In the nineteenth century, when Karl Benz pitched his car idea to investors, he tried to temper their expectations by telling them that demand would be limited because of lack of chauffeurs.
During the 1939 World’s Fair, General Motors predicted that cars would be driving themselves in twenty years.
My point is that for almost as long as we’ve been imagining cars, we’ve been imagining that someone else would be driving them.
We are Google’s chase team.
What we’re chasing now is the red Prius that’s approximately thirty feet in front of us. The Prius planes and curves on the California 1 as adroitly as any NASCAR. Around even the most rollercoaster-like turns, its speed is a steady thirty-five. Our van hurtles and weaves behind it, hugging the cliffs of Big Sur. More than four hundred feet below, waves break and churn against some very sharp rocks.
It’s Monday, midday, and the sky is set with tiny white clouds. Somewhere near Bixby Bridge, Levandowski points out the window, excitedly, at what he says is a whale. He tells us that its tail just breached the surface of the water and disappeared. He presses the binoculars to his chubby face and continues scanning.
Clarke and I refuse to look. Clarke’s behind the steering wheel. His knuckles are whiter than the clouds above and you can see the muscles in his jaw tense with concentration.
Levandowski points out the window again. “There it is,” he says.
This time we look. Clarke looks and I look.
Down in the bay, there’s a big spray of water, a black hump, and then the van’s wheels are crunching over gravel. Clarke slams the brakes but still we’re slung into the guardrail. Clarke starts yelling at Levandowski. I lift my laptop to make sure my pants aren’t damp.
A few minutes later we’re back on the road. Levandowski’s sulking because Clarke threw his binoculars out the window. The Prius is gone.
I log onto Google Maps and see that the blue dot that signifies “Prius” has stopped moving. When we catch up, we see that the car has driven itself into a cliff. The front end’s crumpled. There’s smoke pouring out the engine. The LIDAR mast, with its forty-two lasers and tens of thousands of dollars worth of navigation equipment, has snapped off the roof and is scattered across the road.
Levandowski puts on his gloves and starts collecting the technology. Clarke looks like he’s going to cry. I open the laptop and begin typing the incident report. The gist of which is:
We are fucked.
Next morning we’re called into Page’s office. That’s Larry Page, as in Google’s multibillionaire cofounder and CEO. He’s read my report and basically agrees with its assessment:
Totally fucked.
He digs into his pocket and pulls out his cell phone, which he tosses onto the desk and asks what we think it is.
What I think is that it’s exactly like that part in Star Wars where, during the Battle of Endor, Admiral Ackbar yells, “It’s a trap!”
Because obviously.
Page clears his throat.
Clarke finds a stain on his tie and begins rubbing the hell out of it. Levandowski focuses on a spot on the wall, just above Page’s head, and stares really hard. I slump in my chair. “It’s a cell phone,” I say.
“What?” says Page.
“Cell phone,” I say louder.
“Wrong.” Page picks up the cell phone and hurls it at us. It whizzes over our heads and explodes against the wall.
What that was, says Page, was a smart phone. And in terms of functionality, he says, it makes his Ferrari look like a retarded piece of shit.
Why, he wants to know, can’t we make a car that’s at least as smart as his smart phone?
Clarke looks at Levandowski.
Levandowski looks at me.
I look at my shoes.
Levandowski sighs.
“There was a whale,” he says, and stops.
“It won’t happen again,” he says.
Page doesn’t say anything. He picks up a stapler and begins squeezing. As we hurry out of his office, the stapler flies into the hallway and dents the wall.
Levandowski and Clarke want to go down to the lab to do another postmortem on the LIDAR. I tell them that I’m going to lunch. It’s not even ten o’clock.
“Whatever,” says Clarke. “Don’t be evil,” he says.
Google’s motto used to be something we took seriously, but now it’s something we say instead of saying something else.
“Don’t be evil,” I say back.
But instead of the commissary, I go to the parking lot and call Julia. Julia’s a graduate student at Stanford.
