Transcript of Episode 94 – Shahin Farshchi on Self-Driving Tech

The following is a rough transcript which has not been revised by The Jim Rutt Show or by Shahin Farshchi. Please check with us before using any quotations from this transcript. Thank you.

Jim: Today’s guest is Shahin Farshchi of Lux Capital. Prior to Lux, Shahin co-founded Vista Integrated Systems which built wireless vital sign monitors based on neural interface technology that he developed during his PhD at UCLA. Shahin also developed hybrid electronic vehicles for GM in Detroit, worked as a software developer in several Silicon Valley startups, and researched new techniques for semiconductor manufacturing. He earned his bachelor’s in electrical engineering computer science at UC Berkeley. I love this quote from his Lux Capital about page. He’s passionate about artificial intelligence, robots space cars, and engines, pretty much anything you might find in an episode of Star Trek. All right. Well, welcome Shahin.

Shahin: Great to be here Jim.

Jim: Today, we’re going to mostly talk about self-driving cars, but the as the introduction shows, Shahin is a man of many interests and we may go off into who knows where as regular listeners of the show know that we frequently do. Let’s start talking about self-driving cars. Waymo has finally started full driverless taxi service in Phoenix. How big a deal is this?

Shahin: That’s a huge deal. It’s a huge step in the direction towards offering robotaxis broadly to the public. Of course as you know, these things take time. Autonomous cars have been a few years to come for the past maybe five or six years, and I think this is a huge milestone. Waymo is obviously one of the leaders in offering autonomous cars to the public, and this is obviously a huge milestone and I’m very excited.

Jim: Yeah, I had the same reaction. Everyone has been talking about it, but finally now somebody has done it, and that seems to me a fairly important step frankly for the evolution of our species because they’ve obviously been gigantic, multi-billion dollar investments on trying to develop autonomous vehicles, but as you and I both know as technology product guys, over time, anything you can do today will get cheaper, faster, better over time so that all kinds of market opportunities at smaller scale that may not warrant five billion dollars’ worth of investments will become more and more tractable things, like hotel room cleaners, kitchen main… I’d love to have a kitchen maintenance robot that would do the dishes, God damn it, right?

Jim: I suspect that this threshold will actually be important to get people to start thinking seriously, though of course timing is everything and when the cheaper, smaller, better versions of these technologies become applicable to these other domains. Before we go further, something that gets referenced a lot in the literature, I’d love to go over with you which are the five levels of automation for cars. Some people are a little sloppy about how they reference automotive automation, perhaps Tesla most famously so. Should we go through the five levels?

Shahin: Absolutely, let’s do it. I think there are two levels that matter. Let’s start off with all of them. There is level zero which is no automation, so that would be basically the model T forward all the way up to a four Taurus from the ’80s. There is level one automation which is very basic automation, such as autonomous or adaptive cruise control, lane change departure warnings where the vehicle can take over a portion of the driving task, but the remainder is still under the driver’s responsibility.

Shahin: You have to do everything else to keep the vehicle operating, and obviously you have to be aware of the road. Level two gets interesting which is what Tesla does and other upscale automotive manufacturers offer today, which is taking over all driver functions, not just a portion like adaptive cruise control or lane keeping, but all of the driver functions in a certain environment where the driver needs to be alert and ready to take over at any moment in time, and that is likely going to be the prevalent mode of automation that’s going to be available over the next few years. Then there is level three where the vehicle can take over entirely and the driver need not have control or have oversight over the vehicle for periods of time.

Shahin: I’m sure you’ve seen the fancy videos where the steering wheel folds away and the driver is able to take a nap. That is level three and we’ve seen some manufacturers claim level three functionality, but I think there is challenges that we can get into later that may postpone level three availability in the near term, if not indefinitely.

Jim: Question here on level three. The fancy Cadillacs have something that somewhere between level two and level three, which is that they make sure your eyes are still on the road, but your hands don’t have to be on the wheel.

Shahin: I would put that squarely into level two because in the case of level three, you don’t have to be paying attention to the vehicle, you don’t have to be ready to take over instantly. Whereas in the case of level two, the driver needs to be constantly aware and to your point, the system sees to it that your eyes are on the road because it’s a level two system.

Jim: All right. Level four, this is where the real work is being done today.

Shahin: Level four gets interesting because that’s where the vehicle takes over the entire operation and does not require any level of driver intervention between two points. You can get into the vehicle the same way you get into an Uber and expect to reach your destination without any kind of interference with the actual operation of the vehicle. The user experience going from level two to level four goes from being the driver all the time for level two, where the vehicle is helping you out to level three where you’re the driver part of the time, where level four you are the passenger, and the driver takes you from A to B.

Jim: Though of course, the distinction between four and five is that four has some ring fencing, right? It doesn’t claim to be able to do that thing everywhere and at all time. Am I correct with that that’s the main distinction, that there’s geography, weather or maybe even time of day, right?

Shahin: Exactly. Level four takes you to from A to B, but A and B have to be well-defined, and it has to be under conditions that are satisfactory for the operators of the vehicle. I view level five as somewhat of science fiction, where the vehicle can go from any point on the surface of the planet to another point of the surface of the planet under any condition at all times, and that is certainly a very interesting scientific feat. I expect us to get there at some point eventually, but I’m not sure if that’s a practical or even desirable role, but that’s how level five is defined.

Jim: Level five I think is what most of us have in mind when we think about an autonomous car which is, “Hey, I live out in the country on a remote farm and it’s an hour’s drive to the nearest grocery store over some fairly rigorous mountain roads, and I’d love to be able to get in my car and say go to Martin’s and have it do it.”

Shahin: Absolutely.

Jim: Frankly, the talk from the automotive industry say three years ago seemed to imply they were going to get there by 2020 or 2021, and yet that seems to be vanishing, either off into the distance or into the marijuana smoke of Elon Musk, one or the other, right?

Shahin: Level five makes for great headlines, and it makes for great science fiction. I’m not sure if you grew up watching Knight Rider like I did, but your KITT is a level five vehicle. It will take you from anywhere to anywhere. It will dodge bullets, it will go over bombs. It will get you to where you need to go no matter what safely. I think that’s a formidable goal. I think it’s a goal that to your point, a lot of people aspire to. It doesn’t strike me as a goal as one that is commercially viable and interesting enough to be worth the likely very, very large investment that’s required to get there in the near term.

Shahin: I think you can build a very interesting business by offering a level four product, where like Uber and Lyft today, you provide services between certain points subject to availability in areas that are of commercial interest. Today, you may not be able to get into an Uber between any random point at any random time to any other random point any random time. However, there are enough routes that are popular enough where there are drivers available, where Uber has become and Lyft have become pretty interesting businesses. I think with level four, you can achieve something like that and perhaps even better.

