Three of 600 Chrysler Pacifica Hybrid vans in Waymo's autonomous-car fleet are parked outside SoHo63 in Chandler, Arizona.Caitlin O'Hara

CHANDLER, Ariz.—Across Arizona Avenue from Waymo’s self-driving-car showroom sits the Crowne Plaza San Marcos hotel, which is allegedly haunted. According to employees and guests, the ghost can move plates, knock phones off cradles, even—helpfully!—fold clothes. I passed this knowledge on to a Lyft driver, who retorted, “I didn’t know that, but to be honest with you, what isn’t haunted?” Every western boomtown has its ghosts, each wave of fresh pavement and new money disturbing the traces of what lived and died there before.

Chandler, just southeast of Phoenix, has added 220,000 of its 250,000 residents since 1980. Intel, PayPal, the security outfit Northrop Grumman, and a variety of other big companies have set up shop here in the past few decades, chasing tax breaks, a lower cost of living, and lots of square footage. Nothing in Chandler is supposed to look or feel old. Instead, it is either new or waiting to be new, like the scraped-flat parcels of land near what passes for downtown. And now weaving past every green-glass office park, poolside condo, and mobile-home park is the ultimate symbol of newness, the self-driving car.

Waymo, a subsidiary of Alphabet, began testing cars here in early 2017, and the fleet has been growing ever since, serving a handpicked sample of the community through the “Early Rider Program.” Now it is launching to the public under the name Waymo One. In the coming weeks, a growing number of Chandler residents beyond the Early Rider Program will be able to hail Waymos.

Autonomous cars have been under development by various companies for decades, but the industry’s modern era exploded out of a series of Defense Advanced Research Projects Agency competitions in the 2000s, when a few university teams, funded by the military, began to make huge progress on self-driving vehicles. Breakthroughs in technology, mostly in the laser-range-finding systems known as LIDARs, allowed the cars to understand the three-dimensional space of the world. Mapping projects from Google and others helped encode human rules of the road for machine consumption. And, of course, in recent years, ever more computation and the use of machine-learning models have improved visual processing tasks like pedestrian detection.

A pedestrian crosses in front of a Waymo driving autonomously during a demonstration on November 27, 2018, in Chandler, Arizona. (Caitlin O’Hara)

The driverless car is a monument to our age, synthesizing everything that Silicon Valley can be: brilliant, farsighted, rapacious, data-hungry, convinced of machine competence and human fallibility. It is tech’s will to change the world made manifest, and it will—eventually, almost certainly—reshape cities in the most mundane and significant ways.

In preparing for Waymo’s continuing deployment, Chandler’s planning commission hewed to the mundane: The primary change it considered was a 10 percent reduction of parking in exchange for developers constructing dedicated pickup and drop-off locations. Such civic administrivia could obscure the fundamental, astounding new fact of the world: Waymo One is an autonomous-car taxi service that normal people—hundreds, then thousands—will use to summon robots for a quick ride to Dee’s Dancewear or volleyball practice or Walmart. There will be many firsts along the way to the self-driving future, but this is the end of the beginning.


Last week, I crossed the sunny street, got in to the backseat of a Waymo minivan loaded with far more computing power than an early-’90s supercomputer, and hit a big blue button marked Start Ride in English and braille. There was a guy in the front seat, young and ponytailed, but he did nothing but go along for the ride, hands off the wheel.

The car took off down the street, destined for a strip-mall bar called the Priceless Prime Time, one of a handful of destinations that Waymo allowed us to go to as part of a press preview of the technology. It got to a stop sign at Arizona Avenue, the name for busy Highway 87, which pulled this town into existence as a crossroads outside a still-tiny Phoenix. Traffic surged past us, and the car waited patiently—maybe a hair too patiently—and then pulled out and merged with the stream of dumb machines heading north.

Inside the car, none of the technologies that make the service work were particularly obvious. There was a screen in front of me, and it showed an abstracted view of what the car “sees”: how it synthesizes all of its inputs into a sense of where it is, what else is there, and where it needs to go. The screen is, essentially, a map drawn by engineers imagining what a normal human would want to know about how the robot he has temporarily entrusted his life with understands the world.

It shows the shape of the road in perfect detail, a bright little model of the minivan pursuing the line of its path through the world. Other vehicles are shown as blue rectangles. Every four seconds, the rest of the world comes briefly into existence: a line of palm trees, the curve of the newish Chandler Center for the Arts, a psychic’s squat little parlor, the halal-market parking lot—there, then gone. People are shown as three-dimensional clouds of gray points moving atop a circle; they shimmer like a movie ghost or a school of tiny fish. I found myself looking out the windows to remind myself of the solid volume of the humans dressed in flesh and clothes on their way to the pharmacy.

