Shahin Farshchi of Lux Capital

This is Smarter Cars, a podcast about autonomous vehicles.

This is the finish line. The Stanford racing team has made its way into the history books.

But the most important thing for me.

Is it actually doesn't matter who comes first.

It matters that we as a community achieve it.

Early in the technology, a thousand flowers should bloom.

Welcome to Smarter Cars. This is your host, Michelle Kairouz. In this episode, we talk with Shaheen Farshi. Shaheen has a PhD in electrical engineering from UCLA. He's been an entrepreneur, he's worked for General Motors, and now he's an investor and a partner at Lux Capital. Shaheen, welcome to the show.

Great to be here.

You are a car guy and an engineer and a partner at Lex Capital. Tell us a little bit about your background and Lex Capital and what you've been focused on.

Sure. I grew up in Silicon Valley with a dad who was a big science fiction fan. So I grew up watching Knight Rider and grown men talking to their cars and Star Trek with grown men traveling around the galaxy in their pajamas and became a huge fan of science and engineering and cars and computers in space and greed drew me to do computer science, electrical engineering during the dot com bubble. Unfortunately for me, I graduated in the dot com bust and came to Silicon Valley from Berkeley, which is where I went to undergrad to look for jobs. And I saw empty parking lots and tumbleweed blowing down University Avenue and decided to pursue my passion for cars in hopes of doing Detroit being, I guess, having greener grass or greener pastures than Silicon Valley. I went out to Detroit, didn't quite see it that way. Wasn't too crazy about working for a big company, kind of being Silicon Valley, Bay Area, kind of bred. So came back to California, did a PhD in electrical engineering. Like many grad students, started a company, raised some government money for it, and was advised that the best time to get to know VCs is when you don't need their money. So I got to know a bunch of VCs when we were fully kind of grant funded, and I got to the point where the company had to either raise equity dollars, where I had to kind of invest the next five years of my life to the company, or maybe try something else. And so after having delivered on the contracts with the government, I decided to explore other opportunities. And having gotten to know a bunch of VCs, Lux actually came forward with asking me to join the team. And so I thought I'd try this VC thing. And here I am 10 years later.

Terrific. Well, when did you first start to think that Autonomous vehicles were moving beyond science fiction and academia and into the realm of investable opportunities.

Sure, sure. So during my time at Lux, I had seen many trends, technology trends come and go. And what I noticed about autonomous cars was that not only was it being embraced and accepted or accepted and embraced in the kind of technical community, but it was being viewed as a certainty by sort of the business press and the general non technical community. And I'd rarely seen the non technical community embrace something to this level where it's viewed as a, again as an uncertainty versus kind of skeptical views and the expectations that it's going to take a very long time. You had non technical communities, regulatory communities, even building platforms around the expectation of driverless cars. And that obviously led me to believe that there is a large investable opportunity here. Great.

And Lex has invested in Zoox. Why don't you tell us a little bit about that thesis and how you came to know those guys and formed your investment?

Sure. So I got to know Kyle Voigt at Cruise shortly after he came out of YC and I had the benefit of getting to know him and his team and I credit him for first exposing me to sort of driverless. And we actually spoke about working together. However, that arrangement didn't work out and obviously the rest is history. However, after that, after that engagement, I was given the heads up by a journalist that there is an interesting company brewing out of Stanford that I should go meet that they haven't been able to establish contact with. So he gave me their email address and said knock yourself out. And I basically shared my thesis in Driverless in three sentences and got an email back immediately from the founder and was asked, was invited to tour the company and I flew my entire team because we're based in New York in Silicon Valley. I asked my New York team to fly out, visit the company and. And we made probably the largest investment we've ever made in this company. Because of the magical nature of the team and the bold vision that they were pursuing.

I think the adjectives most used about Zoox are mysterious, stealth and secretive. Can you tell us what Zoox is building and what you thought was so impressive about their product?

Zoox is building a what I expect to be an iconic institution that's going to revolutionize initially urban and eventually all of surface transport. They are taking an approach similar to what Henry Ford and the Dodge brothers did, which was reinventing the vehicle with the intent of it being self propelled. In Zoox's case, they're reinventing the vehicle with the intent of it being driverless from scratch. Today's conventional vehicles go in one direction because we have eyes in the front of our heads. Their suspensions are designed around a human being have to pilot the car. Having to pilot the car, it's designed around the driver, not around passengers, unless it's a limo or a minivan and the fundamental design philosophy changes. Add that to a business model where you are in the transportation business versus today's automotive paradigm. We can talk about this later. Which is based around building and moving plastic and metal off of dealerships. These are two fundamentally different business models which require fundamentally different organizations. And Zoox is being with the from the grounds up is being built from the ground up to build a product that is conducive to this future and build a business around it that is conducive to this future model.