Jim: Yeah. For our audiences make it clearer, an example might be for instance of a level four Uber-like service that it operated in a 50-mile circle, which is not far from what the Waymo thing is. I think it may actually be quite a bit smaller than that, but some defined geographic area where it had high quality mapping, et cetera and it might even have a term of service that they could suspend the service during say a blizzard or something like that.

Shahin: Just like how by the way in the case of a human piloted vehicle, the human can say that weather conditions are not permitting me to take you to your destination, and that’s something that we accept every time we get into an Uber. It doesn’t have to be any different in the case of a level four autonomous car where you get into the vehicle and if there is certain road conditions, certain weather conditions, certain visibility conditions that preclude the vehicle from being able to get you to your destination the same way a driver would, then that trip could be aborted and an alternative destination would be required from the passenger, so that trip could be completed. That still in my opinion could be the basis of a very interesting business.

Jim: Yeah, at least in the corner cases, that could get pretty ugly. You’re halfway to your destination and there’s no squall comes through and the AI says, “Oh, sorry Shahin, we cannot proceed to your destination, nor can we take you back home. Where should we take you? You can go within two blocks of where we are today,” right? There are these corner cases that are somewhat disturbing with that kind of configuration.

Shahin: Two comments there. One, the same can happen with a human driver. There’s technical difficulties, you get flat tires, the car breaks down. The same can apply with a human driver. Then the second point is that all of these autonomous car companies will tell you that part of their technology development includes a mode of teleoperation and a backup service where if there’s any kind of failure in that vehicle, then the vehicle can be teleoperated by a human which would be as though they are physically in the car and/or a second vehicle in the case of a technical failure can come, pick up the passenger, and the passenger simply jumps from that vehicle to another vehicle.

Shahin: They either get to their destination or to somewhere safe. Teleoperation is a huge component of these autonomous car services that are coming around the corner.

Jim: Will that be feasible before we get to 5G in terms of latencies, et cetera?

Shahin: That’s a good question. 5G certainly is a very interesting technology. It is not absolutely required because the teleoperation systems that are being built today are taking into account the latency that exists in most urban networks and areas that they plan to operate in today. I think there are other challenges, but latency certainly is one of them, but latency is something that’s been accounted for and has been engineered into the solution. Going from single digit, double digit, millisecond latency that 5G offers to say 100 millisecond latency that would be the case in LTE is something that a lot of these developers have taken into account and have engineered their products.

Jim: Of course, though something I always point out to people that underlying TCP protocols allow for considerable variance in latencies or so-called jitter. While the mean might be 60 or 80 milliseconds, the specs allow going up to 2000 milliseconds drop off occasionally at 1% range. That kind of variance has to be built into any such system as well, because driving an automobile is not a condition where it works 99% of the time is going to work for you.

Shahin: Correct. Keep in mind Jim that the emergency scenarios that we have in mind are similar to the emergency scenarios that are taken into account in the case of aviation, or in the case of other modes of mass transport where sure, there is a possibility that all the systems shut down, but those possibilities are extremely unlikely. The expectation here is that there is a subset of the systems that will prevent the vehicle from operating autonomously. Therefore, the teleoperation will not be effectively sitting behind a steering wheel like a video game and driving the car remotely. It would be augmenting those subsystems that are no longer functioning and providing the vehicle with a path.

Shahin: It can find itself back or navigate itself to safety. It’s plotting points on a map. It’s identifying certain objects and obstacles. It’s being able to provide an update on the map that could have led to a discrepancy that led to the vehicle to no longer be able to operate. When I say teleoperation and what’s meant by teleoperation again isn’t that notion of a person sitting behind a steering wheel with a bunch of screens around them and literally driving the car. That doesn’t necessarily have to be the case. Usually, it is systems that are remotely augmenting the systems on the vehicle that have run into trouble and in those scenarios, you don’t need a super low latency system.

Shahin: You’re absolutely right, there is that just like a plane where both engines can fail at the same time, and there could be a puncture in the fuselage and decompressing. All of these things could happen at once where the plane falls out of the sky. Those types of situations are remote, and the emergency procedures usually allow these planes to land. I expect the same to be applied here in autonomous cars where the remote drivers help get the vehicle to save people.

Jim: Of course, all those levels of redundancy add cost. I was a civilian student pilot for quite a while. not too long ago and one of the things I learned is that the cost of anything aeronautical was about 5X the equivalent of automotive, because of the fact of the high reliability that was required and the redundancies. Presumably in this case, it won’t go to the 5X level, but they probably will have to throw some non-trivial cost into these redundant systems, so that they do have safe backups.

Shahin: You’re absolutely right, and that leads to my assumption that these level four vehicles, again a vehicle where the humans are strictly passengers will probably be limited to fleets just like 737s are, where they will likely have redundant systems, where they will have to be professionally maintained, and they’ll have to be operated as part of fleets and offered as a service as opposed to buying one and putting one in your garage, although people like Jeff Bezos and Mark Zuckerberg may choose to fork over a large sum of money to have the bragging rights that they own one of these machines, but even the machines that they purchase, just like private aircraft, are likely going to be operated by the professionals.

Jim: That’s interesting. You actually anticipate my next question, which is there are obviously lots of deployment models from every hairy homeowner owns his own, to transportation as a service, to a distributed economic community like Uber, where individual people buy these things and then make money by entering them into a network, et cetera. You already pretty much said it, but let’s focus specifically on that deployment model. In the near term, where do you see most of the action likely to be?

Shahin: Jim, let’s go back to the levels that we’re talking about earlier, and we can talk about this in that context. We all already know that level zero, level one, and level two are consumer products. You go to the store, you buy a Cadillac with super cruise, that’s a level two product. You put it in your garage, you own it, and you would treat it like any other consumer products. When you go to level four, now we’re talking about the redundant systems. We’re talking about a level of liability on behalf of the person who’s operating the service. We’re talking about having to operate these vehicles as fleets.

Shahin: We’re talking about having to employ professionals that oversee these vehicles, just like how you have people in control towers that are overseeing the operation of airplanes while they’re in use. I feel like there’s a huge distinction as it relates to the implementation of what we’ve seen, zero, one, two versus level four. That likely is going to be in the form of again level zero one and two continuing to be consumer products, and level four being a fleet operated model, where large companies with large staffs of professionals own these vehicles and they offer them to consumers as transportation as a service.

Jim: Again, I pointed out the possibility that capital ownership structure could be distributed, i.e. the actual owners of the vehicles could be individuals or smaller businesses who then put them into service on a network, very much the way people who own private jets can put them into a fleet operated, optimized network like sentient or something like that. Do you see that as a possibility?

Shahin: Absolutely and just like private aircraft, you need to comply as an operator of private aircraft with a bunch of rules and do a bunch of things to maintain the aircraft and operate it under certain standards. The same will apply to people who choose to own and operate their own fleets of level four vehicles.

Jim: Some of the other configurations that are already being deployed are less random access point to point, but rather more dedicated route type vehicles. In fact, I believe there’s already one in Nevada that’s in operation that runs from point A to point B. I don’t recall where that is, or what that model is. Where do you see that model fitting into the evolution of this ecosystem?