Given the ubiquity of ride-hailing apps now, riding in a self-driving car is surprisingly natural, even boring. That is, until you notice the wheel turning on its own while the guy monitoring the vehicle sits with his hands on his thighs. It’s enchanting and unnerving, like a spirit is driving the car. When Alphabet, GM Cruise, Uber, and other companies investing in autonomous cars evangelize the technology, they present it as a way to escape the tedious, dangerous moil of driving. But encountering the technology in the field, on actual city streets, doesn’t trigger the exciting pang of future freedom. Instead, it produces a stranger feeling: discomfort that an ethereal intelligence inhabits the human world.

A Waymo car isn’t just a car. It’s a cluster of other stuff originally alien to automobiles: chips and code, hardware and software, laser range finders and cameras, machine-learning models and pathfinding algorithms and mountains of map data, fine-tuned by human hands moving mice. It knows every inch of this place with more precision than any human, has driven more miles than any long-haul truck driver, is one of the most expensive technological projects ever attempted by private industry, and could be worth hundreds of billions of dollars to Alphabet. And yet … none of that is comforting. Quite the opposite.

The self-driving car is humanlike, but not quite alive. Built new, it is made of the immortal memories of 10 million miles its kind has driven and the structured dreams of 7 billion more miles that the car has virtually run around a computer model of the East Valley, outside Phoenix. Turning the wheel, pressing the pedals, it works like the ghost of the technologically extended human, the one who needs a car, the one who parks.


In the late-19th century, farmers had a name for the automobile, this horseless carriage that moved on its own: the devil wagon.

And who is to say that’s wrong? The car, powered by humans’ incredible but inconsistent driving skills and the internal combustion engine, blew cities apart, scattering them around ever-expanding highway rings. Most American places, rural or urban, bow to their logic. Commuters can spend an hour or more getting to and from work. Whole geopolitical regimes are held in place by the world’s petroleum needs. And cars kill 1.2 million people annually across the world, about 40,000 a year in the United States. This number is decreasing, thanks to consumer protections like seat belts, less drunk driving, better emergency medical care, and safer roads, but there are no proven ways to make driving perfectly safe.

This new thing makes new mistakes. On one ride, the car stopped short of the intersection at a red light, then haltingly inched forward to the proper line. There were a lot of cars coming quickly in the other direction, and the Waymo seemed to be worried about them. If you’d never seen a car before, the sight of several thousand-pound vehicles barreling toward you a few inches to your left would probably be terrifying. But the long experience of the automotive world has taught most people that cars stay in their lane almost all of the time, statistically speaking. And in any case, stopping short of the crosswalk wouldn’t save you anyway. The decision was extremely computer, barely comprehensible by human logic.

Another time, I hailed a car immediately after being dropped off. As I stood there on the curb, the Waymo took off, slowly, headed down the street, turned around, and pulled back up to the curb. Robot slapstick. Why drive off and repark when you could … do nothing? It’s in these moments that you can glimpse the odd way the Waymo understands the city.

Odd to a human like me, anyway. It knows every technical detail of a situation—street widths, curb heights, speed limits, velocities of other vehicles, pedestrian shapes—and absolutely nothing else. It’ll leave you standing next to it, drive around, and come pick you up again. Understanding how its robot brain functions is like trying to imagine how a bat uses sonar to navigate or how a heat-sealing snake uses its sensory nose pits. This is an other intelligence that has been taught to understand a slice of our world, but what we are allowed to know of its universe is partial and imagined.

The user interface is visible as a Waymo drives autonomously during a demonstration. (Caitlin O’Hara)

So it’s not surprising that the cars do small things differently from human drivers. When the car dropped us off at the Prime Time bar, it stopped awkwardly near the end of a parking strip, jutting out into the parking lane at an angle rather than pulling up parallel to it. It wasn’t dangerous, but I’d just literally never seen anyone stop a car like that in my life. The decelerations were jumpy in the way you induce when you get in to a rental car and aren’t used to the brakes. The accelerations, meanwhile, were smooth, but remarkably fast. The goal seemed to be to get up to and stay precisely at the speed limit as quickly as possible. And why not? Maybe it’s humans who have been riding the wrong acceleration curve to maximize speed and minimize energy usage.

Only once, on our very last turn of the ride, did the car cause me to worry. We had to make a left across Arizona Avenue’s two lanes. Oncoming cars streamed along in bursts. For a second, the Waymo seemed to see an opening and began to move—but then, rapidly, jammed itself to a halt. It felt a little like riding in a car with a new driver. A moment later the street completely cleared and we proceeded through.

