Safely Scaling Autonomous Vehicles with Reid Hoffman, Chris Urmson & Stefan Heck | Greymatter

So with that to intro each of their companies, I'm going to ask how they ended up deciding to start them. And I'll start with Chris, which is you've been down this road once, right? Yeah. Recognize it has a lot of potholes, bumps, construction, a bunch of other things. What made you decide to start Aurora?

I left Google after about seven and a half years. It was an amazing place to be and I see a lot of actually familiar faces in the room here from folks I knew at that time. And I left, I was kind of had enough of that place and it was time to go find something else to do. Didn't know exactly what to go do. Spent two, three months talking to interesting people and getting a sense for what could be out there and kind of taking off the rose colored glasses that you wear whenever you're inside institution and looked out of the world and saw that there was this really unique moment in time where the automotive industry was facing an unprecedented set of challenges and looking for help. Where the technology was close to ready but not there yet. So there was an opportunity for a new company to be built. And if that company had the right ethos, we could attract the right people and had the right model of working with industry, we could actually build something meaningful. And so that seemed like a good time to take a shot on goal. And you know, we've been building a company and really glad we did.

So Stefan, same question with a different vector. So as opposed to I've already gone down it, discovered all of the places that this is actually a super difficult problem that I'm beating my head against and I'm going to do it again. Yours is instead, okay. I was building this whole like kind of resource revolution right off the book. Kind of like how to have green practices in business be actually the better practices for business. But now I'm going to go to this really early stage wacky tech startup scene and jump off that cliff.

Do it from scratch.

Yes. So what led to that?

Well, two things really. I was at Stanford as you described, and I would bike from my house to Stanford every day. And I was doing research on disruptive technologies and how they scale and deployed. And one thing you realize very quickly is outside of Google and a few companies, all of the data that's being used in transportation is on average 5 to 10 years old. So if you look at the government planning data that we're using to design highways, to design our infrastructure, the traffic models that inform our public transit, none of them include Lyft or Uber, none of them have any concept of autonomy in them. And usually the data is very sparse and very old. So that was the intellectual reason to say there's got to be a way to get real time data to actually inform this transition, driving both sharing and autonomy and optimization. The more visceral one is very simple. At least once a week, while biking to campus, which is not that far, it's like two and a half, three miles, I would nearly get run over by somebody who was distracted on their phone. And so it was that classic entrepreneur like, there's got to be a better way to do this than have people drive while looking at their phone and kill or nearly kill people, especially people on bikes, who.

That's a little bit different. I'm not solving it. I'm solving it so I don't get run over. I'm solving it so I don't get run over.

That's right.

So how would you guys describe the current state of play for what the road looks like for getting these vehicles network connected? And why is being network connected so important? And then what are the kind of some of the key things to look ahead all the way from current distraction, which I guess we'll start with, to autonomy?

First of all, it's an urgent problem, right? It's an epidemic of road fatalities in the US Bigger than all the things that get all the newspaper attention, terrorism, war, diseases, even the drug epidemic. Far more people die in automobile crashes than any other cause of death. And so we've got to solve that problem. There's definitely the potential to solve it long term with autonomy, but there's also an opportunity to solve it more near term by actually just making human drivers better and taking the first 25, 30, 50% of those fatalities out of the system. Of course, there's opportunity along the way to learn how humans actually drive, which I think we'll come back to because that's a very interesting dimension. It's not the classic DMV rule book, right? While you may have read the DMV rules to go pass the test at some point, that's not actually how you learn to drive. You learn to drive by sitting in the car with somebody and trying it out and gradually learning how to perceive. And so that's, I think, the stage we're at. I would call it the teenage years of autonomy, because we've got a lot of early vehicles out there, some more mature than others. Obviously none of them are flawless, yet they're a little bit like teenagers who make lots of mistakes while driving it takes about six years and 2.8 collisions for the average American to learn to drive. So if you have kids, that's what you have to look forward to.

Thanks for that.

Yeah.