Shahin: I think most of the early level four deployments are going to be these fixed routes. I think Waymo obviously is taking a huge step forward with the service, and I believe it’s Phoenix, correct?

Jim: Correct, Mesa, Phoenix area.

Shahin: Yes, so it’s my expectation that most of these providers of autonomy services are going to be point to point as they harden the technology. Let’s say after a year or so in operation, they will offer the service as a random point to random point service. I think it’s just a process of refining the technology and not just the technology itself, but also the service and the operation that’s needed to run these vehicles, so the staff that maintain the vehicles, charging of the vehicles. It’s easy to start with one popular point to another popular point before you go to random.

Jim: Let’s go down this road a little bit. I have it later in my notes, but let’s do it now. One of the things that’s interesting as I step back and thought about this in my preparation for the call today, is that the nature of these beasts must be very different than the human brain. Of course, we know that about different categories of AI that the human brain is more general than AI has been so far. For example, a human is a general purpose driver. A human does not have great detailed knowledge of the route. I can imagine you’re 16-year-old. We might say the only route you’re allowed to drive is from home to school period, right? You got to be home before the sun goes down.

Jim: Yeah, maybe a brand new driver may operate like one of these early autonomous vehicles, and a mature adult human experience driver is general purpose doesn’t need the route information other than very generally, right? I think that’s an interesting clue that the technological road these folks have taken is extremely different from the way humans drive cars.

Shahin: That’s correct, and that’s because humans have benefited from many million years of evolution. We have two eyes on our heads, and we benefited from our ancestors having evolved from having to stand around and analyze their environments to be able to quickly generate intuition and run from danger, and be able to distinguish between fixed and moving objects without having to put too much thought into it. Unfortunately, AI has only existed for the past 10 years in its current form, and it relies on a very large amount of sensory inputs and a very large amount of computation and very large amounts of previous sensor data and previous training to be able to properly infer where it is and what action it should take from the sensor data that it’s receiving.

Shahin: In the near term, these machines are going to be equipped with many, many sensors with highly detailed maps, with a very large amount of compute power and a very large amount of historic training data before they can do what your teenager can do by spending a couple sessions behind the wheel with you in the passenger seat. By the time this technology matures, which I think will be in the not too distant future, we will have that, but until then, there’s going to be the need for a lot of sensors, detailed maps, very, very extensive training, training on many, many millions of scenarios that takes place either in simulation in silico, or in real life by driving these vehicles around town and a lot of compute on the back end.

Jim: I do a lot of my own work in the field of artificial general intelligence, or at least the attempt to get there. Clearly, this is not yet anything like general intelligence, even within the domain of driving. This is a hard code/trained kind of model, much like the deep learning systems that AlphaGo Zero, et cetera, which is interesting that they can get this far by what’s essentially brute force. Is anybody working on a more general solution in the sense that the way a human plays chess is very different than the way AlphaGo plays chess? If you listen to Elon Musk, it sounds like he’s talking about a general solution, less sensors, et cetera. Any thoughts on that distinction between brute force and generality and the approaches the players are taking?

Shahin: The general approach is certainly one that’s possible. In fact, there’s a handful of groups that have approached this from this angle, where you just put a camera on a car, drive it around for a few hours, and then eventually the vehicle can train itself and the vehicle can drive itself. We’ve seen that take place successfully. The challenge is that for autonomous cars given how much is at risk in terms of people’s lives at risk here, this has to work 99.99999, I forgot how many nines it is percent of the time for it to be as safe, if not safer than human drivers. That’s where it gets complicated because you need to start implementing rules in order to prevent those kinds of corner cases from taking place.

Shahin: If AI was as mature as the human brain, then of course you can train it for the same amount of time that you would train a human, but since we’re nowhere close to that, you have to spend a lot more time training and in addition to that, you have to help it by to use your term brute forcing some aspects, so that you can get to that level of safety and assurance that’s needed for these products to be viable.

Jim: The corner cases are what it’s all about I think as it’s turning out and probably were underestimated with some of the more optimistic forecasts three or four years ago. While I was doing research for this episode, I came across one example where someone says, “Hmm, the car has to figure out what to do when a flock of wild turkeys is stopped in the road, for instance,” right? It doesn’t happen very often, but it’s quite a mess. As a person who lives in a rural area where I see wild turkeys almost every day and other wildlife, I would add that ducks in the road is different than turkeys in the road, which is different than geese in the road.

Jim: Somehow humans are able to use our general intelligence, a small amount of cultural war about the difference between a turkey, a duck, a chicken, and a goose. Poor Mr. AI doesn’t have that 400 million years of evolution, and so he needs to have some rules, right? Turkey, you do one thing. Frankly, you just beat the horn, the turkeys will fly away. The geese might not. They’ll just stand there and look back at you and hiss, right? You ought to have a different algorithm for geese than you do for turkeys. That’s pretty far out on the tail, but not that far out on the tail. I think those are some of the things that have turned out to be the real challenges in these projects, because even level four needs to figure out what to do when a flock of wild turkeys is in the road right in front of you.

Shahin: Exactly, and if let’s say something falls off the back of a truck, should you just go ahead and run into it, or do you need to divert away from it? Those are things that humans are very good at doing that machines can’t quite figure out. Okay, was this a hard object that just fell off this truck, or was a soft object that just fell off this truck? This object that’s in the freeway, can I just go ahead and run over it, or do I need to swerve and avoid it?

Jim: Is it a plastic bag?

Shahin: Exactly, is it a plastic bag, or is it a piece of concrete?

Jim: Exactly, that’s the famous example. Is it a plastic bag or a concrete block? Tell you what, makes a hell of a difference if you run over one or the other.

Shahin: Correct, or is it a empty happy meal box, or is it a piece of concrete? That we can identify as humans very easily, but machines have a harder time doing that. Is this piece of expansion on the freeway, is that the actual direction the freeway is going, or is it just a small piece of expansion concrete that doesn’t lead anywhere? We’ve seen some cases where Teslas have been veered off the freeway because they’re fooled by lines and portions of concrete that are inconsistent on the freeway.

Shahin: Again, that’s why level two is prevalent right now and probably would be prevalent for the foreseeable future as consumer products, where you need highly, highly sophisticated machines to offer level four transportation as a service where the humans become simple passengers, and that also leads us into level three. You may ask, “Well, why aren’t we talking about level three?” I think what we’ve talked about so far leads us to the conclusion that level three probably isn’t a practical solution. The reason is that if you’re offering a vehicle that has a steering wheel and the steering wheel folds away for certain periods of time, then you start asking the question of, “Okay, well who’s liable for the operation of this vehicle during the time that the steering wheel is folded away?”

