaker, any automaker to sell large volumes of vehicles in the middle of a city just because there's so many incentives and it's only getting worse.

But there's so many incentives not to own, right? Everything from parking to taxes, insurance, and so on, it's quite expensive. And so this is really just a natural progression of things that, you know, many of these automakers, especially ones that have been around for a long time, they've had to go through many transformations. I think this is just one in a long line of them. when I started this show two and a half years ago and every time I would do a autonomous vehicle segment I would sort of have this running joke that like the wager is we'll see self-driving cars on real roads in a real way by 2020 because that's what I had always heard for the last five years or so and and I know that generally

in the industry that's sort of you know by necessity that's sort of been tamped down a bit recently And of course, the COVID situation is changing that as well. So I'm not going to ask you when you think that my bet is going to pay off. Thank you. But yes, yeah. Let me ask you this way. When you started, not Argo, but when you started in this technology, did you assume that by 2020, like a lot of other people,

that we'd be further along than we are now? No, I was never really sure. You got to remember, I came into this, not so much as a diehard roboticist, but as a software engineer that came out of a totally different industry that was kind of interested. I wanted to learn more. But I think I was always a little bit of a skeptic in the earlier days. I'm not anymore, obviously. I mean, I think it's no longer if it really is a question of when. But in 2004, the world was very, very different. That's when I got my start at Carnegie Mellon. What was interesting to me,

was. I mean, it just sounded almost inconceivable that you could even make something like this work to me, somebody who didn't get a PhD in robotics, and I'm surrounded by PhDs, right? I'm surrounded by people who have already had anywhere from five to eight years of experience in this, if not more. And in some cases, people who already had 15, 20 years of experience doing out to automation. And it was really eye-opening for me at how little in 2004 actually really worked very well. And I realized early on that there was a long road to travel. And I actually thought that the first deployments was really going to be some sort of driver assistance that would make use of a lot of the AI techniques we were using then.

And it turns out that that actually has its own set of issues, right? And I believe the path we're on is the right one, but it's going to take some time for us to get there. And that path is full self-driving, no involvement from a human. Right. I've heard that from other folks, which is that instead of just doing like some super cruise control or whatever, a lot of people realize at the same time that you can't go half measure, you've got to go full measure. It's not going to work. You've got to solve the problem entirely before you even try to deploy it. Yeah, and I think there's a world for both. The thing that you can't count on so much is that a human is going to be paying attention. all the time from a from a from a let me let me say that part again. Sure. The thing that you can't count on is for a human to be involved in another task like reading,

email, something else. And then it'd be expected to come back and in a matter of just a very short time span, a fraction of a second, a second, be able to then resume control of the vehicle. Because you have to regain some sort of situational awareness of what's around the car, right? That particular state is a very difficult challenge to solve. I'm not going to call it impossible because I've learned in my career that, you know, there's always a potential solution out there. But, you know, there's really a stark contrast,

and I think the industry needs to get better with the vocabulary and be more precise, right? that there are systems that you must pay attention all the time and monitor the system, or it is the system takes over for you within some sort of area that it's able to operate in. It's one or the other, that in-between state where you can check out for moments of time, but then the system has to re-engage you, that's the hard part. That's really hard. And I think back in 2004, we didn't know enough to the, you know, we didn't know enough to that that was going to be problematic.

Well, and that lines up with what I've heard from other people on the show, too, which is that, and correct me if I'm wrong from your perspective, that the biggest thing right now still holding back this technology is the human element in the sense that, like, you know, if you could construct a perfectly closed system, like the algorithms can handle that or whatever, but it's the fact that there's pedestrians that don't cross a crosswalks, that there's, you have to be aggressive and nudge your way into traffic sometimes. It's the, it's the fact that we still will have for the foreseeable future human actors in the equation,

that that's the thing that is the hardest nut to crack still. That's right. And just put simply, how do you, how do you know what's going to happen three, four, five, six seconds from now? That's, that's really difficult to do. Now, it turns out that as humans, we have evolved to be really incredible predictive systems. We are able to anticipate from small little movements, whether somebody's paying attention or about or a little bit antsy and want to jump out, right, even though it's not their turn to maybe go into a crosswalk. As humans, we pick up on very subtle cues.

And the challenge is how do we get a computer to do the same thing? And the good news is that in the last several years, I think we have techniques to do it, but it takes a huge amount of data to sort of train these algorithms so that they're accurate enough to be effective. I had Gary Marcus on here, the AI researcher, and his whole big thing, I don't know if you know him, is that AI hasn't developed real common sense yet. And that sort of sounds like the same thing, is that what you're describing is thinking, intuiting what would happen six seconds from now, you can't know, like a computer wants

the perfect information to make a decision, but common sense is I have enough information to make a reasonable decision. That's well said. That's right. So is that another holy grail for cracking this? It's just giving these cars some sort of common sense? Well, we like hard problems. There's also the saying that common sense is not that common. Well, I mean, it's not like human beings, even human beings with this common sense don't seem so smart sometimes when you're out on the road with them. Well, this is, this is absolutely solvable with what we know how to do today. Just the challenge, though, is that it requires, it requires a lot of examples. And when I say example, what I mean is it requires presenting the software with a lot of different versions or scenarios of, you know,

people and bikes and other things in and around the car doing different things in order to refine this prediction algorithm. And that's why it's kind of going full circle now in our discussion, but that's why road testing is so important is you can only simulate that to a certain extent on a test track. You really want to get the real world examples because there's just going to be so much more variation. Is it one of those things where it's like that weird tipping point where you can't get these robots out on the road in a large scale until you've got it as good as you can possibly get it. But once you get it out there, and so then there will be millions of miles driven every day, then as soon as that can happen, like this system will perfect itself overnight.