I’m sleeping with her.
She’s not my wife.
Julia and I met when we were demoing an early version of the Google Car. Both of us were in the Prius’s backseat, letting it drive us around Stanford’s campus. I was looking out the window. There were old brick buildings, green trees, the California sky.
It took me a while to realize that Julia was looking at me.
She was smiling.
“Things are in the saddle,” she said, “and ride mankind.”
She told me that that was Emerson. “‘Ode, Inscribed to William H. Channing,’” she said. When we got back to her apartment, she told me that she was getting her doctorate in literature. She was writing a dissertation about representations of machines in poetry.
“Consilience” is what she wants to call her book. She says that it doesn’t matter that the title’s already taken. In fact, she says that’s kind of the point.
Julia doesn’t answer. I hang up before the phone goes to voicemail.
Instead, I press the icon on my phone that dials Tracy’s number. Tracy who is my wife. Even before the phone starts ringing, I know that she’s going to pick up.
Ten months ago she went on maternity leave and she hasn’t gone off. Whenever we talk about her going back to work because, you know, we’re making half of what we used to make, and so hard times, we end up getting in a big fight.
“Hello,” she says, and I hang up.
I’m in the car and halfway to Stanford before I realize that she’s calling me back.
When I knock on her office door, Julia looks up from her desk, first surprised and then disappointed.
“Oh,” she says. “It’s you,” she says.
There’s a catastrophe of papers spread out in front of her. She’s wearing these reading glasses that magnify her eyes.
I ask whom was she expecting.
She scans the hallway and closes the door. “These are my office hours,” she says. She asks what I want.
What I want, I tell her, is to take her to lunch.
She looks at the clock.
“Brunch?” I say.
We go to her apartment. It’s a one-bedroom in Pacific Heights. Out the window, you can almost see the Golden Gate Bridge.
“You’re sad,” says Julia. She’s holding my penis but she’s essentially right. What I feel most of the time is something like sadness.
My wife has noticed it too. “Do we make you unhappy?” Tracy says sometimes after we’ve put the baby to bed. “Do you realize you frown all the time?”
I tell Julia that we lost a car.
“Another one?”
She sounds truly disappointed. But then she really believes in the project. When we first met I gave her the investor’s pitch, the one that made them go nuts and give us all their money.
In America there are forty thousand automobile deaths per year. Ninety-three percent of those are caused by human error.
Every ten years that’s the population of Oakland gone, in twisted metal and burning rubber, just like that.
At Google it’s our conclusion that, when it comes to being drivers, people basically suck. The Google Car is our hands-on-our-hearts attempt to do something good.
The truth is that we’re developing the Google Car because the technology’s ripe. In ten years, autonomous driving is going to be the industry standard. Every car company, and most of the tech ones, has something in R&D. Drive around Silicon Valley for long enough and it’ll begin to seem like every third car is being chased by a black van.
Back at the lab, Levandowski and Clarke are yelling at each other.
It turns out that in a very technical sense there was no accident. The Prius crashed because it was supposed to crash.
Telemetrics show that something small, perhaps a small animal, ran into the road in front of it. Once the LIDAR picked it up, the car swerved, which is exactly what it’s supposed to do.
The question that Levandowski and Clarke are now debating is whether or not swerving is a good design feature.
Like is it better if the car swerves and crashes? Or if it continues plowing forward regardless of what’s in its way?
Levandowski has a big heart so, of course, votes for swerving. He wants to know what happens next time when it’s a kid or someone’s dog.
“Don’t be ridiculous,” says Clarke.
Clarke says that Levandowski’s argument is stupid. He says we can adjust the LIDAR so that it picks up bigger things like dogs and children.
The real question, he says, is if we’re going to let an eighty-thousand-dollar car crash itself every time a squirrel or whatever runs into the road. “What would a real person do?” he says. He says that’s how the car needs to react.