Shahin: If the answer is that it’s a third-party operator who is liable for the vehicle the same way the operator for level four vehicle is responsible, then you also have to have all the redundancy in all the systems in place to make the level four vehicles super expensive. The question becomes, “Well, why bother with a level three vehicle that has all the expense and the complexity associated with level where the driver takes responsibility for a portion of the time? Well, why not just make it a level four vehicle?” Now again, there’s always those unique cases where you have a billionaire or any enthusiast who wants to have a level four vehicle that they can take over every so often, but that’s not going to be a common mainstream product.

Jim: Yeah, I was going to say there may be some limited level three vehicles that people are talking about already, which is level three on the interstate for instance. I wouldn’t be surprised to see those.

Shahin: Possibly, but from the viewpoints of the manufacturer that’s building and offering these vehicles and from a viewpoint to the consumer, I’m not sure if these are super attractive, again because they’re likely going to cost a lot of money to build and a lot of money to operate. Again, there needs to be redundancy, there needs to be absolute safety for the scenario where the driver is no longer able or required to oversee the vehicle and at the same time, it is not fully autonomous like a level four vehicle. The question is, “Well gee, what am I getting with this part-time level two, part-time level four functionality?”

Shahin: Again, unless you’re a high-end auto manufacturer, just like how you can buy a hybrid Ferrari today, you’re spending half a million dollars on this vehicle and you get good gas mileage I guess, but that’s not the point. The point is that you’re getting the cutting edge technology in a super rare vehicle. I wouldn’t be surprised if these high-end brands also offer these super high priced level three products, but I don’t expect them to become mainstream.

Jim: I would push back a little bit, which is there is a distinction with respect to redundancy that at level three, you can fall back to wake the driver up, ring a bell, pull the car over and say, “Sorry, I can’t proceed, but you can” and then fall back to human control, which let’s say a fleet car with no steering wheel, no brake pedal. You don’t have that option, so you have to build a whole bunch of these recovery scenarios that we went into earlier. I would say that level three may be a little bit more viable than you think, so long as people think through how do you wake the human up and make them take control before you go forward, but you literally pull over and stop until that happens.

Jim: It’s certainly going to be tricky, but I do think that there could be a niche market in there and as you say probably the higher end, but it may not be Ferrari. It could be Audi, Mercedes, BMW guys of that ilk. Certainly, Cadillac is claiming they’re going to do it. I don’t know, we’ll see.

Shahin: That’s a good point. There is a future where there are varying levels of level three. There is the ford pinto level three where to the example that you gave, as soon as the vehicle has a problem, you get water splash in your face to tell you to get ready and take over the vehicle in a matter of a fraction of a second to avoid a catastrophic scenario, or you can have the Rolls-Royce level three experience, where you have 30 seconds or 20 seconds or 15 seconds.

Shahin: You get a nice massage on the seat that slowly wakes you up because the vehicle has the ability to maintain control and avoid catastrophe for 15 seconds, but every fraction of a second requires many, many thousands of dollars or perhaps even hundreds of thousands of dollars in redundant systems and perhaps even external system oversight, so that it has the luxury of being able to give the driver time to react and take over the vehicle. Again, like a lot of this whole notion of telling the driver, “Hey, driver you don’t have to pay attention. You can read a book, but you’re going to get slapped in the face immediately because of an impending death scenario coming up.”

Shahin: Just looking at consumer behavior, I don’t see how that’s an attractive consumer product versus just a level two product where you know that you need to be paying attention to the road, versus a level four product where, “Hey listen, we’ll get you from a to b and you pay for the distance that you travel.”

Jim: It’s interesting. From my perspective not to say that one person’s point of view makes a market, but I really have no interest in the high end level two, like the current Cadillac product where you have to sit behind the wheel, you have to sit up straight, and your eyes have to be tracked to make sure you’re looking out the window, but you don’t have to have your hands on the wheel. That doesn’t strike me as any value-add.

Jim: On the other hand, what you just described where I could be reading the book or answering my email behind the wheel, but not with my attention on the road, and it had let’s say a… let’s pick a number 8-second smart enough to handle the or look for eight seconds into the future, and psychologically I need to be prepared to take control on eight seconds notice, which is actually quite a long while, while I’m otherwise completely distracted. I would pay for that. At least for myself, I see a niche at the medium high end of level three and personally not interested at the high end of level two.

Shahin: I agree, I agree. I think eight seconds would be something that’s very appealing as a consumer product. I also think that eight seconds would take us into that territory of high redundancy and the need for external oversight that would make the vehicle likely extremely expensive. The question for you is, would you pay $300,000 for that eight seconds versus $90,000 for a model S that’s level two? My guess and I’m not good at predicting future consumer behavior, but I think I’m not sure if people would pay that extra amount and have to subscribe to a subscription, where you have that service overseeing the vehicle. In case you don’t take over, you don’t end up killing someone and have to pay all that money for that luxury. I’m not sure if I see that.

Jim: Yeah, always every advanced product has that, capabilities price space, right? Where are you in this n space of M capabilities plus price, and if we look at the history of these things, you basically start out with a very elite market as you point out. Maybe it is $300,000, but there is a market at 300,000, I guarantee it. I still remember the first big screen TV I bought. We just put a huge edition on the house, and we want to do something special. I spent $10,000 on a 73-inch big screen TV. This would have been in the year 2000, right? 10,000, Jesus Christ, right? Today, you could buy a very nice quality led screen TV for that for, I don’t know, $700 or something like that.

Jim: When they were $10,000, only maniacs bought them, very, very tiny. I bought it at a specialized big screen TV store that was fancy and bespoke, and all this shit they came out and set it up and all this stuff. Now you order it on Amazon, or you go over to Walmart and throw in the back of your truck. One could imagine the same thing happening for high-end system level three, where initially it’s $300,000 and only maniacs buy it. Let’s see, that’s 20 years later, probably happened faster because this is pure compute. Let’s say it happens in five years, the price gets down to a tenth of that, so the premium is 10% instead of 3X. Maybe your $90,000 S goes to $110,000 or something like that.

Shahin: Absolutely, and there was a time when you had to purchase in the 80s, you had to pay $70,000 which is the equivalent of probably $150,000 to $200,000 in today’s dollars 40 years later to purchase a car with airbags and anti-lock brakes. Now fast forward 30, 40 years, those are all standard. Every single car has that. Today, you may have to fork over a few hundred thousand bucks to get that kind of experience where you’re not slapped in the face, or get water pour on your face to wake up and take over the vehicle within a second, and people pay that kind of money.

Shahin: Just like how people spend tens of thousands of dollars to get oil changes on their Ferraris, they would spend to have that service to be able to have the luxury of a few seconds, five, six seconds to be able to put away their phones or put away their laptops or newspapers or put down the champagne, and be able to take over driving the vehicle. I feel like that’s certainly possible, and it takes some time beyond that again, just like how airbags and ABS took a few decades to become completely mainstream, until every vehicle that you buy has those varying levels of takeover time for the autonomous features.