And really excited about it.

One of those is usually totaling the car. So you better hurry up, Chris, see.

What we can do.

Yeah. My daughter's 11, so I hope she will have autonomous cars before she'll beat her before she's ready to drive. I think yours are a little bit older, so you're getting time crunch. But what's important is, and this comes back to your question about connectivity in the human case. Each of us has had to do that on their own. Maybe you had a parent, maybe you had a friend take you out, but that's kind of it. There's no scale effect in the world of autonomous cars. There's a huge opportunity for every car to learn from every other car's experience. And that was the notion of Fornado creating a shared data platform where all the anonymized safety data actually is available to everybody and everybody building autonomous cars. We can all learn from both the good driving of the best drivers and from all the screw ups that happen on the road every day.

Do you think teenagers is the right description for current autonomy? What is the road ahead?

I guess I wouldn't have used teenagers. What we're trying to build at Aurora is we're coming at the network side of this a little differently. We're thinking about can we build a platform that spans across the industry and supports it, and where we can leverage the lessons learned and the experience of those vehicles out there in the fleet, bring that back and make all of those drivers incrementally better. And from our perspective, the opportunity to dramatically improve the safety and efficiency of something as fundamental as transportation is just too big an opportunity intellectually to miss out on and fun to work on. And we hope ultimately it'll be a huge win for society and a huge economic win.

So one of the challenges, of course, that we all know, is that it's impossible to get to zero, fatality, it's impossible to get to 100% safe, etc. And so one of the questions in navigation is that people tend to go, well, new is dangerous, and I don't understand how new is not dangerous unless it has zero, which is you get regulatory stupidity, you get bad reflexes. What's the way? And I'll start with you, Chris, and then move back to Stephan on this one. What's the way that we as a society, we as governmental Leaders, we as business leaders should try to help navigate this and set the benchmark for when should we really be saying, okay, this is the time that we press accelerate?

It's certainly not my place to answer that, right? This is a conversation we have to have as a community, as a society. But I think the first step is recognizing that the status quo is broken, right? 38,000Americans every year, 1.2 million people worldwide. If we wait for perfect, that will just get worse. Unless Nauto comes through and cuts it in half, of course. But I think without that, without some kind of technological impact on this, it's only going to get worse. And so we cannot let the perfect be the enemy, the good. I think from a very pure statistical point of view, the day these vehicles are incrementally better than the average driver, then, you know, if we were completely passionless about that, that's the day we should put them on the road because they will get incrementally safer from there. And that'll be one life saved, that'll be 10 lives saved, that'll be 100,000 lives saved. Very quickly there is this challenge that as people, we associate with individuals, right? And it's much harder to talk about the thousand lives saved than it is to talk about, you know, little Sally who ran out in the road and something tragic happened. And so we need to build that relationship and have that open relationship with the regulators so they understand. They already understand the size of the problem, but they understand that we're approaching it reasonably. We need to have this dialogue with the policymakers. And I think it's for other folks in the room, like the media here, to help contextualize these events so that the general populace can kind of understand what's going on and understand what the pros and cons are in a really balanced way, not sensational way.

I agree with everything you said, Chris. I'd add two other nuances to it. One is we talk about the average human driver. That's a fiction, right? The spread of human driving ranges all the way from 20% of you here in the room may go your whole life without ever causing an accident. You might get rear ended, you might be involved in an accident, but it's not your fault. And 15% of you statistically are going to cause 80% of the damage out on the roads. And so I agree with the notion of the average, but we should take the really bad 15 off the road much more quickly because autonomous cars are going to be better.

Actually, by the way, this may be the first time, where the 80, 85% of Americans are above average. May actually be accurate.