Levandowski heaves himself off his stool and begins pacing around the lab. He’s muttering to himself.
“I thought people were the bug,” is what I pick up.
Both of them want to know what I think.
What I think is that we’re the guys who drive behind the car that drives itself. You want to ask a moral question, go talk to our CEO.
I pick up a dissected piece of LIDAR and twirl it between my fingers. It’s a mirror attached to a long metal stem.
I say that I guess I’d choose swerve.
Clarke walks out of the room. “What?” I call after him. “I’ve got a kid,” I say.
Of the three of us, Clarke’s the youngest. Google hired him right out of MIT. But in terms of the technology, he probably knows more than Levandowski and me combined. In college he won some kind of big government award for designing a helmet that could mind-read mice. You put the little helmet on the mouse and then put the mouse in a maze. The helmet would tell you a fraction of a second before the mouse turned a corner if it was going to turn left or right.
Levandowski settles down next to me.
We begin picking through the guts of the LIDAR. There are wires and sensors and motors and motor parts. He asks me about lunch.
“A three-hour lunch,” he says.
Levandowski knows all about Julia and his stance is that he disapproves. This is because Tracy and I once invited him over for dinner and now he thinks that Tracy’s great.
The whole time they joked about me like I wasn’t there.
Tracy passed him the green bean salad. “You should see him with the diapers,” she said. She pinched her nose and held the other hand about three feet in front of her face.
“At the office,” said Levandowski, “you ask him a question and he practically dives underneath his desk.”
On average, Americans spend one hundred hours a year commuting back and forth to work. Stretch that out over a lifetime and that becomes eight thousand hours. Which is one full year of life.
Start adding in all of the other car trips you take and the number skyrockets. The conservative estimate is that by the time you die, you’ll have spent eight years behind the wheel.
The only things you’ll have spent more time doing are working, sleeping, and watching TV.
“If you hate your commute so much,” says Tracy, “get another job.”
She’s put the baby to sleep and we’re finally getting around to dinner. The microwave beeps and she puts chicken on the table.
“But who will provide us with our bounty?” I say. I gesture at the gray chicken that looks an awful lot like last night’s chicken.
“You’re resourceful,” she says.
This is pretty much how it goes every night. I complain about something and Tracy deflects it back at me.
“You wouldn’t let us starve,” she says.
I’d complain about the chicken, but I already know how that one goes. Tracy will ask me what did I expect, exactly. She’ll say that it’s hard work raising a kid. Then she’ll remind me that it wasn’t a decision that she made on her own.
She’ll talk about the long and frequent conversations we had, before we even started trying to make a baby, about pulling the goalie from the net.
I liked the idea. The metaphor I guess you’d call it. By pulling the goalie we were choosing to let fate decide.
“Who talks like that?” Tracy said when I told her this. “We’re deciding,” she said.
Later that night, as we undress for bed, Tracy asks me about the phone call from earlier. She wants to know why I called and hung up.
“Phone call?” I say. The baby’s sleeping in her crib at the foot of the bed. We’re whispering so that we don’t wake her.
“You called,” says Tracy. She shows me the Caller ID.
“Huh,” I say. I tell her that the phone must have dialed itself.
She wants to know, on a scale of one to ten, how would I rate that answer. “One being regular bullshit,” she says, “and ten being complete bullshit.”
I tell her that I’d rate it zero.
“Zero’s not an option,” she says.
“Then how would you rate it?” I say.
She turns off the light.
So far, Google Cars have logged more than two hundred thousand miles and still they keep crashing.
For a while their biggest problem was stop signs. The cars couldn’t figure out that people were essentially aggressive drivers and that they rolled through intersections instead of stopping completely. In four-way situations the accident rate was something like seven in ten.
I was the one who came up with the workaround.
By making the AI a little more aggressive we could ensure that Google Cars would roll through stop signs before other cars. Human drivers would be forced to yield to them.
Focus groups basically hate them, and why shouldn’t they? They’re more like something from The Terminator than The Jetsons.