Jim: I think that’s an important takeaway is this stuff’s not a continuum. We talk about the five levels or the six levels. You want to count level zero and things are not going to evolve in step functions. A lot of stuff’s going to evolve along a continuum through two to three to four, though four as we pointed out earlier is more of a discontinuous change. That’s the chasm shall we say. Below four, that will probably be continuous variation and market experimentation. Some things are just not going to work in the market and some things will, and some things won’t work today, but they’ll work in seven years and we should be interested in watching that as it happens.

Jim: Next, I’d like to go into, I should ask you this before we started, but do you have a reasonable knowledge of what the major players are up to and where they’re at, guys like Waymo and Cruise and what have you?

Shahin: All the major players, if you think of them as GM Cruise, Waymo, Argo, Zoox and by the way, we were investors in Zoox, we recently sold the company to Amazon and aurora, all of them are testing their fleets of vehicles and trying to train their vehicles with real life scenarios and varied types of traffic situations, and are expecting to offer their services whether it’s A to B or point-to-point in the next year or so. That is something that’s pretty much consistent amongst all these players. Besides that, the traditional OEMs, the Daimlers, BMWs, Toyotas of the world, all have their internal autonomous car efforts.

Shahin: We know that most of these companies have also partnered with these Silicon Valley startups as well to be able to accelerate their efforts. In addition to that, the tier ones and OEMs are working on increasing their offerings as it relates to level two driver assistance. We’re seeing two parallel efforts going on. We’re seeing the addition of level two driver assistance across the line, and we saw Tesla and Cadillac being some of the early players in that space. Daimler is catching up, BMW is catching up with them and then the same companies are partnering with the Silicon Valley, Aurora, Zoox, Waymo, Argos of the world to offer these level four, fully autonomous, point-to-point transportation as a service offerings.

Shahin: Both of these are happening in parallel. Now who’s going to get there first? What’s going to happen first? I think we’ll have to see, but Waymo obviously has been a leader. Waymo is obviously an established leader in this space, but that’s simply because they’ve been very public with a lot of what they’re doing. Other companies have not been as public on what they’re doing, and I think you should expect to see news coming out of those other companies, companies like Zoox and Argo and Cruise in short order.

Jim: Yeah, Argo’s interesting. They keep a pretty low profile, but they’ve brought in a shitload of money and some pretty high profile partnerships with people like Ford and Volkswagen.

Shahin: That’s correct.

Jim: Any insight into what their secret sauce is if any?

Shahin: That’s a good question. I think that the secret sauce is going to come down to unit economics. I think ultimately, all of these players will be able to offer an autonomous transportation solution. I think where the technology and the differentiation will really come in is their unit economics. Meaning, how much it costs them to move someone from A to B on a per mile basis. I think that is what’s going to make or break these technologies. People spend too much time emphasizing on who’s going to be first. I think the person being first isn’t necessarily going to succeed. I think it’s the person who can operate the best unit economics on a per mile basis.

Shahin: If you account for how much it costs to produce the vehicle, how many sensors it requires, how much computer it requires, what kind of battery storage it requires and in addition to that, how much human effort you need to keep these vehicles on the road? How sophisticated a teleoperation system that you need to make sure that these vehicles are safe? In addition to that, what kind of software goes into the system that makes the operation smooth and the vehicle fast while safe? All of that goes back and how many for example you have on the road at any given time. All of that goes back to what your cost is on a per mile basis, and whoever can offer the lowest cost on a per mile basis, while offering the best experience of the passengers in my opinion will be successful.

Shahin: I believe that we will see multiple companies offering multiple products after the public, but again I think that the devil is going to be in the details as to who will be successful in the long term, and that all comes down to cost per mile for moving passengers from A to B.

Jim: What I would call in a different context value engineering, right? Who gets the capabilities price space? Who puts their dart down in the right place on that capability versus cost? That makes particular sense if the buyers are highly economically motivated, such as a big fleet like say a Uber or someone who’s going to compete with Uber. At the end of the day, price is going to be very important. This morning when I was prepping for this call, I did a back of the envelope guesstimate of how much the driver costs in an Uber, and number I came up with not counting insurance was probably about two-thirds. If you throw an insurance, it might be 75%, assuming that self-driving cars are actually significantly safer than human-driven cars.

Jim: Yeah, there’s clearly a big incentive if you can get the driver out of the loop for transportation as a service, as long as it’s not more than three times as expensive. There will be a great marginal profit contribution, the less expensive you can make the capital and the lower operating costs for the back end network.

Shahin: It’s not obvious because right now, you put a human behind the vehicle, behind the wheel of car and it’s their car typically, and there is no need for a constant oversight. The only service that’s needed relates to billing and complaints and customer experiences after the fact. Whereas with an autonomous system, you need engineers highly, trained people overseeing the operation of these vehicles. These vehicles typically have to be owned and managed by the operators of these fleets, and they need to be cleaned. They need to be recharged and they need to be maintained. It’s not immediately obvious that the unit economics are going to be better.

Shahin: I actually expect the union economics to be worse in the near term relative to human drivers, but just like every other technology is going to mature and the unit economics are going to improve over time.

Jim: The question is, is there an entry point when your unit economics are too high? There is certainly for tests, right? A company the size of Google can certainly lose money for a number of years running small and even medium-sized tests, but if you go to a large-scale distribution, how do you do that before the economics are better than the human alternative?

Shahin: It’s my expectation that the autonomous car companies will offer a unique product to convince customers to perhaps pay more. It could be that, where you’re offered a special experience. The notion of arriving in a robotaxi, notion of being picked up in a robotaxi is something that consumers would be willing to pay for, the ability to blast music as loud as you’d like, cool down the cabin as cold as you’d like. Those could be attributes that would make people want to pay more for a robotaxi ride that perhaps even they would for a black car today on Uber, or you may have these super cash rich companies that can afford to offer the service at a loss in the near term to bring the price closer to conventional ride sharing.

Shahin: I think it’s going to be a mixture of both. I think it’s going to be a mixture of operating at a slight operational loss, while also offering a unique special experience where people would want to pay more. Because if you look historically, cars were probably initially not as cheap to own and operate as horses, and eventually cars became more attractive than horses. Going to the movies was probably more economical than owning a VHS or a Betamax player and buying movies for the home. It took time and new business models emerged that made those new technologies attractive and mainstream, and I expect the same for low taxis.

Jim: Yup, interesting the theory of market penetration expressed a little skepticism on how big the market is for people to take the driverless car just because it’s status simple. Maybe some, but most of the time, you take an Uber, nobody sees you come or go, so you don’t get much in the way of virtue signaling, right? Maybe a little bit and of course, the ability to lose money for a while if we think about it, even the current operators like Lyft and Uber are still burning cash at a pretty high rate. Of course, it may turn out to all be a fool’s errand.