May be accurate. Although the problem is most of them think they're good and they're not really. So that's one challenge. The other one I would say is geographic. Right. We see in the data, first of all, driving is very different by geography. Boston versus San Francisco, downtown versus a suburb. And so I think this is really, it's not a black or white. Do we turn it on? It's by geography, which of course you've done very carefully both in your past and now. And I think that's the discussion we need to have when, under what conditions, in what geographies for whom do we turn autonomy on? Because we can make a big difference relatively quickly here. But I think if you're aggressive and careless and don't have proactive engagement with regulators, you can do something that's, that's well beyond the capability today and that will cause a huge backlash. Right. Regulation happens because of Sally getting out in the road or a school bus getting hit and that gets disproportionate attention to the fact that yes, in the ordinary traffic circumstances, a lot of humans killed other humans. And so I think we have a responsibility as participants in this industry both to explain, as you said, but also to advocate for an approach that doesn't get out ahead of our skis for what's possible today.

Absolutely. We need to be looking to bring the technology to market very responsibly.

So both of you have also taken approaches which are like how do we partner and work with the OEMs? But describe a little bit about how you see the OEMs partnering with them well and how that will actually help solve these problems.

Making cars is hard. Right? And having a startup is hard. And so we think we can be world class, best in the world at delivering the self driving capability, that platform. But we're not arrogant enough to think that we could be best in the world at building cars or building a transportation network or all these other slight, you know, as an insurance company. So let's go out there and focus on the one thing we can actually do well and let's go find other great companies that will allow us to get to market quicker because they know how to do that already. And so that's been the way we've approached industry today and it's certainly not a panacea. Right. The culture within our company is very different than the culture within our partner Volkswagen. They're both amazing companies, if I say so myself. But we operate in different ways for a variety of reasons. So putting the energy into those relationships and managing the interface and managing the expectations is a hard job. And it's something we have to work through every day with them. I completely agree.

The skill sets are very different. To really get a full autonomous car system deployed takes a lot of very different capabilities. Running the network, managing the user demand dispatch, building the vehicle itself, and the intelligence. Like Aurora, we're really focused on a piece of that system. Right. The cloud and the higher levels of the intelligence in the vehicles. How do the vehicles coordinate? How do they learn from each other?

Agreed.

What's exciting to me is they do all believe, as you opened up, that this is a once in a generation or once in a couple generation transformation, that they need to be out ahead. And that's part of how we selected the partners. Right. The ones we work with are investing heavily, see this as an upside opportunity and they understand that it has to be a safe deployment and they have to share the safety data to accelerate. Because to really get to safety takes hundreds of billions of miles of experience ultimately to get to that, not just to as good as humans, but to get near perfect. And if everybody has to do that on their own, it really takes more risk, more time, and we let people die along the way. And so the reason they're working together for the first time in 100 years in some cases. Right. I mean, not often do you see GM and Toyota working together. They did in the Nummi plant, but that was a very short period of time. And then we've got BMW and a couple other OEMs that are not public yet. They are excited because there's this potential to accelerate all of their progress and yet build differentiation on top. And that's a very unique approach, but I think it is fundamentally a collaborative approach that says we have to move as an industry and do it together.

I think this room knows this next point, so I'll ask us to go a little bit deeper because most people who aren't in this room, who aren't deep in the autonomous vehicle revolution, the platform change, the what can change in society think, oh, like, you know, tomorrow it'll be all av.

Right?

And everyone who's knowledgeable and thinks about it says, no, no, Right. There's a whole like, how many vehicles do you build up? How do you get through the learning curves? How do you do the regulatory stuff? How do you get the partners with partnerships with oem? So what it means is there's going to be Many, many years of a mixed environment. And we already touched on it a little bit with the safety expectations. But what are the other things that people should think? Okay, here's how we should be navigating a mixed environment. And this is what it means, both from the human driver side and from the AV side. And we'll start with Stephan on this.

