The Speedy Rise and Fall of Robot Babysitters

During the winter of 2017, an 18-year old college student named Canon Reeves spent much of his time trailing a knee-high robot around Fayetteville, Arkansas, as it delivered Amazon packages to students. The robot, created by a start-up called Starship Technologies in 2014, is basically a cooler on wheels; it uses radars, ultrasonic sensors, and nine cameras to make deliveries. Reeves’s job was to monitor how it handled various terrains, field comments from the public, and press the off switch if necessary. He also took photos; many students asked for selfies with the bot, he said. “People would also ask if it could deliver beer.” It couldn’t.

As advances in autonomous technology have placed a huge number of self-driving machines on our roads and sidewalks, a side hustle has materialized in recent years: robot-babysitting. In Phoenix, human attendants will remotely monitor Google's upcoming Waymo robo-taxis, using the car's cameras to evaluate and adapt to passenger or road challenges. State safety regulations typically require that autonomous vehicles be accompanied at all times by humans. These professionals’ job titles range from “robot handler” to “safety driver,” but they have essentially the same responsibilities: monitoring robot behavior for safety and performance, and answering questions about the technology.

The unusual nature of the job lends it some cachet, despite the mundane nature of the work itself. (Job requirements posted by the driverless-car start-up Cruise Automation include: “Able to drive or spend time sitting in a car for six to eight hours a day.”) “I’ll tell my grandkids someday!” Reeves told me. The pay is good, too. Starship Technologies starts its robot handlers at $15 an hour, and Cruise pays $23 an hour, more than double California’s minimum wage of $11. Still, many of these above-minimum-wage jobs aren’t likely to be around for too much longer.

Broadly speaking, robot-babysitting jobs fall under the umbrella of careers in automation, which include maintenance, engineering, and programming. The demand for people with this skill set is considerable, with an expected 20 million to 50 million new jobs expected in this category by 2030, according to the McKinsey Global Institute. In the year that ended in June 2018, the number of postings on Indeed.com’s recruitment boards advertising positions in automation had almost tripled since the year ending in June 2016.

Automation workers include computer scientists, IT workers, and administrators. “Those people are in scarcity, and there’s an extreme demand,” Mike Ramsey, a research director at Gartner, said. But they also include less skilled workers—college students, like Reeves, or others without a degree. As autonomous technology gets more sophisticated, Ramsey expects to see a bifurcation of demand for these robot-keepers. Even as demand increases more technical roles—AI analysts, systems testers, and vehicle technicians—advances in tech and loosening legislation will lessen the need for blue-collar robot roles, like Reeves’ job with Starship Technologies. “We are in this squishy period where the human has to be heavily involved in testing, and that is likely to end,” Ramsey said.

What interacting with robots might reveal about human nature

In some cases, companies have already managed to get rid of their robot babysitters. Over the last year, a 34-year-old entrepreneur named David Rodriguez spent hundreds of hours following a machine called the KiwiBot around UC Berkeley’s campus while it delivered Soylent, Chipotle, and Red Bull to students.

Created in 2017 by a group of Colombian entrepreneurs at Launch, an accelerator program based at Berkeley, the KiwiBot exhibits Pixar-like sensibilities. To retrieve orders, the app prompts students to give the robot a thumbs-up or a wave; the bot’s digital eyes will wink or roll depending on its mood. Rodriguez, who heads business development for the start-up, was tasked, early on, with monitoring the KiwiBot for problems—even carrying it, should the motors fail. Since April 2018, though, the KiwiBot has been largely babysitter-free, and the majority of human interactions involve technical checks and loading food into the bot. To eliminate that grunt work, the team is developing a restaurant robot to collect and load orders—which could happen as soon as 2019. However, Rodriguez assured me that his staff won’t be out of work. Everyone holds dual roles in the company; greater bot autonomy just means employees will shift their focus to accounting, engineering, and design.

Some observers note that certain kinds of robot-babysitting—the kind that is monotonous and doesn’t require much education—can make for thankless work. The safety drivers who sit in self-driving cars have described their roles as “exhausting” and “demanding,” and many told me about the constant pressure to stay hyper-alert at all times. “It’s incredibly hard to sit in a chair and stare at a computer without doing anything for eight hours,” Ramsey said. “But you do not need a Ph.D. to do it.” In March 2018, the field of robot babysitting took a beating when a self-driving Uber in Tempe, Arizona, hit a 49-year-old named Elaine Herzberg. Dashcam footage showed that Rafaela Vasquez, the car’s safety driver, had not been looking at the road when the accident occurred. Investigators are deciding if Vasquez will face manslaughter charges.

Following the accident, a number of executives began pushing to shelve the program altogether, according to the New York Times. Uber suspended all its self-driving tests in the United States, and many of its 400-plus test drivers had to scramble for work. In May, it laid off all Arizona safety drivers, followed by more layoffs in July in Pittsburgh and San Francisco, which dropped Uber’s total test-driver pool to about 55 people. Despite this, Uber said testing will resume this summer and is actively recruiting self-driving-truck drivers and autonomous-vehicle engineers. “Our team remains committed to building safe self-driving technology, and we look forward to returning to public roads in the coming months,” said an Uber spokeswoman.

Driverless Cars Are Like Elevators

Robot babysitting was the first job that Jordan Zagerman, 21, ever held. In 2017, while completing an associate’s degree at San Francisco State University, he followed a delivery robot around San Francisco’s Parkmerced neighborhood, supplying condoms, chips, and soda on demand on behalf of a company called Dispatch. He also managed Dispatch’s Snapchat account and designed branded sweatshirts. Then, after four weeks, came the “it’s not you, it’s me” email. “Over the weeks, people were progressively less surprised with the robot,” he said. “Every robot handler was let go.” The experience convinced Zagerman that he needed to better prepare himself for the future. That fall, he moved to Philadelphia to start a bachelors degree in user-experience design at Drexel University; when he graduates, he hopes to return to Silicon Valley, but this time for a more white-collar, technical position. “I left,” he told me, “to make sure I wouldn’t get phased out with autonomy.”

McKinsey estimates 10 million to 800 million jobs globally could be lost to automation by 2030. In the long term, it’s inevitable that robot-babysitting gigs will go the way of elevator operators and lamplighters. But they’ll also birth new robot-related roles. “A huge number of jobs will be created as autonomous vehicles are loosed into the environment,” Ramsey said. In 2016, Bosch started training students from Schoolcraft College, a community college in Michigan, in autonomous-vehicle repair; Toyota has trained students in maintenance as well. “We might even see a return to low-level jobs where people come and fuel the car for you,” Ramsey said. “Until we can wirelessly charge, someone needs to refuel them.” The hardest-to-automate industries, as it happens, are the ones that require looking after humans: childcare, education, health-care aides. Robot babysitters might feel like they have scored the job of the future. But in fact, real babysitters might be better positioned.