Jim: If it turns out that there’s several operators you hit something close to the sweet spot with respect to the technology, and then which actually then makes it easier for multiple fleets to compete with each other, and there’s not that much economy at scale which is an interesting analysis in itself, it may turn out nobody makes any money at this, or at least not enough to be worth the very large valuations that are going into these companies. What do you think about that?

Shahin: That’s a very possible outcome. It’s very possible that these products are offered, and people aren’t willing to pay extra for them, that a ton of cash is lost in the process of trying to make these products mainstream. The whole industry suffers as a result and the technology is mothballed for perhaps a decade until those economics make it attractive. This wouldn’t be the first time we’ve seen this. There’s been other occasions where technology was just too early for its time, and the unit economics just didn’t make sense, and people continued using what they were using before. If you look at the dawn of the automobile, the automobile came out probably 50 years before it became mainstream, because it was coach built, it was handmade, it was unreliable.

Shahin: In some countries, it required a dedicated operator that most people couldn’t afford, and it took decades until it became mainstreamed. The same could happen here, and we’ll have to see what happens.

Jim: I guess that’s why you guys exited Zoox, right? Said, “Hmm, balance of return and risk, I think we’re a seller.”

Shahin: In the case of Zoox, we had a very interested buyer who obviously has very vast resources available to it and the question was do we allow this fantastic partner carry the vision forward, or do we keep it as a private company that has to keep on going out and raising money to get it to where it needs to go? For us, the priority was making sure that the company was in a good place, and the team was able to meet its ambitions and goals. It felt like as if moving forward with the acquisition was the way to get them there. We expected and continued to expect Zoox to be a fantastic organization and a super valuable company.

Shahin: It was a bittersweet decision when we decided to sell to Amazon, and we expect that the sale will put the company in a fantastic place, and again help the team achieve the vision that they set out to achieve.

Jim: Yeah, there’s always lots of different reasons why we sell or not sell our companies at various times. That’s a good description of that mix of motivations. A couple of the outlier players in the industry. I’ve alluded a couple times to Elon Musk and his different approach to all this. I mean if you listen to what he says naively. you’d think he’d be at level five by the end of the year, right? I can’t find anybody else that believes that, though he does have one unfair advantage which I don’t know what his actual name for it is, but I’d call it shadow testing. He claims that they have running in the background on many Tesla’s their self-driving software, that even when you the human are driving are acting in parallel in a shadow fashion as if they were in control.

Jim: By comparing this as if shadow AI with the behavior of the actual humans, there is an opportunity to perhaps learn a shitload faster. Maybe he’s got something there, I don’t know. What are your thoughts on that?

Shahin: I’m sure he’s got something there that is certainly an advantage, being able to run the software while all these vehicles are running, and training the AI with live humans in real time is certainly an advantage. Now the question becomes how much is the advantaged over folks that are using tools, for example, like simulations instead and where investors in Applied Intuition, which allows all these level four developers to test many, many scenarios at hundreds of times real time in silico, in a simulator to train their vehicles. I know that it’s an advantage. The question is how much of an advantage is it? In a day and age where these types of powerful simulation tools are available, I would question the extent at which how much Tesla’s advantaged.

Jim: I think we just don’t know, right? Because if I think about it from the cognitive science of education, it turns out having a teacher in the loop in something like real time is quite valuable. The higher fidelity of the teacher, the quicker you can learn. As I’m thinking about this, if I think about the cognitive neuroscience in particular of learning, having that real time linkage between the shadow software and the behavior of the driver could be spectacular. I haven’t really thought about it till just now. I’m going to defer saying Elon Musk is a pot smoking idiot, which he’s obviously not and maybe this advantage is big. We’ll see. Time will tell. We’ll see if he really can release something like level five or at least true level four by the end of the year as he’s claiming.

Shahin: I think level five is certainly a goal that’s worth talking about that would get investors excited. Going back to my earlier comments, I questioned whether or not that’s a desirable goal. If I had to choose, I would rather have a level four vehicle that can operate in most of Los Angeles under most weather conditions with unit economics that are better than conventional ride sharing. I think that would be a far more desirable outcome than being able to get a level five vehicle that can operate in random parts of the country and in super extreme weather conditions, where a whole lot of people wouldn’t be getting rides anyway.

Shahin: If I had to choose, I would choose the former, and I would encourage the business-minded companies that are developing this technology to focus on the former versus a science project, which is what I view the latter, level five.

Jim: Then let’s talk about the other one which God knows what they’re actually up to if anything. There’s long been speculation that Apple’s been working on this, and it would seem like a natural for them. It’ll initially be an elite product, probably it’ll be a brand opportunity at tremendous levels. It’ll be able to capture potentially a high-end premium price. All the things Apple likes, but at least officially, they’ve said they’ve stopped working on it, but people are not sure if that’s true. Do you have any insight into that?

Shahin: I have no idea Jim.

Jim: Fair enough. I’m in the same place. All I hear is the rumors and we know that they are sometimes operate at multiple levels of secrecy and attempts to camouflage what they’re doing. I don’t know, though I have to say I suggested many years ago they should have bought Tesla when Tesla was financially struggling, but they didn’t and it’ll be interesting to see if they missed this gigantic opportunity.

Shahin: A little story on Apple. I have a friend who was a investor in a company that was providing a component to Apple, and Apple gave them the general specification. There was all this buzz and rumors as to what the phone would be, and they thought that their product would be going into a phone, or into the phone where their product actually ended up going into the MacBook Air. Even there are suppliers who could be making components and systems for what could become a vehicle maybe under that impression, but it may end up becoming something completely different, so we’ll see.

Jim: Yeah, you can’t quite count them out. I have a good friend of mine who I can’t say who he is or anything else, but I can say he worked directly for Jobs for 10 years and developed some of the key software products for the Mac. He told me some stories that were mind-blowing in terms of the ways Apple goes to much greater lengths than anybody else in the technology space at least to keep people guessing what they’re up to. We shall see, we shall see. Let’s move on to another topic. You mentioned it in passing, and it happens to be a passion of mine which is a simulation. I call them simulation in the loop businesses, and you mentioned your Applied Intuition company. I actually had two startups that I was involved with.

Jim: I was chairman of one and a founding investor and director in another, where we used the spice analog circuit simulator in the loop with the advanced evolutionary AI to develop tools for chip design, in a computer chip design that is, and we successfully sold both companies. The first one to Synopsys and the second one recently to Siemens. I have been completely fascinated by the simulator in the loop business model. Maybe you could talk a little bit about your thoughts about that and your experiences with Applied Intuition. Did I get that name, right? Was it Applied Intuition?

Shahin: That’s correct, that’s correct. Simulators have extremely powerful tools across industries. You’re seeing simulators be used more often in the development of cars and planes and other more sophisticated, more complex products which has collapsed the price and the design cycles associated with these products. If you look at products today, the design cycles are a lot shorter because of the benefits that simulations can bring to bear. There’s a lot less testing that’s required today versus yesterday or several years ago, again because a lot of these scenarios can be done in simulation and get you the same level of fidelity for results.