So you talked about the long period. While there's human driven autonomous cars. We're a mixed environment forever, unless we all agree never to walk, bicycle, walk our dogs, anything like that on the road and across crosswalks. It's fundamentally a mixed environment. So the exciting thing to me is if we think about it that way, you start thinking about how do the vehicles fit into the rest of the human environment. And on the one hand, it's exciting because it allows us to rethink cities in some of the ways that is part of the excitement of this. You free up parking, you free up road space. Now suddenly we can decide we have more parks, or we have more housing, or we have more green space or other public facilities. So that's an opportunity to redevelop cities that we haven't had in about five generations. The last time that really happened was streetcars and elevators that transformed cities at the turn of the 18th to 19th century. The other piece of that, 19th to 20th century. Yes, the other piece of that is how do we want the vehicles to behave? Because if you look at real urban driving, not highway driving, highway driving is pretty prescribed, right? You got to stay in the lane, you can't cut other people off. But otherwise it's all guided. But in real city driving, even just my ride coming here, people cut each other off, people yield. There's all of the normal social dynamics about who's higher status, who's in a rush, who's aggressive, who's friendly is in play. And so if you watch even just a simple thing, four way stop signs, yes, there's the rule of the law that says whoever arrives first shall go first. There's actually a second rule that I found more than half of Americans don't know, which is if you're synchronous, then you yield to the person to the right. That's generally not well known in the population. But the reality is, it depends what kind of car do you show up in? There's people who are friendly, who wave you through. So those are the interesting dynamics. And that's why I think we actually have to adapt this technology already to the human world. Both because there's this 20 year transition period, while there will be human driven cars, maybe even longer. But also because actually we want that because it's not going to be safe. And you can see it in the autonomous cars that just play ultra safe, right? They actually will wait for the crosswalk to completely clear because a backup behind them, which has its own set of other ethical considerations, congestion effects. But the other thing that happens is somebody will try to go around them. And we've already seen, even with the autonomous cars we have today, that there's humans gaming them. So unless they actually know that, and that factors into the way they're designed, we increase risk. Another great example is speeding. All of you break the law. I know that for sure because basically on most roads, humans drive a little bit above the speed limit. Now our infrastructure engineers actually know that. So they set the speed limit at the 85th percentile of the average expected speed, which means they're all expecting 10, 15% speeding as normal human behavior. Now if we deal with the regulators by saying, well, the autonomous car will always follow the law, sounds good at first glance, but actually a really bad idea. Because what happens, they're the bottleneck. Everybody else tries to pass around them. You get these really crazy high speed passing maneuvers that increase the risk. So the best thing for the autonomous car to do is like the rest of us, cheat a little bit, go a little bit faster, but be in the flow of traffic. So that's where I think it's interesting. How does it fit into the human practices?

Yeah, yeah, I think I would completely agree. I think the next 20 years is going to be the hardest. This is most of the energy we put into developing this technology is really about dealing with people doing things that aren't normal. Right. If all the other human actors in the road were staying in their lane, behaving the way they're supposed to, taking their turn at the intersection, now, we would have solved this a few years ago, and you know, Bob's your uncle. But the fact that we have to deal with the fact that that person will at the last moment make a swerve in front of us, that's really where the hard part is. And that's what leads to the extra sensing on the vehicles, the extra software computation. My disappointment and optimism is this is as hard as it gets. And then over time, these systems will get simpler, more accessible and more available.

I'm waiting for the day when the humans actually learn to drive better from the autonomous vehicles.

I completely agree that there's going to be a reckoning that Happens where people see these vehicles on the road and behave a little bit different around. But I'm also actually optimistic about that because one of the things that's thrown out there is that the self driving cars have to behave like the other people in the city.

Right.

We have to follow their driving conventions. Well, the amazing thing about people is they are incredibly adaptable. The secret hypothesis I have is if we drop a whole bunch of good drivers, good self driving cars out in the world, you know, the human drivers are going to react to that and be like, okay, well I'm actually not going to make this crazy left turn here that I shouldn't. Right. They'll start to kind of blend in. I, I'm, I'm optimistic more than you are, I think.

