Future Man

Driving With | John Krafcik

The head of Google's self-driving-car program on Caterhams, moonshots, and why (rejoice!) you'll still be driving long after the robots take over.

By Sam Smith
Photos by Robert Kerian

One rainy day last spring, I pulled into the parking lot of a quiet office building in Mountain View, California. The address was unlabeled on Google Maps; my iPhone thought I was at a shopping mall. I was in a borrowed, 260-hp Caterham 7 CSR 260 with the top down, and rainwater went up my nose, because that's what happens in a Caterham when you refuse to put the top up. The car's charms make you immune to suffering. Or maybe you just get so much suffering that you start to enjoy it. Sevens tend to plant your feet firmly in the present.

Which is partly why I drove one to see John Krafcik. Krafcik is CEO of Google's Self-Driving Car Project, arguably the most futuristic thing going in the car business. The other reason is because Krafcik famously owns a Caterham himself, plus a 2006 Porsche 911. In a happy paradox, the guy taking humanity's collective keys owns two of the greatest sports cars we've managed to build.

The self-driving program is part of X (previously Google X), the tech giant's semisecret "moonshot factory" for projects of social good. Krafcik started there last fall, after executive stints at TrueCar and Hyundai Motor America. The son of a tool-and-die maker and a 30-year industry veteran—"I never wanted to do anything else"—his career began in the 1980s at the joint Toyota-GM NUMMI plant in California. He later studied at MIT, where he helped assemble The Machine That Changed the World , a five-year study of the Toyota lean production method.

We spoke at X but also in the Caterham, the latter in the hills above Palo Alto. Emails with Krafcik prior to our meeting included discussion of a first-gen Toyota MR2 on San Francisco Craigslist and the subjective merits of Volvo wagons. Those topics came up organically, because that kind of weirdness is basically unavoidable when you work at this magazine. (Or are trying to bring on the robot-car future, apparently.)

Sam Smith: One of my co-workers pointed out that a Caterham owner working on Google's koala car is like a master chef becoming head of Soylent. Why did you want this job?

John Krafcik: [Laughs.] Many, many reasons. I've always fought for great safety technology. What was great about Google was that opportunity to have a big impact. You know the data? 1.25 million people die on roadways every year. It's pretty amazing that we, as a global society, aren't more pissed off about that. We should be really angry. And we would be really angry if we could see one 737 crashing every hour.

SS: More than 33,000 people a year in America. Cars are generally safer than they once were, but that number's still huge. How did we get used to it?

JK: That is the question. How have we come to accept that as an appropriate tax on transportation? I think part of it is because only four to five percent of the time those fatalities are the fault of the machine. The rest, it's human error. It's really tough. Technology can help us solve that. I felt excited about the opportunity to help. Recognizing that it was in no way going to impede us as car enthusiasts, our ability to drive the cars we want to drive, on the roads we want to drive on.

SS: You have to assume Google works differently from a carmaker on something like this.

JK: It's one of the distinguishing characteristics between how traditional OEMs think about autonomy and how we have the liberty of thinking: Do I have to go in this stepwise fashion? Moving from L1 to L2 to L3 or L4 … [carmakers] have retail networks they need to take care of. They have customers they need to take along on this journey. The Google approach ends up being very freeing. One of the things that's different is, we're solving for pure L4 autonomy. The no-steering-wheel, no-brake-pedal approach.

SS: The press has you aiming there because every form of partial automation is too risky. Because of the "handoff moment," where the machine asks a human to take over when it can't handle something. Is that why? Or is it just chasing the purest version of the problem?

JK: I think it's a little bit of both. It primarily comes from the research we did in 2011, 2012, when we perfected an autopilot sort-of highway-cruise mode. We let dozens [of Google employees] use these cars, self-driving on the highway, to commute to work. We had cameras in the car so that we could … make sure that they were being safe, and just do some learning. We discovered things that were pretty frightening.

SS: You're not talking about nose picking in traffic.

JK: It's the conundrum of this technology: The better you make it, the more relaxed the human agent behind the wheel becomes. And the more relaxed you become, the less attentive you are. It's in that moment, according to Murphy's law, that the car will ask you to take over. But you're turning around, hooking up your phone to your laptop to get some juice, and you're not prepared. We had enough experiences like that to realize the hand over really was just fundamentally challenging.

SS: But that issue is going to be real before we're fully automated.

JK: There's no question it is a more challenging problem to solve. Highway driving is probably an order of magnitude or two easier than the city and suburban driving we're doing right now. Traffic is always flowing in the same direction; differential speeds are relatively low; there aren't many obstacles; there aren't pedestrians or bicyclists. The handoff … the most robust solution is an L4, where you don't have to worry.

SS: What's Silicon Valley's Achilles' heel in a situation like this? Detroit has always held that tech companies underestimate the car business, but Tesla pulled some air out of that.

JK: In general—and I would say this specifically does not apply to Google—I think there isn't sufficient understanding for just how hard it is to make cars. There's not sufficient depth of experience in the complexity of the automobile.

SS: Is that endemic?

JK: I don't know. It may not be. I mean, you've heard the guy down the street [Elon Musk] say, "Yeah, it's hard. This is harder than we thought." This team hasn't had that issue, because it's been focused on perfecting technology that would bolt onto or into automobiles. It's never really felt competition with Detroit.

SS: Even so, there's a general perception that everyone working this problem is in a race to solve it.

JK: We feel a sense of urgency, for sure. But it comes from … every hour that goes by, it's another 737 crashing.