Shahin: If you look at semiconductors back in the day, the amount of effort that went into making a chip that had let’s say 10,000 transistors was probably more than what goes into making a chip that has billions of transistors. A lot of that goes back to the availability of simulations and to the comment you made earlier on startup, a lot of the statistical analysis that accounts for many variations and process technologies that gives you a certain band of expected performance for the chips that are coming off of the line.

Shahin: In the case of Applied Intuition, obviously you need to train your neural network with many scenarios and the benefit of having a company like Applied Intuition as a partner for your simulation needs is that they are building a vast array of tools and a vast array of opportunities to test many scenarios and train your network on these many scenarios as opposed to just simply trying to reinvent the wheel in-house. If you think back in the days, companies like Digital Equipment and IBM had their own electronic design automation tools that were built in house.

Shahin: You look at the speed at which the industry has moved by virtue of the likes of Synopsys and Cadence and Mentor Graphics being available to these fabulous chip companies, and thus accelerating the adoption or the creation adoption of electronics broadly, I think the same will be applied to autonomous cars with Applied Intuition, which is the availability of these tools that can greatly collapse the time and the effort that goes into training an autonomous vehicle.

Jim: Is Applied Intuition at least for this time focusing only on the autonomous vehicle market, or are they thinking about a simulator and the loot businesses more generally?

Shahin: They’re focusing on autonomous vehicles broadly, robotics broadly.

Jim: That’s pretty big. You throw robotics in, you open the field up tremendously. I mean robotics is conceptually at least a much broader field.

Shahin: Exactly Jim. Exactly. They are in the business of providing training simulation tools for autonomous machines broadly. It could be a machine that’s on the road as a car or other vehicles, and we’re very excited about what they’re doing.

Jim: Very interesting. Now of course, people like Waymo do a lot of simulation with a hundred times as much simulation as they do road miles, at least that was the last time I checked. I don’t know what the ratio is currently. It may have shrunk a little bit. They seem to think you need both.

Shahin: You absolutely need both.

Jim: Our chip design tool stacks, we realized our biggest barriers to going to market was learning how to integrate with the tool chains at our customers. We were part of a 10-piece tool stack essentially to design a chip, and I suspect you run into something similar with Applied Intuition. Have you guys thought through how you integrate into their work process, where they have to meld a number of different approaches including simulation, product design, value engineering, and physical real world testing?

Shahin: That’s a great point. If you look back at silicon engineering, it took some time for there to become accepted and broadly embraced process from basic schematic layout to physical layout, to physical verification, to manufacturing, to testing. I expect the same to happen with autonomous cars. I think we’re still very early in the life cycle of this technology, and it’s our expectation that applied and tools like applied will become the de facto tools part of this process. Now if you look at semiconductors, I think a lot of the consolidation had a role there. I remember when I was a grad student using Cadence, I was baffled by how the Cadence interface was so different for every step in the process.

Shahin: Then later it occurred to me that the reason why it is like that is because they bought all these other companies and assembled them under their single environment. It’s my expectation that there’s going to be a general acceptance as to how you go about building an autonomous vehicle, or even a robot in general. There’s going to be tools that are going to be the accepted tools and processes put in place where these companies have a major role, and that will obviously accelerate the introduction of these products to the market.

Shahin: Just like how today a product like a smart speaker, which may have taken years to develop and manufacture, now can be developed and manufactured in months by a young entrepreneur, whereas you needed to be a high ranking manager at a large company in the past to be able to pull something like this off.

Jim: Yeah, that’s interesting. This may actually make the economic problem for the players in the autonomous auto industry more difficult if it essentially lowers the activation energy for somebody to enter the marketplace.

Shahin: Absolutely. If you think about Philips and the compact disc, that’s the story that I love going back to, they invested a large amount of money on the materials engineering and the process development to be able to offer and manufacture compact discs. What they did successfully was reduce the amount of time and money that it takes to build a factory to build compact discs. Well, guess what happened? A lot of other folks also started building compact discs using slight modifications, or variations of the Philips technology. They ultimately didn’t make money on a innovation that they invested in, and that was successful and became mainstream because they succeeded to the point where they enabled everybody else to do the same.

Shahin: The same could happen for autonomous cars where companies like Applied Intuition reduce the NRE. They reduce the time and the money that it takes to offer a product, and they reduce the cost of operating the product on a daily basis for their fleets, so that there could be many small, local companies offering level four robotaxi services.

Jim: I had not thought about that, until I again was doing the research for this call and stumbled across your company. Whenever I try to get involved in an industry, I always try to scope out those kinds of dynamics. Are there force fields that will make this a loser for everybody? The famous example was in 1982 when 104 companies were launched to build the brand new Winchester hard drive technology. Predictably 98 of them failed, right? It was too easy to enter as it turned out. There weren’t sufficient barriers to entry. What are some other interesting companies in the autonomous and/or robotics field that are doing what Applied Intuition is doing, which is providing a middleware of technology that enables people to enter the market more rapidly?

Shahin: Another big problem Jim in autonomous cars is the sensor calibration problem. You need a sensor for detecting position, detecting velocity. You obviously need cameras, you need GPS, you need inertial sensors. All these sensors need to talk to each other, and they need to be calibrated so that the information that they receive is a cohesive whole for a computer to process on the back end. One of our companies, Ava is doing is collapsing many of those sensors into a single device that can perceive the environment the same way we do. The way it does this is that it provides a very high resolution map of its environment, the same way Lidar does, but with greater resolution and greater sensitivity.

Shahin: In addition to that, it provides velocity information for every point that it picks up. Think about the example that I gave you earlier, how our ancestors were able to run away from the mountain lions that were coming out of the bushes without having to look and think and observe and figure out what was going on. They had the instincts to immediately run away. This technology offers the same thing for a vehicle, where the sensor itself can detect is this a toilet, is this a blanket, is this something that I should be veering away from, is this a child that’s standing by the crosswalk, or is that a fire hydrant?

Shahin: Being able to make those very important distinctions at the sensor level without putting all of the onus on the compute on the back end, and also creating the simplicity of a single sensor being able to detect all these things is extremely attractive, and I would view it as an enabler. You are effectively buying a perception solution when you buy an Ava sensor, and that will greatly simplify the challenge of designing an autonomous car, and it would be an enabler for many of these companies trying to enter this arena.

Jim: Ah, that’s a good one because that is a hard problem. It’s conceptually a hard problem. In fact, in cognitive neuroscience, it’s still not exactly known how we integrate our multiple modalities of sensation, sound, sight, and touch of the three that have specific regions in the brain, each of which have neural maps for their inputs, but how they’re integrated into the sensorium of our consciousness is still not known. It’s an area of active research of course, and the analogous problem clearly applies to the integration of the various sensors on a autonomous car. Having that solution turnkey clearly reduces the complexity of the problem and hence the activation energy to enter the marketplace, so that’s exactly what I was looking for. Excellent example.