Well, one of the things that. Funny, this conversation is actually making me think about one thing I hadn't thought about before. Which is good because I've thought about for example, the fact that humans will actually be semi psychotic against AV cars. Right? Because they're like, oh, I don't care, you know, some humans, yes, some humans, but by the way, some humans who are on the road. Right. Relevantly, there is an interesting thing. Given that all these sensors will include cameras, as you can see, the aviation network actually in fact developing a reputation score as to which of these drivers are capable or not, et cetera, et cetera, that are off them and actually in fact adapting to actually the human driver behavior. Do you think that's a likelihood, a weirdness or a. I think it's going.

To threaten insurance even in the next few years because there's this huge asymmetry between the record you have as a human driver of what you did and the record. Every single AV has what happened. Right. And so if you're not gaining data from the vehicle to prove what really happened and video, all the other sensors are going to be amazing evidence that's going to tip the scales. And I am longer term optimistic that humans will learn good driving patterns. I'm skeptical that near term, while there's a small number of AVs that the humans will collaborate in a good way. Yeah, we'll see lots of testing of. Can I cut them off?

I wonder about that. One of the theories is oh geez, you'll never be able to go down the freeway because people will walk out in the freeway because they know the self driving cars are going to stop. Well, it turns out if I'm driving down the freeway and somebody steps in front of me, I'm going to stop right there's. A social norm that prevents people from doing that today. And there'll still be people in these vehicles riding along and they're going to be just as pissed off about the person blocking their road when they're in a self driving car as they are when they're driving themselves. So I think there'll still be a social regulatory mechanism that's kind of keeping this under control. But I'm also not a psychologist or a sociologist, so we'll see how it plays out.

It'll be interesting to see how the adaptation goes. I mean, it is interesting because we'll now actually have a map by which we can actually in fact say, here's actually in fact good and bad driving behavior as we get more and more cars with all these sensors. Because there will be a way to backtrack to that with the equivalent of body cam, kind of like, oh, here's actually interesting evidence.

And I think this is really powerful because the fact that we can actually understand what happened. We know human recall of these high speed events is really bad. And I think this is part of the premise of Nauto. But the fact that we'll have these on these vehicles, it will allow us one to make the vehicles safer because we'll understand what happened. We'll use that data to train the vehicles and improve them. But we'll also be able to resolve conflict and say, you know, actually the vehicle did make a mistake, let's fix that and you know, make the world safer. Or you know what, like there was no way I could avoid that guy blowing through the red light and crashing into the side. Here's the light, it was red. Let's attribute responsibility where it belongs.

The other thing that's interesting about that is we as humans misjudge the risk, right? Not only that, we don't track the high speed events and most crashes are over in a tenth of a second or a twentieth of a second. Even something as simple as tailgating, we don't perceive the elevation of tailgating risk correctly. So we have lots of examples of nauta devices in vehicles that have detected tailgating where we know scientifically the risk is elevated. Humans will go much closer before they think it's dangerous. And even if I show you video and evidence and braking distance, so it's hard. Same thing for distraction. Almost all of us think, oh, a quick glance doesn't matter. And yes, if it's one second or less, it doesn't matter. As you cross two seconds, the risk goes up exponentially. We have humans who are watching TV who are checking their phone, who are texting 12, 13, even 18 seconds at highway speeds. And so we underestimate just how bad that is, of course, terrible. And I think the autonomous cars will teach us that record feedback will show. And then I think the other interesting thing is what's the long cycle feedback to the law? Because our current laws don't ever text and drive actually doesn't make sense. Right. If you're stopped at a red light, there's almost no risk from you checking your phone or checking the map. But if you're doing it at 65 miles per hour for 18 seconds, then yes, please, please don't get off the road.

That's a terrible idea. Yeah.

So how do you think this is going to change the insurance industry? Right? Because obviously predictions about the future are usually wrong. But it's, you know, like how do.

You see game though?

Yes, exactly. So how do you see that with both the mixed environment and the increasing amount of av, how do you think insurance is going to change?