So they're looking at mobility as a service rather than selling units off a lot. And the car itself, I've seen some photographs and pictures of a very futuristic looking vehicle, perhaps with no windshield and no steering wheel and something about the wheels having their own motor. Can you explain a little bit about that?

Sure. So without having to go into the details of the car, of the design of the vehicle, like the Dodge brothers, the vehicle has been designed with the intent of it being a driverless, B providing a magical and memorable experience for the passenger, and C to be operated as a fleet within a fleet model. If you look at today's conventional cars, the vast majority of vehicles that are sold are sold as consumer products. And sure, a small fraction of them are operated in fleets, but they're not designed specifically to be in fleets. Whereas in the case of Zoox, with the expectation of offering a robo taxi service, the vehicles are being designed from the grounds up again, to be driverless, to offer a very magical and memorable consumer experience, and to be operated in fleets. And there's the infrastructure that's being built around that whole operation, which is again very unique and very special.

So is there an advantage to building the whole car from scratch with autonomy? It sounds like Zoox is doing that. I think some of the electric car companies, Neo and Tesla, are doing that. But you don't see that many folks building their own cars at this point, at least that we can tell. What do you see as the advantage of doing that?

So it's a gargantuan task and I think I'm not in the heads of the traditional automakers. But my guess is that if you're a traditional automaker and your numbers Your business is measured by how many units you sell. It is very difficult to completely scratch your existing vehicle architecture with the expectation of building and operating probably 10, tens, if not hundreds of thousands of vehicles, whereas you're building and selling many millions of vehicles today. Which is why my belief is that it's companies like Zoox, like Nio, which are positioned to be the first and be the best at offering and operating transportation solutions, versus current conventional automotive companies that are built from the ground up to be consumer product companies and be rewarded and measured by the speed at which they take metal and plastic in and push vehicles out. The analogy could be the airline business. If you were to tell Boeing tomorrow to become an airline, they're not fundamentally built to offer customer service, to do marketing and branding, to deal with the logistics associated with operating flights, hiring pilots, getting food services, dealing with, dealing with weather, dealing with ticketing. I mean, that whole. It's a very different beast. And I believe the same applies to driverless cars and these driverless car services, where you can't be a traditional auto manufacturer and suddenly decide to flip the switch and become an operator of a driverless car service. You can be a conventional operator and perhaps build automotive technology that would feed into these operators. So the operators would partner with an auto company, for example, Uber would partner with a BMW, for example, to procure vehicles that BMW builds with the intent of being driverless. And BMW invests in the new vehicle architecture, and we very well may see that. But again, if you're a car manufacturer, if you're a BMW and you're selling hundreds and hundreds of thousands of dollars units, or millions of units of vehicles, it is very difficult to invest the huge amount of capital and time that's needed to come up with a new vehicle architecture.

Right. How should we think about the traditional car companies trying to partner, at least on the technology side, if not on the mobility service side? But. But I think we're starting to see certain tech companies say that they're going to build like a technology layer and to partner with traditional car manufacturers. How do you think that combination will work? Obviously, a very different approach than building it from the ground up. But how do you think the technology will fit in? It's hard to know what full stack means in this context, Michelle.

It's still the early innings. So you have the full stack building the vehicle from scratch with the technology and the operation on top, like Zoox, somewhat like Nio, and then you have the players who are building the bits and pieces, who will partner up together. So you have Uber, who has the distribution, the reach, the customers, and they will very likely partner with with a automotive company to provide the chassis and the systems that are needed for the driverless technology to be mounted upon. So my expectation is that the conventional automotive companies, which are more B2C today, may over time transition to a hybrid of B2C and B2B where they are providing vehicle chassis to the companies that are integrating these technologies. They can very well be the tier one automotive suppliers today, or they can be companies that specialize in taking chassis and putting autonomous technology on top of them. They may also be the developers of this autonomous technology and then they partner up with the Ubers and the Lyfts of the world who have access and distribution to customers. Again very early. But a fit finger in the air, yes, would be something like that. You have the fully vertical, the vertically integrated folks on one end and then you have the current players today working together towards offering a solution on the other end of the spectrum.