But accidents are down by more than 50 percent.
Next morning I’m standing over the griddle making pancakes when Tracy comes into the kitchen. Pancakes are Tracy’s favorite and are something I have a true talent for. Whenever I make pancakes they always come out perfectly round and perfectly thick.
Tracy puts the baby in the high chair and asks me “What’s the occasion?”
“Can’t a guy make pancakes?” I say.
“I’m not complaining,” she says.
She stands next to me and tugs on the collar of her bathrobe. She’s not wearing anything underneath.
We watch the butter deliquesce and the batter congeal. The smell of pancakes rises into the air. I put a short stack on Tracy’s plate and tell her that sometimes phones really do dial themselves.
“There’s an article on Google News,” I say.
“Are we still arguing about that?” says Tracy.
“Then what are we arguing about?” I say.
Tracy shushes me. She kisses me on the cheek and then on my lips. And before I know it my hands are sliding inside her bathrobe. “You know,” she says, my ear in her mouth, “you’re not a bad guy when you give it half a shot.”
“Neither are you,” I say. My hands are moving south and my palms are becoming sweaty. There’s heaviness of breath. Pancake batter spitting.
“Well,” I say. “Not a guy,” I say.
I tell her that she knows what I mean.
She does, she says.
“I’m an adult woman and you’re an adult man,” she says. “With adult responsibilities,” she says.
Then she lifts the baby from the high chair, wipes food off her face, and puts her in the crib for a post-breakfast nap.
She asks if I’m going to join her in the bedroom. But by then my pancakes are burning. I’m already late for work.
We’re standing in the parking lot outside of Google HQ. It’s overcast and there’s a flock of seagulls in the trashcans. You have to shout to be heard.
Clarke’s holding a baseball and is waiting for the Prius to come around again. We’ve rigged this one with a modified LIDAR. When it’s close enough, Clarke tosses the ball. It bounces off the Prius’s hood and the Prius keeps going.
Then Levandowski tosses a soccer ball. Tires screech as the Prius jerks out of the way.
We do the test several more times. Each time the Prius hits the baseball but swerves for the soccer ball. When we present the results to Page that afternoon we tell him that the Google Car can now calculate between catastrophic and non-catastrophic collisions. In the latter case, we say, the car can now decide that the less damaging decision is to collide.
AR Detection is how Clarke pitches the new feature.
AR meaning Acceptable Risk.
Page squeezes his stapler a few more times. It’s only when he puts it back on his desk that I realize that I’ve been holding my breath.
He tells us that he wants us to start doing road tests.
“This weekend,” he says.
Out in the hall I notice that the dent in the wall has been patched but not repainted. The plaster’s a different shade of white.
Levandowski’s upset about something. You can tell because he keeps making these “hrm” sounds and playing with the knot in his tie.
Clarke asks him what’s wrong.
“Acceptable Risk?” says Levandowski.
Clarke asks him what he’d call it.
Levandowski says, “Hrm.”
Then he says that baseballs and soccer balls are one thing but that he wants to know what happens to the kid who’s chasing the ball.
“Is that an Acceptable Risk?” he says.
Clarke says that nothing happens. He says that that’s the whole point.
By then we’ve walked across the campus and are at the garage and there’s the question of who’s going to sign the car out for the weekend. Clarke says that he will. But Levandowski says that he doesn’t trust him with it.
“Not with Mr. Acceptable Risk,” he says.
While they argue, I sign the waiver and pocket the keys.
Driving the Prius home, I detour through the suburbs. It’s evening and in each neighborhood there are kids chasing kids, kids chasing soccer balls, dogs, kids on bikes. It’s a lot of information to process.
I wonder, what if Levandowski’s right?
For humans there’s a tenth of a second lag between receiving and processing information. Computers are three hundred thousand times faster.