So my model is that driver insurance becomes product liability insurance because person in the vehicle is not actually responsible for the action for the most part. And that over time the cost of that actually collapses. Because unlike estimating the risk of me as a driver who you might get to sample 10,000 miles a year, and comparing to Stefan, who I'm sure is an excellent, safe driver, who probably also drives 10,000 miles a year, if you can compare that to a fleet of vehicles that have a consistent driver who are driving, you know, maybe 10,000 miles a minute when we get to deployment, you'll really be able to understand the statistical rate of these events and then they'll price that in. And companies that aren't willing to price that in, they're going to be non competitive in the marketplace.

I think there's additional effects. Right. You get the knowledge base of insurance today is actuarial tables of historical risk. All those rules are changing as we start inserting autonomous cars, even as we just start inserting ADAs and other safety features. You know, forward collision warning is significantly reducing rear ending. So that's a huge impact today on loss. Insurance companies aren't really ready to process live streams of real data, which is what they're going to get very shortly. So that totally changes the business model and the actuary will become less critical and the real time algorithm programmer. And then of course the knock on effect is, I agree with you, the pricing overall the pricing will reduce, but the dynamism of the pricing has to increase. Right. When and where are you driving? Each mile we're driving is not actually the same. If you drive six in market in San Francisco, you're 1,000 times riskier than crossing market at fourth in market or second in market. And so even down to that micro scale, there's actually huge variability in risk. And then of course you all know driving at rush hour is much worse than driving in off peak times. So there's a huge opportunity there actually price not only for that highly predictable autonomous vehicle driver, but actually to price for the actual route, the time and the destination you pick.

Yeah, it's kind of fun to think about the fact that there's both a massive aggregation benefit, but also a micro, almost a micro transaction benefit.

So I'm going to ask another couple questions. One of the things on the Internet was a serious worry about digital divide that be the digital haves and the digital have nots and that will actually have a path of social stratification. The natural kind of utopian view in AV is it makes like transport a lot cheaper, available to anyone, transform space in ways that even if because of the economics of real estate, you have to live a little further out, actually in fact that's no longer as horrendous and you have a free year market of mobility. Is there anything in terms of an equivalent of digital divide that we should be attuned to to make sure that we're steering more towards utopia and dystopia? There is.

I've been an advocate for years now for what I call asus autonomous, connected, electrified and shared. We should not deploy autonomy by itself because if Autona comes in as a high end luxury feature, we will go and see that the average people per vehicle will drop. We're at 1.2 today in the U.S. we could go. Which means you're stuck in 101 next to a car that's running errands for some millionaire getting their tie, dropping off junior at private school. Right. That's pretty much your dystopian scenario. And everybody else who can't afford that car is. Everything gets worse, Pollution gets worse, congestion gets worse, safety, depending on how many deploy, might get better, might get worse. If you start combining this amazing technology of autonomy but actually put it into shared vehicles that are on call. And I don't need to own a car for a poor household. Transportation is 25% of the household budget if they can pay for that per trip. Now people suddenly who can't afford a car to go commute to a job have access. We enable parts of the population that haven't been able to drive because they're too old, too young, or have other disabilities, now suddenly they have access to services in a way. I think one of the things we miss is in the world of autonomy, things will come to you rather than you having to go to them. So that enables people who don't have the mobility suddenly to have access. So deployed the right way is a shared infrastructure. And sharing, by the way, is also about the fixed places. Right. If we have vehicles in our basement that are shared in the same facility, that makes mobility access much better than if I only have that if I live in a, in a private home out in the suburbs somewhere. So this is one where I'm really optimistic. But we do need to give it the right nudge to say, don't just deploy it as an individually owned classic, traditional car model luxury feature, actually steer it towards integrating it with the rest of the transportation system, including other modes. Because having it integrated into bikes and scooters as one service enables you to have even more efficiency from the system.

I agree with everything you said and I think the one thing I would add is I think that this is one of the cases where the economics actually align really well. Well, right where initially these vehicles are going to be more expensive because we're going to put computers and lasers and radars and cameras and stuff on them. And by bringing that to market, amateurizing that across utilization in a, in a ride sharing business or a taxi business or whatever you want to call it, public transportation, I think you get a huge, huge economic win there. And so the economics are going to push it in that direction, which is very exciting.