So it might be, it might take three different companies to put together that full solution. A car manufacturer and technology layers and then perhaps a mobility provider.

Correct.

Interesting. I read an article last week about Aurora, which is founded by Chris Urmson and some others. Said it wants to provide a full stack solution, but defined it as working with an automaker or supplier to design and develop the right mix of sensors, software and data services to deploy fully autonomous vehicles. And I sort of thought that wasn't quite full stack. It's not really making the car. And so it sounds like maybe we're going to see some in between players that are not building it from ground up.

But also, and there's an opportunity for these players to have a major role. I think that my belief is that the market and the opportunity is so large that we're going to have multiple winners valued in the hundreds and hundreds of billions. My expectation is that the likes of Aurora, Auto X Drive, AI, these software full stack software companies will have a major role in partnering with the automotive companies and these distribution companies in enabling these services.

It doesn't seem like the auto industry has ever been a winner take all business. So I think it makes sense that there would be multiple players who could be successful with multiple configurations. Is there anything different about autonomous or about rideshare that you think leans more toward only having one or two winners? Or do you think it's sort of will be an open field?

So as far as distribution is concerned, I wouldn't expect there to be too many entities that are interacting with the consumer. My expectation is that you're going to have the incumbents like Uber and Lyft to partner with the right groups and offer autonomous solutions, whether it's for consumer preference or whether it's for improving their bottom lines. And there's going to be these new entrants like Zoox or others who are full stack, who also want to have direct line of sight to the customers. But I wouldn't expect there to be too many. What I do expect there to be many of are people who are within that value chain, outside the vertically integrated players. So the people who supply software, the people who supply security, the people who provide teleoperations services, the people who offer design tools. It is my expectation that you're going to have many, many entities that are feeding into that supply chain. And the question you bring up which is, is it an interesting business? I would say that it's up to those entities. Today when you look at a vehicle that comes off the assembly line, there are multiple players that are in that supply chain that capture various portions of the margin associated with building that car. You have the product, people who like the OEMs, who do the marketing and do the supply chain and do the branding and the design. Then you have all the suppliers that feed into that supply chain. You have the distributors and the dealers. So today there are people who have a higher portion of that margin because of their access to customers or because of their negotiating position or bargaining power, balance sheets or whatnot. But that whole paradigm is going to change when we go on from a per vehicle model to a per mile transportation model. So in that per mile driven, the question will be to your earlier question, who's going to capture the most margin? Is it going to be the people like Uber, like Lyft, like Zoox, who have distribution? Or would it be the people like Aurora and drive AI who have the powerful software that's going to be running on these cars, similar to how Microsoft Windows won the majority of the margin at Microsoft Office obviously in the PC era. So that is still to be determined. And again, we're very much in the early days. But it's important to make that distinction between the per vehicle regime and the per mile regime that we're going into.

And it seems like there's also another hardware layer with the sensors and the LIDAR and these other components which maybe aren't part of a traditional auto supplier chain today, but need to become part of it. And it's not clear whether that will be provided by Aurora or you know, partnered with software, or whether that'll come in some separate form or some separate deal that the automakers make. You know, Wired magazine came out with a piece last month that basically said Detroit's winning, not the Silicon Valley car companies. You know, GM bought cruise and Detroit's the only one who's actually going to make the car. Do you see the young tech companies getting bought up by Detroit companies? Or how do you feel about the merger and acquisition piece of this?

So we're already seeing signs that the broader technology establishment is interested in automotive technology. We saw the NXP acquisition by Qualcomm. We saw the Movidius acquisition by intel, the mobileye acquisition by Intel. So you're seeing conventional technology companies become more cognizant of the value that can and will be captured in this autonomous era as we move from this conventional per vehicle model that we discussed to this per mile model that we expect in the future. I do not expect that the opportunities for MA for these technology companies that are in the space to be limited to the conventional OEMs and the tier one suppliers, but rather across the entire technology landscape. I would not be surprised if the likes of Cisco or Broadcom or, well, Broadcom has now been acquired, but other kind of large technology companies with very large balance sheets, with plenty of cash, paying high multiples or paying kind of again, offering large acquisition dollars to these companies.

What are your views on the question of full autonomy? Like Google and Zoox have said that they'll pursue going straight to a level four, level five vehicle versus a more gradual approach of assisted driving technology that gets better and better over time, perhaps the Tesla approach.

