Lee Hower
Editor’s note: The NextView team periodically holds internal “shootarounds,” where we discuss a startup topic or trend and try to make sense of it for everyone involved: entrepreneurs, investors, and consumers. We’ve decided to share these publicly. Below is a lightly edited transcript of our discussion on drones. You can also read past shootarounds on voice interfaces and drones.
LEE HOWER: I think everybody here is enthusiastic about autonomous vehicles. There’s a whole bunch of things that it means for the automotive and mobility industry. We can talk about that and what could or could not happen in the next years or next few decades. Then there’s a obviously broader, societal impacts of if-and-when full autonomous vehicles are here. What does that mean for drive-in, quick-serve restaurants, for instance? What does it mean for city planning? What does it mean for where people are going to live What does it mean for the insurance industry? What does it mean for lots of things?
We’ve obviously made an investment now in Optimus Ride too, so I know we’re all generally bullish here.
DAVID BEISEL: So maybe it’s helpful for this meeting to start by discussing why fully autonomous vehicles are further out than some people think.
ROB GO: Part of the problem is people-based. Unless you’ve started to look really hard at this space, it’s tough to discern what’s what. The National Highway Transportation Safety Administration put out these guidelines of different levels of what is truly “autonomous.” When people say driverless cars or autonomous vehicles, they think, “I get in my car at my home or apartment, I close my eyes, and I wake up at my office as usual.” But there are wildly different points of the view as to where we are now and how quickly we arrive at that future. There are also some companies in the marketplace who have an incentive to promote what they’re doing as autonomous to generate better positioning and buzz and PR value, which creates more confusion.
LEE: It’s silly to express it, but one thing to note is that companies like Tesla or Ford have an incentive to say, “We’re going to have fully driver-less cars — a full fleet — by 2020.” They have incentive to say that to make people think that Ford is ahead of the curve relative to Chevrolet or Volkswagen or Toyota, or whatever. But personally, my belief is that we’re much further away from fully autonomous cars than we think.
One, the reality of driving and roads and all the things we have to deal with is infinitely more complicated than people appreciate. If you just stopped and thought about the last 100 trips you took in a car, I promise you there’s a whole bunch of things that happen with a reasonably high frequency of situations that are hard to automate or build rules around. It could be something as simple as someone who leaves their garbage can a little too far out in your neighborhood, and it blocks the road. It can be as simple as when you’re driving behind a bus that stops to let kids off. You have to stop in that situation, but what it was a garbage truck? An autonomous car doesn’t actually know to go around it. It’s not smart enough to realize it’s okay to actually go around one vehicle versus basically being parked there for the next however many minutes.
TIM DEVANE: What have you experienced driving a Tesla, Lee? Isn’t there a self-driving component to it?
LEE: It’s kind of scary to be honest. Or put another way, it works great 95% of the time, but the 5% of the time that it doesn’t work, it’s fucking scary. Let’s say you’re driving on the highway and you’re in the right-hand lane. When you come to an exit ramp and if you have the self-driving setting on, it sees that the stripe goes off to the right like that (motions hand away to the right) and it just veers you off to the road. It doesn’t occur to the car that, “Oh, I actually would just keep going straight on this highway and not go right because that’s just an exit.”
Or, like, I drive on 93 to the office and back home. As you know, they put the Jersey barriers up for the HOV lane. I’ve had times where I’m pretty damn close to the Jersey barrier, but you just drive carefully as a person. But the self-driving setting has veered me quickly away from the barriers because it freaked out that it was getting too close. But sometimes you just have to drive close to those things because of all the various components that make up that one stretch of road.
ROB: I feel like one of the reasons Tesla has an advantage though is that they’ve been building a data set around semi-autonomous driving much faster than anyone else. They think like a software company and not a metal-bending company.
LEE: Correct. But there’s a second thing that could delay truly autonomous vehicles becoming mainstream quickly, and that’s cost. Right now, with those Google self-driving cars, the cost to build one of those is about $400,000, once you put in all the stuff. And yeah, there are however many people in the world who have $10 million in the bank who might pay $400,000 for a car, and of course the cost will go down with the increase in volume of manufacturing. But if it costs hundreds of thousands of dollars to make a truly autonomous vehicle, you might see it in commercial settings first and not consumer. There could be a commercial trucking use case before there’s a steady influx of passenger vehicles onto the road.
DAVID: So let’s talk about why we’re bullish as an investment approach. Is it a huge market in shipping relative to passenger?
LEE: The reason why I’m bullish is that–we talk about the fact that we take risks as seed-stage investors, but if you win, you win big. This is just about one of the biggest markets in the spheres of commercial activity. It’s hard to wrap your mind around actually, because we talk about things like software markets where something plays in a $10-billion market, or it’s plus-size clothing where it’s a $20-billion market. But the transportation market is just absurdly big.It’s over half-a-trillion dollars.
Last year, a little over half-a-billion dollars was spent on brand new passenger cars in the United States. I’m not talking about any other countries. We’re not talking about trucks or commercial vehicles. We’re not talking about used cars. We’re not talking about taxis or anything. We’re just talking about brand new cars sold to the consumer. Half a billion.