That means that by the time I’ve seen a kid running into the street, have sent the electrical signal to my foot telling it to the hit brake, and have received the signal back from my foot saying that the brake has been pressed, a computer would have done all of it three hundred thousand times.
And yet, even for computers, there are gaps in the information.
Blind spots.
Thousandths of a second when the information is still traveling, when it’s neither been processed nor received.
As I continue driving I find that I’ve detoured my way into San Francisco. The Golden Gate Bridge, once hulking in the distance, is now the dominant part of my skyline. Its two towers, like pyramids on either side of the bay, are shining brightly, orangely, in the sun’s remaining light.
Soon I find myself outside Julia’s apartment. I idle on the curb and watch the sun set. The streetlights blink on.
I kill the engine.
I press the intercom and ask Julia to let me inside.
At Google, in the lab, we literally have a checklist of things we want to make sure the technology can handle before it’s ready for the public.
The list goes:
Left turns: check
Stop signs: check
The list is pretty mundane but at its bottom is the meta-category city driving. Because not only are city streets the most difficult type to navigate, but also because people in the city are the most unpredictable.
The litmus test of our litmus test is San Francisco.
It’s like someone traced over a Pollack painting.
When Julia opens the door, she’s wearing a dress. It’s a short, shiny, tubelike thing that’s designed to show off, among other things, her breasts.
I ask what the occasion is.
She says she’s going to dinner.
I ask if she wants a ride. Then we’re in the backseat of the Prius. There’s the glow of the LIDAR’s GUI and we’re watching the screen’s little hourglass, waiting for the AI to boot up.
I ask Julia where we’re going. My fingers are poised over the touch screen, ready to punch in the address.
“I’m going on a date,” she says.
Which, of course, is obvious. Besides the dress, she’s wearing a honey-and-something perfume that I’ve never smelled before.
I’m not surprised, but still I feel something.
Gutted, I guess.
She asks me what I ever expected. “I mean,” she says, “have you even thought about this? Our situation,” she says and gestures between the two of us as if what we have between us is a situation.
“About how unfair this is to me?” she says.
Out the window the Golden Gate’s towers are lit up. There’s black water in between them. Cars driving across the expanse.
“I’m not breaking up,” says Julia. “If you could even call it that. We can still keep doing what we’re doing.”
I make a noise of assent.
She names a restaurant on Embarcadero. The car starts driving. We go past other parked cars, under streetlights, up a hill.
After I drop Julia off, I let the car drive me home. I sit in the backseat and watch the moonlight on the water. Even in the dark you can see the waves breaking toward the shore.
There are other cars on the road though not that many. As we pass each other, my car and the other cars, I wonder what this looks like to them, the other drivers in the other cars. This driverless Prius, with its steering wheel jerking itself, and its man in the backseat gazing out the window, his forehead pressed against the glass? What will it look like in ten years when all the cars are like this?
All those autonomous cars with their LIDAR masts, AR augmentation, collision detection units, lasers, radars, HUDS, GUIs, infrared cameras, inertial measurements, and 6-D stereo-vision systems? Cars that never get lost and can drive themselves, down to the millimeter, exactly in the middle of the lane? What will it be like for us when we we’re no longer driving ourselves?
I think about Tracy and the baby and myself riding safely in the backseat of a driverless car just like this one. I think about how tomorrow we can all be taken for a ride.
For a while I’m buoyed by the thought.
One of the first things that Julia taught me is that science is the expression of what we’ve learned to keep from fooling ourselves.
“Poetry,” she said, “is also the expression of what we’ve learned to keep from fooling ourselves.”
We were naked in her bed and I’d asked why she was studying what she was studying. Why poetry and machines? I asked if she ever read just for fun.
I told her that when I was a kid I used to read for fun all the time.
“When I was a kid,” I said, and she put her finger on my lips, shushed me. “You’re not a child,” she said.
She began reading to me.
She read me that Emerson poem.
She said, “Things are in the saddle and ride mankind.”
She said, “Go, blind worm, go.”
And I did, eventually.