So last question for me. Do you think that cities will ban human driving cars at some point, in which case the prediction of what the year range of that will be, and then which continent is that likely to happen on first?

Well, the second part's Easy Europe, because there's already cities that are talking about it today, Vienna, Helsinki, and they're doing it today by planning private driven cars. So there's only professional transit drivers. The step from that to saying only avs or shared avs is a pretty quick one. So I think we'll see that. To come back to your time question as early as middle of the next decade in just a few places like that, downtown urban cores, and some of these older cities, because remember, these are mostly medieval cities that weren't even planned for cars, let alone the scale of traffic that we have today.

Speaking of Vienna, speaking of Vienna, Vienna.

Is the City with the highest public transit modal share, 60% of all trips trying to go to 80, which has never been done before in the world. And that's why they're thinking about banning passenger, individually owned passenger cars downtown.

I agree, Don.

The tougher question is when will it happen in the US that one's hard.

So I think that there will be probably a city in the US probably at the same time frame. Right. There'll probably be a city in the US where they're like, you know what, we've seen it over in the city. I want to make it, you know, a flagpole thing to attract people to my city. Right. I don't think it's New York, I don't think it's San Francisco, but I think there'll be some up and coming city that says, you know what, we want to be different. We want to make this a distinguishing feature.

Pittsburgh.

Unclear about Pittsburgh.

So Elisa has a microphone, she'll choose a question or two.

Hello. My question is what are your thoughts on security? Because as Stefan is, you're working on fleet training, I assume, And Chris, you would be using that for training lots of your cars. Right. I mean the backdoor vulnerabilities are very large and if you spoof your external security, you know, you get access to your kernel and it can be very, very dangerous. So what's your approach on decoupling your mission critical and non mission critical aspects of your systems and ensuring. Are you thinking of a different architecture, operating systems or what's your thoughts on this?

You want to go first?

I would love to go first, absolutely. Security is important. Security is not, as you point out, an issue with self driving cars is an issue with connectivity in general. Right. And homogeneous fleets of software out in the world. Our approach is to think very carefully about the architecture in the vehicle, to bring best practices to bear to recognize. Unlike most things in the automotive industry, this is something that you have to be able to play push updates over time and you have to participate probably in industry standards or industry groups to track these things. But I don't think there is a magic bullet. If you have one, please come talk to me.

It's the real geek question.

This is. We do not have plans to go to a more microkernel approach today.

Only thing I would add is in addition to the control part, which I think the core of your question, there's other aspects of security. Right. What's left behind or in these cars or where are they sent? Yeah, that's a big question that concerns the government in terms of security right now, not having these roaming around in the wrong place with the wrong cargo. Yeah, yeah. And then we're also very concerned about the privacy element because we're collecting data on very large numbers of drivers. You don't want the individual attributes or their particular trips to be identifiable. Right. So while the fleet for us, the commercially will see its drivers, the moment it's put into the shared pool, we remove all of those individual identifl attributes. Which is actually trickier than you think, because remember, there's a lot of sensor data reconstructing who it might be. There's a lot of GPS traces. If I look at two days of any one of you as GPS trace, I can pretty much tell who it is. Right. So there's a lot of work that has to go into making that happen. And then we're very focused on not retaining data that has that risk. So I think how much you retain the learnings in a form that doesn't put individual security, security at risk is really, really important. And in very near term, like all the safety things, we also have an immediate security benefit because in a passenger transport service, we protect both the driver and the passenger from other passengers and from each other. Because you have a camera in place, we have an automatic system and a manual button that the driver or passenger can trigger to cause an alert to go off to the dispatch center. So that's bringing security forward for today and making it a little bit more likely that you'll be willing to share that autonomous car rather than have to own your own.

Okay, so let's please thank Chris and Stefan.

That was fun. Thank you.