So it is a responsible and well thought out strategy, in my opinion, to separate what many call level three, which is driver assistance or periodic portions where the vehicle takes over. With level four, level five, which is no steering wheel, I feel like as if incorporating Level 3 or Driver Assistance or what I like to call pseudo autonomous is a very challenging problem, which is different from the problem of pursuing level four and level five, which is without a steering wheel. When you're doing pseudo autonomous, it is a feature. It is a feature, a technology that you install on conventional vehicles. And it's my expectation that there will be many, many millions of vehicles coming off of a assembly lines that have level three or pseudo autonomous functionality as a feature. On the other end you have level four, level five, which is fully autonomous, no steering wheel. And it's my expectation that that's going to be limited to the operators. You won't be Purchasing a level, unless you're a billionaire, you will not be purchasing a level four level five vehicle because you can just simply summon one with a ride sharing app. It makes absolutely no sense to own one of those vehicles again unless you're incredibly wealthy. And so the challenges associated with offering a level 4 level 5 service are very different from offering a level 3 solution. The level 3 solution has to fit within the conventional automotive supply chain. It has to fit within this current automotive marketing and channels and product definition. Whereas level four, level five is an entirely different ball game. Level four level five vehicles aren't necessarily cost constrained. You can blow $150,000 as an operator with redundancy and reliability on level four level five, whereas level three, you have to be very careful to be able to integrate it with the rest of the vehicle, has to be very easy to manufacture, easy to install and fit within the budget of a conventional vehicle. You can't have level three as a $30,000 option on a conventional car. It has to be a $3,000 option on a conventional car. So the challenges are very distinct and the approaches are very distinct, which is why these companies have responsibly decided to take very unique approaches for the two paths.

And how do you feel about consumer adoption? I think there's a lot of concern with Tesla and with some of these technologies that rely on the drive, the human driver, to take back control, that there's going to be a disenchantment, some accidents and things like that that will make people wary of autonomous technology. So how do you feel from a consumer adoption perspective? Would you like to see level three or would you prefer to see it go directly to more of a fleet managed Level 4 or Level 5?

I expect there to be both Level 3 and Level 4 and Level 5 products in the near future. I do expect there to be accidents. I expect there to be tragic accidents and tragic losses as a result of technology malfunctioning, as a result of human error. We see people abuse technology all the time. There is accidents as benign as people walking into each other checking social media while they're on their cell phones, to people using their phones in their cars and getting into very tragic accidents. So I think the abuse of technology and carnage as a result of the misuse of technology is probably as old as technology itself. And it's naive to not expect level 3 pseudo autonomous technology to be abused where people get out of the passenger seat and jump into the back seat, for example, you know, while the car is in level three mode. And tragic accidents do occur as a result of that. Level four, level five as well. You should expect these, these fully autonomous vehicles to get into accidents. And I guess the question really is for the public as to how much they choose to punish these technologies versus letting them mature, knowing that human drivers are killing tens of thousands of people per year. And they should be more forgiving. But again, it's very difficult to be forgiving when a loved one is involved in a tragic accident. So I completely understand and empathize.

Yeah, it's funny when I mention autonomous vehicles to my mother, she, she said, well, what about the trolley problem? I heard about the trolley problem on npr. You know, how are the cars going to decide who to kill? And I thought, gosh, if my mom is asking this question, you know, this is out there. How do you feel about some of the ethics issues and other issues around the fact that these cars are going to get in accidents?

I think, well, so there's two questions. There's the question around accidents, which we talked about, which are inevitable, in my opinion. Then there's the question around the trolley problem. And the issue that the trolley problem brings up is one where an algorithm is the deciding factor between human life or death. An algorithm decides whether the passenger should die or the pedestrian or the people on the outside of the vehicle should die. And that's a classical, I guess, you know, technology ethics problem that I unfortunately don't have the answer to.

Do you, how do you feel about letting AI and deep learning, you know, solve some of these problems around these algorithms? In other words, instead of rules based code with an engineer sort of pre programming ahead of time, allowing the AI, you know, through the training and watching millions of miles of human driving, coming to its own conclusion about how to handle some of these tricky situations.