DAVID: There’s a second point, which I think is worth reflecting on too, which is: To what extent is the pathway from where we are today to fully autonomous vehicles a pathway of vertical integration versus not? If you look at historically, when people built the first cars over 100 years ago, there were no companies that made engines and components and whatever.
LEE: Right, Ford’s first factory — the main, huge one, where they still make cars — this was called River Rouge. It’s like the size of a city. Literally, ships would come in there with iron ore. They would make the steel, then turn the steel into pieces of a car, then assemble the car, and at the other end of the whole complex, finished cars would drive out onto boats and trucks and get shipped around the country. That was out of necessity, since the auto industry didn’t really exist yet.
TIM: Google is taking a page from that, no? Their thinking seems to be that the only way to get a fully autonomous vehicle to market is purely vertical like that Ford example.
LEE: Exactly. “We’re gonna build it all ourselves and not actually do any interim steps between today’s car, semi-autonomous, and fully autonomous.”
Here’s what’s more likely though. If you look at the rest of the automotive industry, there are companies that put cars together, but they buy lots and lots of pieces of technology and components from lots of other companies. I am of the belief that autonomous vehicles are likely to be something like that — if not at the beginning, then pretty quickly after.
ROB: Right, different companies are going to solve different parts of the challenge. And that’s a part of what makes the investment opportunity and the opportunities for entrepreneurs to build meaningful companies so intriguing.
DAVID: You started talking about the perception of computers versus humans earlier, with the Jersey barriers, Lee. Where do you see that heading?
LEE: Right, we talked about the potentially useful data gathered by, say, Tesla with semi-autonomous vehicles today, to inform fully autonomous vehicles tomorrow. But that only helps if the computer is perceiving the same thing as the human. If the computer is perceiving something different from the human, it’s actually non-useful data. In other words, you can put cameras on top of a car that can see 50 times further than a human being can see, or you use lasers that can cut through fog.
Conversely, I can look ahead and parse what I see in different ways than the computer. The computer can sense a truck in the right lane next to me if it’s 100 feet ahead, but it actually can’t perceive very well when it’s just 20 feet in front of me. One of the freaky things that happens is that Teslas are programmed to try and be in the middle of the road, because that’s what you should do. But if there’s a giant semi that’s actually kind of overlapping your lane a little bit, or it’s windy and swerving, the safest thing for you to do is actaully be way the hell over to the left side of your lane, not be in the middle.
The point being, telling a computer what to do based on what humans perceive is not necessarily relevant.
DAVID: I’m looking forward to having my fully autonomous RV so I can have my mobile office and live anywhere.
ROB: Trying to get to Burning Man?
DAVID: Ha, well, yeah, but mostly, I meant to help me live and commute much further from the city center.
TIM: The opportunity cost of commuting almost goes away. You could see people doing that from Boston — live down on Cape Code and just show up to the office after a productive morning in your vehicle of doing work, nothing missed by commuting that far. Maybe you’d get to shower and get ready and get dropped off it you own a self-driving RV too.
LEE: That may well happen. There’s a couple different versions of how you get there. Autonomous cars can communicate with each other, such that they coordinate their efforts. The concept is called convoy. The second is you have smart roads. Even if a car can’t communicate with another car, a road can basically communicate to, basically the road can say, “The average speed here is 10 miles an hour, the average speed here is 15 miles an hour, the average speed here is 40 miles an hour.” And so on.
ROB: Because you have full autonomy by then, people might be willing to be in cars more often too. Seniors, kids. People who generally don’t want to travel far. I mean, just think, would we put our kids into one of these vehicles alone and send them off somewhere? Regardless of that specific use case though, if being in a car was more pleasurable and not dead time, you can actually see people being in cars more.
DAVID: I mean at some point there’s a geometric volume data problem.
LEE: Yeah, the amount of data that can theoretically be generated by an autonomous vehicle in 1 minute is 500 GB — it’s absurd.
DAVID: So you’re not actually going to share all of that data with every car, but you might share a detailed set of data that are within a 100 yards of your car and a different set with the overall system.
LEE: If you believe, as I do, that being 100% autonomous all the time, everywhere, wherever, is really hard, but getting to 98% or 99% is vastly easier, you could imagine the first iterations of “fleets” where cars are all autonomous, but a very small percentage of fraction of a percentage of time, they get into a position where the computer has a hard time figuring out what to do and so a human interjects remotely.
A split-second save may not apply here, but for instance, a remote human “driver” could assess a situation where a driverless vehicle is stuck behind another car. The human could determine if it’s a school bus stopping every few blocks that you can’t drive around or if it’s a garbage or delivery truck that is fine to pass. And then the person can get the vehicle back into a place where the computer can take over again. Totally doable.
One last point about these ideas of “fleets” would be the auto insurance market. Some people believe upwards of 40% of it could evaporate with autonomous vehicles, in which case, unless they could figure out something else to do, some pretty big insurance companies would go out of business. There’s a question of, if it’s all fleets, you would have a different kind of insurance model. But part of this is what happens to laws. If the owner is still responsible for what the car does, even if it’s part of a single group of cars moving and communicating together, it sort of doesn’t matter. But if the company who made the car is responsible, that would be a change.
Ultimately, stepping back again to the topic in general, we feel it’s going to happen, but for any number of reasons, it will take awhile to be fully autonomous and fully ubiquitous. But right now is a big opportunity for both investors and founders.