But from what I observed in the tightly controlled point-to-point drives around town, and what Waymo’s data presumably show, this technology has become safe enough to drive many, many miles without getting into an accident in a low-density suburb of Phoenix. I did not worry about my own safety or anyone else on the road a single time in the hour of riding around. And in one case, the car made a tremendously impressive move. Right after a right turn, heading south, the Waymo announced it was changing lanes—“Left,” the voice rang out in the car. Then the computer spotted a big truck that was trying to merge into the Waymo’s desired lane. In an instant, the car said “Canceled” and smoothly guided us back into safety in our original lane, before waiting for the left lane to clear and getting over, unscathed.


Waymo—the car, the computer—can do enough. Now Waymo—the company—has to mesh its product with the city and build the system of the self-driving fleet. It has chosen the hermit-crab route, borrowing preexisting pieces as it stealthily sets up shop.

In the far north of Chandler, across the street from a U-Haul center, there’s an alley. On one side sits Casas del Campo Mobile Home Park, packed with trailers and RVs. On the other is a Budget Rent a Car facility that has become a Waymo storage-and-refueling station through a partnership between the two companies. Around 5 a.m., the humans who operate the robot fleet come in, bearing coffee, getting ready for the day. They head out to serve the people until a shift change in the afternoon, when they clean out cup holders, stretch their legs, and fuel up at a 6,000-gallon gas tank tucked behind a wall. At 10 p.m., the cars come back here, drivers chitchatting. For now this is a job—well, a contract position, usually run by the staffing firm Adecco. There’s no particular line of work that seems to qualify people to monitor the robot driver, because it simply was not something that was possible until a couple years ago.

Just across the alley from the facility sits a trailer with a palm tree and a concrete deer family posed in the front yard, and a plastic Christmas tree on the porch. Inside an ashtray, the butts of dozens of Marlboro Reds splay like seeds in a sunflower. This is Joe Rak’s house, and he’s had it for 35 years.

Rak’s got a gray mustache and a mullet. When he sits on his porch, smoking, what he looks at now are the big sensor arrays on top of the Chrysler minivans that Waymo uses. Rak fought the installation of the facility with the city. Not because he has anything against self-driving technology itself, but because of the refueling facility’s huge gasoline tank, which, as a man with a Marine Corps sign hanging in his house, he referred to as a “500-pound napalm bomb.”

He contends that the city ignored its own building codes, which don’t allow fuel tanks within a certain proximity to homes or other fuel tanks, to allow Waymo’s plans to proceed. But when Rak brought up this prospective legal complication, the city told him that the code had been changed, and dismissed his concerns, he said. In the meeting minutes of the Planning and Zoning Commission, one member “mentioned there’s a large investment from Budget on these types of things so he’s not too worried about it.” Rak maintains that he’s not the kind of guy who goes around making a stink, but he did because he felt the commission was “bullshitting” him.

“This was in reference to where I live. And I don’t like to have a 500-pound napalm bomb right there,” Rak told me. “And they’re going, ‘It’s not a napalm bomb,’ and I’m going, ‘6,000 gallons of gas,’ you know?”

Joe Rak smokes a cigarette on his deck overlooking Waymo’s refueling station in northern Chandler, Arizona. (Alexis Madrigal / The Atlantic)

RoseMarie Horvath, a city attorney, pointed me to Resolution 4698, which the council adopted May 12, 2016, and which eliminated the portion of the fire code on which Rak’s claim rested. The entire section on aboveground tanks was eliminated when the city imported, whole, the “2015 edition of the International Fire Code.” By the June 24, 2016, edition of the published code, the section was gone, long before the refueling station was built.

Which is not to say that putting a huge tank of gasoline right near some homes is an ideal situation. This is one reason the city required the company to get a use permit, which it did. “In my personal opinion, it’s just money. The money was thrown at the city. They jumped on it,” Rak said. “And screw everybody else.”

And the truth is that many types of development require some people to sacrifice. Each economic boom must bury previous uses of the city. And Chandler didn’t grow its population and economic base so spectacularly by being unfriendly to business interests.

For the city and for Waymo, the technological project of the self-driving car is grand, but building a real service requires smaller, grubbier things to be done. These are the things that separate a technology demonstration or a pilot program from a commercial service. So the partnership with Avis Budget is just one of many that the company has created in advance of this commercial launch. Another is with AutoNation, the world’s largest group of car technicians, to act as mechanics. Another is with the local mall and a Walmart for special drop-off zones.

These latter arrangements don’t seem significant, but right now it is the pickup and the drop-off that pose the greatest challenge to an actual commercial service. Humans know intuitively where and how to stop; self-driving cars do not. That means Waymo has had to individually delineate curb areas on every single street in its service area as “safe” for picking up and dropping off.