SS: There's really no business end game? Licensing, sales? You're just tackling this massively expensive thing because it's there?

JK: Right now, we're focused on solving the problem. Which business model is the way to ensure the most social benefit, we don't know yet.

SS: If the goal is so altruistic, why so little public cross talk? If testing miles equal knowledge, then Google knows more than anyone else. But you've got few partners, and the data is mostly kept private.

JK: I think you'd be surprised at the number of conversations we have and the common approach on problem solving, trying to solve some of the regulatory issues at the state and federal level. There are various mechanisms … one of them is the great work of the DOT and NHTSA. They've pulled together … all of us working in this space, and are hoping to align the state policy so we don't end up with a patchwork quilt of 50 different state [regulatory regimes] on autonomous driving.

This is probably one of the more efficient ways to go about doing it. Don't burden an individual carmaker. Don't imagine that carmakers could work together to solve this in a collaborative way, 'cause that never happens.

SS: People keep trotting out a metaphor: For enthusiasts, the best end scenario will be like horses. Parks where you go drive on your own, with everything else done by machines.

JK: I think this whole thing is going to unfold over a very long period of time. In our lifetimes? [Automated cars work] best immediately in cities. It opens up so much goodness. Because no one really likes to drive in the city, and cities don't like having all those cars parking in them. Something like 14 percent of L.A. is parking. I think there's obvious undeniable benefit to solutions like that, that help you get from point A to point B where you really don't want to be driving anyway.

SS: So what's the biggest hurdle on the horizon?

JK: One of the toughest things that we have to figure out is, what's the right time to introduce the technology? One point of view is, it's ready when we are equal to the average human driver across all scenarios.

"We have amazing capability to adapt and innovate. " I think we'll always have cars that are a blast to drive."

SS: That virtually immeasurable thing.

JK: Right, how? If you could, you could say, all right, so they're at least as good as us. Let's get them going, because the one thing for sure is that the improvement rate is going to be better for this technology than a human. Because we learn from every car, in every interaction, across the entire fleet nearly instantaneously.

SS: I can't help wondering what else keeps you guys up nights.

JK: The challenge is, society—policy and the regulatory environment—might not warmly embrace [the technology]. We should, if we can come up with metrics that say, look, we're better than the average human driver. But it's going to be hard to do that, because right now, again, it's human error compared to machine error. As a species, we're more prepared to accept human error. We should be agnostic to these things. We should be choosing the thing that leads to the greatest net social value.

SS: Why a Caterham and a 911? You once said that you'd never get rid of either.

JK: I love those two cars for completely different reasons. The 911 is the perfect manifestation of an initial idea that was sort of okay but not brilliant. But through 50 years of refinement, Japanese kaizen, you make this awesome machine. The Caterham is the opposite of that, right? Colin Chapman sort of put the Lotus Seven down, almost perfect from the start. And it received no kaizen, very little care, throughout its life cycle. It too is now over 50 years old. And it too is just an awesome car. It came out nearly perfect.

SS: You said earlier that it's in a shop in L.A., waiting on a motor swap. Why not just get a different Cat?

JK: Sentimental. I don't think I can ever get rid of either of those two cars. I worry that I'm going to feel the same way about my Polestar.

SS: Polestar? Like, Volvo?

JK: Do you know this car? Have you driven it?

SS: I do. I get it and I don't. Is yours the Smurf-blue one in the parking lot?

JK: I actually bought it in black, and I regretted it. So I wrapped it in Rebel Blue. It's sort of weird.

SS: Might be an understatement.

JK: [Laughs.] I know. They made 80 of them.

SS: I sometimes wonder how much longer we have with stuff like this. Every few years, somebody writes that this golden age of automobiles is about to end. Then Dodge spits out a Hellcat, or the Bugatti Chiron happens. How long before that tide goes back out?

JK: I don't think it can.

SS: Really?

JK: Yeah. We have amazing capability to adapt and innovate, and maybe the center of gravity moves from internal combustion to electric or even hydrogen fuel cell. Who knows? I think we'll always have cars that are a blast to drive. What will go away is the drudgery. It's not the Pacific Coast Highway on a sunny Sunday morning. That's part of the myth that we create, both in my old world and in your current world. It doesn't really reflect most people's experiences.

SS: People far more educated than I am believe that it's eventually going to become harder to drive yourself. Whether through taxation or costly insurance policies, it's just going to be deeply discouraged.

JK: I think it's super far off. We humans travel, in the U.S., this pretty amazing number—3.2 trillion miles per year. About 10 trillion globally. I think there will remain human-driven miles for many, many years. So I don't think that's anything we'll see.

SS: Did you ever go look at that MR2 from the ad?

JK: Nah. They didn't reply. [Face falls a little.]

SS: The look on your face when you said that was like, "f***ers."

JK: No, I totally wanted that car and they stopped corresponding with me by text. Actually, the text string is hilarious.

Interview has been edited for length. Special thanks to Caterham distributor Beachman Racing for loan of the CSR, and for going above and beyond the call of delivery duty.


Levels of Automation*

* According to the National Highway Traffic Safety Administration-the outline used by Google's X.

L1 An assistance program, like electronic stability control or cruise control, where a human driver does most of the work. The most common form of automation.

L2 Assistance programs steer the car and control both drivetrain and braking. This includes most currently available "super cruise" systems, like Tesla's Autopilot.

L3 The car can drive itself in a specific environment but requires a human overseer.

L4 The car drives itself in a specific environment, no human intervention needed.