Yeah, I think that's, I think that the scale helps. But at the end of the day, the data that comes back still needs to be handled with some amount of manual care. It's not like this thing becomes an all-knowing being and learns on its own as you feed it millions of miles every night. It still requires curation. It still requires analysis. It still requires a lot of validation to make sure that it's learning the right thing. So, you know, oftentimes there's this like fallacy out there that you need billions of miles in order to, in order to train these things. And, you know, that if you just had a fleet of millions of test cars, you could build the algorithm tomorrow. And it really doesn't work that way because of that time and care that's still involved. So there's, there's actually the opposite effect that you could actually drown yourself in data and slow yourself down. It's about getting the right data, the right set of variations that's really powerful. That's an interesting idea that someone just turned me on to recently, and it was about this concept that a lot of algorithms have been broken recently by the COVID moment because our

behaviors have been so outside the norms. Yeah, absolutely. The algorithms are, you know, throwing their hands up in the air. So that makes me think, like, is there a potential scenario where even when AI autonomous vehicles are on the road in a major way, even 30 or 50 years from now, like that will be a job. There will still be engineers somewhere that will always be tweaking stuff. Yeah, totally. We're always going to be improving and tweaking things,

and there will always be another mile to map to test on, to understand. But you're totally right. I mean, on the COVID thing, that's really interesting insight, but it's true, right? we've we've we've we've sort of tripped the the the circuit breaker on like every single model because the behavior is completely changed i mean look at the supply chain it's totally turned upside down that the endpoints are are more and more distributed it's not businesses it's it's homes now right as everybody's working from home living at home full time uh it isn't it was never wired to do that from the beginning and so everybody's having to adapt i'll be honest with you i mean i look at what's happening and and um it's incredible how well we have adapted, isn't it? I mean, I think it really is. Well, not only, you know, the supply chains haven't broken down as much as they could have. Right. The internet hasn't broken down seemingly at all. But, you know, the key of that is, is that the system was optimized for the

paradigm that we had for, I don't know, at least a decade. And so when you're talking about things like supply chains, it's a matter of, you have to design your systems in a way, not just for black swans, but to allow for, yes, you can optimize for efficiency that'll be efficient 99% of the time, but then you're going to lose a ton of money if you can't also optimize for that 30% anomaly or that 90% anomaly, you know, that sort of thing. That's right. And with self-driving, 80% of our work is dealing with the black swans. Right. All right, this last question is way out of left field. And so if it's, If it doesn't hit, I might even completely cut it. But I recently got turned on to that Denzel Washington movie Unstoppable about that runaway train.

Yeah, I've seen that. That's good movie. Well, so when I was researching you, like, you early on worked at Switch and Signal, which is... Switch and Signal, yep. So, like, your whole job was, like, controlling that sort of dark territory, like, those stretches of track that aren't governed by signal systems. Yeah. So we managed systems that kept track of kind of where those things were. They call them sort of permits, if I remember, because this is a long time ago.

But absolutely, yeah. I mean, there was these massive command centers that had visibility into railroad operations where it was lit up, so to speak, but also where it was dark, where there was no equipment. And so you had to keep track of that stuff very diligently in the command center. So number one, my question would be, did you feel like that movie? I mean, obviously, it's an action movie. It's a popcorn movie or whatever, but did they get at least the reality of managing trains a little right? I think they did from a crisis management, incident management perspective.

Obviously, everything's dramatized and exaggerated for movie-like effect. But there are command centers that are very similar to what you see in that movie. And in fact, the first project I worked on was actually for the Port Authority in Allegheny County here in Pittsburgh to revamp their control system for our version of a subway, which is called the T. And it was just, it was a fascinating experience because you realized just how important software is in areas that you just never really thought about, right, as an average kind of consumer. there's so many mission critical systems that run out there that software is absolutely essential. And so it certainly served as a good foundation for what I do now. Well, that was my second question. What was in essence, that problem, that intellectual problem, is that what you've been doing ever since just like blown up into infinity?

Oh, for sure. Yeah. You know, I joked around with a collie here. With the quarantine, I've had an opportunity to get a lot of time back in my life because I'm not traveling anywhere. So I actually have been, I've actually been doing some stuff in Python, a little bit of programming, and like little things that, and I realize I'm having a lot of satisfaction from it because this is actually something that totally has no, like, safety or mission criticality to it, the stuff that I'm doing. It has nothing to do, really do with the

self-driving system. So I'm thinking, wow, this is a pleasure. It's a nice change of pace. I've never had I've never had this luxury before. Well, another word for that, Brian, is fun. Yes, I guess. Well, I think it's all fun, but... Well, consequence-free fun, maybe. Yeah, that's right. Brian, thank you so much.

Thank you. It's great being here.