The big-box store or strip-mall parking lot is the next-level challenge. I went out to a Walmart a little south of downtown in the mid-morning. Dozens of people swept into and out of the store, cutting in front of cars, weaving unexpectedly into and out of parked vehicles, trailing children, pushing carts. Here is where unwritten human rules supersede the technical guidelines for how parking is supposed to work. Waymo’s minders did not allow me to take the car to a busy parking lot, but one local, Jim Guccione (yes, he is related to the Penthouse publisher), told me that’s where they really struggle. “They get confused a lot in parking lots,” he said.

Waymo spokespeople told me that they’d already learned a lot about these problems. For example, people arriving at Walmart want to be dropped off right out front, but leaving, loaded with bags, the pickup spot should be in a less trafficked location. Still, if Waymo’s early trials run into problems, expect them to come in the parking lot.


Before it was inhabited by cars, the Phoenix metro area was inhabited by indigenous people. The name the group was given by the archaeologist Harold Gladwin was Hohokam, drawn from an indigenous dialect and meaning “those who are gone.” Out by the airport sits the Pueblo Grande Museum, once home to a large pre-Colombian village. Inside, a football-field-size mound testifies to the long duration of the Hohokam’s habitation. Another trace, more ghosts. Investigations of the human remains on the site have found that the village population of perhaps 1,000 people “was suffering from malnutrition, despite their utilization of all available resources.” Their old territory in this valley now holds 4.7 million people.

These days, there is no end of resources for cities. Money, goods, and information come from all over the world, channeled into population centers. It’s not always an elegant operation. Outside any city, there is usually some cheap place near the freeway where logistics facilities stretch for miles, truck bays awaiting workers who will carry their goods to retail. In Chandler, this place is in the southeast part of the city, near the intersection of Interstate 10 and Highway 202, right before settlement falls away into desert that extends past the border and on to the Gulf of California. This is the backstage of the city, and nestled down there, next to a juice distributor, is the 68,000-square-foot Waymo depot. At any given time, there might be 30 cars inside and a bunch parked outside. The actual facility could not be more anonymous, but it is the core of the operation.

“It’s been cool to see the growth of it,” Daniel Chu, Waymo’s head of product, told me. “When I first went there, it was an empty building, and now it is bustling 24/7 with rows and rows of cars.”

Waymo has hired what Chu said was “hundreds of local residents” to support the company’s extended beta test. Now that Waymo One, the commercial service, is beginning, the depot will have to keep growing. The company has ordered tens of thousands of cars for delivery over the next few years, although presumably they won’t all end up in the Phoenix suburbs.

Daniel Chu, the director of product for Waymo (Caitlin O’Hara)

The company is being very, very cagey about how quickly the service will, should, or can grow. When it launches December 5, the Waymo app will not simply be available in the Android or Apple app stores. Not everyone who wants to hail a ride will be able to do so. The business will slowly grow. Waymo is not making any public commitments to an expansion trajectory.

For now self-driving cars are in a strange moment between impossible dream and everyday conveyance. For people who’ve been watching their evolution, it has become a foregone conclusion that cars will become more autonomous, and the real debate is how quickly, and how much, they’ll disrupt existing transportation modes. Most of these discussions are abstract, based on theoretical assumptions about the usage rates and access and user behavior.

And that’s why Waymo One is so highly anticipated. It’s a lot easier to imagine an organized network of service-and-fueling stations than it is to actually install one next to a mobile-home park or build out a warehouse by the freeway. Academics might imagine that future autonomous vehicles will pack themselves in like sardines into much smaller parking lots, but right now self-driving cars must simply contend with chaotic swarms of humans. They need to ace the Walmart parking lot.

A Waymo is as powerful a stimulant of the algorithmic imaginary as Facebook’s News Feed rankings or Google’s search-result choices. How does this complex system work? Every single person encountering a Waymo—inside or outside of it—develops folk theories of how the car operates.

A couple walk past a Waymo Chrysler Pacifica Hybrid during a demonstration in Chandler, Arizona. (Caitlin O’Hara)

Though people use humanlike language to describe the car (“Waymo picked me up!”), they are not really anthropomorphizing. They do not expect humanlike behavior out of the thing. Rather, it’s more like behavioral pareidolia. Any time the car does something you would do as a driver, you impute similar reasoning. When it doesn’t, it becomes a machine again. We keep half-recognizing our own kind of behavior—and then it slips away. Our brains see other brains everywhere, too, even when we know we’re wrong.

In fact, the illusion of similarity between human and computer drivers is probably the most potent thing in Waymo’s push for technological acceptance. If enough people can simply imagine that there is nothing all that different about self-driving cars, then Waymo can slip in to society. And once the robots are doing the driving without human minders, they can optimize in the way Silicon Valley does, grinding down the price through operational efficiency and increasing the availability of transportation that robots can deliver. This is the ghost of car-human past and capitalism future, available now in a desert boomtown and coming soon to a low-density suburb near you.

We want to hear what you think about this article. Submit a letter to the editor or write to letters@theatlantic.com.