I think so long, Michel, as a computer is making the decision, regardless of how the computer has been programmed, whether it's been subject to linear programming by a developer or having been design through an algorithm or processes like deep learning and AI, I think they're going to be viewed the same as a computer making a decision by the general public. And we have typically been hypocritical as it relates to technology. We are extremely critical of, of accidents that happen as a result of technology, but are more forgiving of accidents that are a result of human error. And my belief, my expectation is that as technology evolves and as it is proven, this may or may not happen, but hopefully one day, if it's proven that AI is indeed safer than people, that AI indeed, from a Statistical perspective can reduce injuries and death, then people will become more trusting of AI. Unfortunately, we're in the stone Age of AI today, and I think it's going to take a long time until it becomes conventional wisdom that AI is indeed safer than human judgment.

A lot of the policy discussion has been around how do we know when a car is safe enough to put it on the road? And how would regulators know? How would car companies, who might be asked to self certify that a car is safe? How will they know? How should we think about being good enough from a safety perspective?

I think there needs to be a new process by which autonomous cars are measured versus the process that's being used today. Today, when a product is tested, it is tested at a certain point in time. That doesn't quite apply to AI because as the product ages, it benefits from all of the learnings of all of the vehicles on the road. So the level of safety of a vehicle that comes off the assembly line is significantly poorer than the next vehicle that will come off the assembly line and significantly poorer than that very same vehicle after it's being driven for many hundreds of miles, because it's benefiting again from the learnings of the entire fleet of vehicles. And so with that in mind, I believe testing needs to be redesigned. And I think rather than testing to see whether or not the cars are safe on an absolute basis, they should be tested to see if the cars, if they are safer than humans on a relative basis. And I think it is mathematically possible to predict the scenarios where humans are prone to error and see if the vehicles are able to overcome those issues. And if it becomes established that the AI at any point in time is now able to overcome the issues that can tend, that can cause humans to make errors, and that the cars or the AI is safer, then I think the AI should be used. So the argument that the regulators can make is these vehicles are not bulletproof, they will cause accidents or will be involved in accidents. However, the rate at which they will be involved in accidents is less than the rate at which they will be involved in accidents had they been piloted by humans. I think that is a rational argument for a regulator whose responsibility is protecting people to make. Because my belief is that regulators will be doing their constituencies a disservice by not allowing for technology to proliferate that results in fewer incidents than human piloted cars, whom everybody knows are extremely dangerous.

What about the idea that I think maybe Zoox has indicated it's exploring, exploring about remote operation, like having a remote Operator who can help a car out when it gets into a tricky situation, sort of. You know, I think what you're talking about with safety are these corner cases, these edge cases, which are the problem, not the ordinary driving. What do you think about, you know, would it increase safety to have remote operators able to help out?

I think remote operators are a must. I think the nature of the operation will be a deciding factor for regulators to determine whether or not the vehicles are safe enough. I think the nature of the operation will be a key factor in competitive advantage relative to the other fleet operators. And I believe, again, it's going to be key.

It certainly would suggest that fleet operation would have that advantage over a car you just bought yourself in your driveway. If the fleet, you know, has this backup of a remote operator, it might help with consumer adoption.

My belief is that for the foreseeable Future, any Level 4, Level 5 service will require a level of remote operation. Now, the question is, how many people do you have for a given number of cars? I think that is going to drop over time, but in the near term, we may even have one or two people overseeing every vehicle for the first, let's say a couple years, and then it will probably become one operator per vehicle, and then maybe one operator per 10 vehicles, one operator per 100 vehicles. And. And I think, again, the success as it relates to regulatory compliance and as it relates to safety and customer experience will very much rely on the nature of these teleoperations. And I don't believe that you can successfully run a level 4, level 5 service without a teleoperations component.

I think the California regulations that are being proposed now are suggesting, at least for testing, that a remote operator would be required to keep an eye on vehicles that don't have a driver. Final question. What do you think about timing? Everyone loves to play this game of trying to figure out when this is all going to happen. Does being first to market matter, or do you think companies are better served waiting and getting it right? How do you feel about timing, and do you think it matters whether people are first?

I expect driverless transportation to be made available to the broad public in limited geographies in the next three years. My expectation is that it won't necessarily be the first to market who is the winner. And you've seen this many times in history. The first personal computer was not Apple. Apple also didn't offer the first smartphone, yet it was able to establish itself as a leader in both of those businesses. My expectation is that the player that offers the most unique, most memorable, safest experience will be the winner. And that player may or may not be the first to market.

Well, I appreciate you taking the time today. Thanks for talking with us.

Thank you, Michelle.

Thanks to Shaheen for joining the show. You can find the show notes for this episode and all episodes at our publication, smartercars on medium.com. thanks again for listening. We'll see you next time.