Shahin: Think about how Qualcomm made the communications components of any device a solution. When you buy a Qualcomm solution, you get CMA, you get a reference design, and you obviously buy their chips and put them in their reference design, and you get a communication solution. We view Ava as similar to Qualcomm in that it’s offering a perception solution, in that when you are buying this sensor, you’re not just getting a sensor that can detect certain things. You are able to put now perception into your vehicle that’s enabled at the sensor level, and we view this company as being what Intel and AMD were to making devices compute and making them intelligent to some level and how Qualcomm connected them.

Shahin: We expect Ava to make them perceptive, and this applies to autonomous cars and robots as a whole.

Jim: Very cool. I’m going to have to look into that company. That makes a lot of sense. Can you think of other niches of that sort that are out there in the autonomy field, where there’s an opportunity for entrepreneurs to essentially provide a middleware solution that makes sense to the whole industry and particularly lowers the activation energy for people to work in autonomy?

Shahin: I think there’s a lot of emphasis right now on the vehicles and the machines themselves and going back to the earlier conversation, there are a lot of other aspects associated with offering a level four service that will have a major impact on the bottom line for the operator. It is the maintenance of the vehicles, the charging of the vehicles, assuming that they’re going to be electric. It’s going to be the teleoperations, it’s going to be customer service. I think there’s a lot of secondary aspects that folks don’t talk as much about because they’re not as sexy and I think entrepreneurs, you can think about which to your point reduces the activation energy and will ultimately benefit everybody.

Jim: Those strike me as interesting opportunities. I know my own entrepreneurial career, particularly back in the early days when I was actually starting and building companies myself, I actually liked businesses that had at least a little bit of dirty work in them, right? Something that was actually annoying and difficult, because a lot of big companies can’t get their shit together to solve actually annoying problems, or don’t want to, or have internal politics that keep them from even attempting it, and dealing with a certain amount of is actually an interesting barrier to entry, particularly against really large players.

Shahin: Speaking of dealing with shit, there is the policy aspect and the liability aspect. If you look at airlines today, if a plane crashes, the airline doesn’t get sued out of existence. There are certain requirements of them that the government imposes, and they can purchase insurance to pay for those events because they are predictable. So long as there isn’t some kind of negligence as a result of the investigation, then the airlines can survive these catastrophic events. It’s not clear as to what the analogy is for the inevitable scenarios where these vehicles strike humans, destroy property, and can cause huge amounts of damage.

Shahin: It’s not clear as to what the limit of liability is for the operators, but people that can figure this out and help the operators comply with the various states and cities that they’re going to be operating in also could be an interesting business and something that can also benefit the industry.

Jim: My only one idea I had about autonomous vehicles, this was four years ago when there was all that hype, I actually did a little exploration of the possibility of putting a property and casualty company together, recruited in one of the world’s best actuaries, et cetera because we realized there were some very interesting attributes that probably driverless car. By the way, this was part of our naivete, but also part of the reason we decided not to do the business. We were focused on the individual owner of essentially level four or level five cars, and we took the business assumption that they actually would be let’s say 60% to 80% lower casualty claims including liability.

Jim: Hence, if one were an aggressive first mover, one could grab market share. However, after about a 3-month study, we decided not to move forward with the business on the grounds that topologically, it struck us that the operators and/or car companies themselves were in a preferential position to offer these services, because they had better information than anybody else, particularly early where the underwriting risk was the highest.

Shahin: It’s my expectation that the ride-sharing companies are also going to be major players here because the ride-sharing companies A, have access directly consumers and B, they’ve been operating in these geographies for many years, and they’re able to determine going back to the conversation on unit economics where to most profitably deploy these vehicles. I would not underestimate them. The automotive companies and the tier one manufacturers certainly have a very important role here, but I think as it relates to actually where the rubber meets the road, no pun intended, I think the ride-sharing companies will have a major role to play and could be advantaged as it relates to deploying these cars.

Jim: As far as I can tell, Uber is like the Netflix play, right? I suspect they have figured out by now that with humans in the loop, the damn thing will never make any money. Just like Netflix, most people don’t remember this, but Netflix used to be in the business of schlepping CD-ROMs, DVDS back and forth by the US mail. Holy shit, an insane business. I don’t think it was ever profitable, but the smart people at Netflix realized that, “Look at the inevitable decline in the cost and capability of bandwidth. Someday not too long in the future, we’ll do this online and our marginal cost will drop by 95%. We’ll have one of the great business franchises of all time,” and they pulled off that play brilliantly.

Jim: No people thought they were idiots for quite a long while. Maybe that’s what the Ubers and the Lyfts are up to.

Shahin: Correct. If you look at Netflix, they didn’t necessarily have to invent the internet, or invent broadband to offer their service. What they achieved with the mailing service is establish a customer base, and that customer base ultimately made it very valuable and easy for them to switch to the online service. Right now, if a Ford or a Daimler or Hyundai decides to offer a direct-to-consumer robotaxi service, it probably would be disadvantaged relative to an Uber or Lyft whose apps are already installed on most people’s phones. They already have a very, very good sense as to how profitable or unprofitable certain geographies are, and which geographies would benefit from using robotaxis versus conventional human piloted cars.

Jim: That’s one of the companies we didn’t talk about was Uber They of course were real early to deploy, at least nominally level four cars, but we haven’t heard too much about them recently.

Shahin: It could be a strategic move on their part to just watch and see the technology develop. Into the earlier conversation if companies like Ava and Applied are successful, then it should reduce the barrier for these technologies become mainstream, these products become mainstream. It could be a move on their part, a strategy on their part to watch and adopt or acquire the right technology that they would need to offer the service when the unit economics are attractive. They would become potential buyers of team or solutions that make sense for them.

Jim: Of course, the other side of if you’re correct about them controlling the access to the customer, they may also be able to operate as monopsonist buyers, right? Where if you have just a small number of buyers, the buyer actually has the leverage over the supplier, and that could be their strategy too. I just looked it up and their market cap is $58 billion, not bad for a company pissing away money by the bucket full. At least the markets believe that one or both of those strategies has a bright future for them.

Shahin: Possible.

Jim: Yeah, we’ll see. I wouldn’t buy it at that price. Fuck no, but clearly people are. I’ve been wrong too by the way as the one that poo pooed Bitcoin and when it was at $500. Oh well and a good friend of mine went in heavy and great for him. Anyway, we’re about out of time here. I really want to thank you a bunch. This has been everything I was hoping would be as a interesting and deep investigation into the field of autonomous vehicles and some of the ecosystems and technologies around it. It’s been great to have you here on the Jim Rutt Show.

Shahin: It’s been a pleasure Jim, loved it. Thank you very much for having me.

Jim: It was great.

Production services and audio editing by Jared Janes Consulting. Music by Tom Muller at modernspacemusic.com.