Evil Geniuses: The Unmaking of America: A Recent History (42 page)

There’s a whole new subdiscipline of technologists and economists predicting and debating what work can or can’t be automated partly or entirely and, depending on the cost, what jobs will or won’t be done mainly by smart machines by what year in the twenty-first century.

Even discounting for digital enthusiasts’ habitual overoptimism, the recent rate of progress in AI and robotics has been astounding. The exponential growth of digital data and cheapening of computer power reached a point in the last decade that allowed so-called deep learning on so-called neural networks—extremely smart machines—to achieve remarkable technical feat after remarkable technical feat. A common task in creating AI software, for instance, is training a system to recognize and classify millions of images. In the fall of 2017 that task typically took engineers three hours to do, but by the summer of 2019 it took only eighty-eight seconds and thus cost 99 percent less. AI can now do decent translations, carry on complicated conversations, beat anyone at Go and the most challenging video games, recognize faces as accurately as people can, and diagnose some cancers better than doctors. To me one of the most interesting recent accomplishments is an AI that designed new AI software as well as or better than engineers could design it. All of that is why the funding of AI start-ups quadrupled just between 2015 and 2018, to $40 billion, and why the total investment put into in AI businesses in 2019 reached $70 billion.

The debate among technologists tends to focus on when they’ll manage to create artificial
general
intelligence, machines able to figure out any problem and carry out any cognitive task that a person can. People at Facebook and Google and Stanford and elsewhere say they’ll do it by the mid-2020s, that they’ll then have machines “better than human level at all of the primary human senses” and “general cognition” (Zuckerberg), true “human-level A.I.” (the head of Google’s DeepMind). The state of the art right now is “narrow AI” or “weak AI,” software that can merely beat human champions at
Jeopardy
or predict the shapes of cellular proteins or drive cars. But most jobs are fairly “narrow” and don’t require a lot of high-level creative problem-solving. I used to hire freelance transcribers and translators, but in the last few years I’ve replaced them with software that does the work a little roughly but well enough to serve my needs. The first industrial revolution took off around 1800 when steam technology improved from impractical to okay after James Watt designed an engine that captured 3 percent of the energy of the coal it burned instead of his forerunners’ 1 percent. (Two centuries later the giant steam engines that still generate most of our electricity still operate at less than 40 percent efficiency.) The good enough beats the perfect.

As for when and how many particular jobs will be taken over by machines, either disembodied AI or robots, the estimates range widely, but pre-pandemic most predicted that between 15 and 30 percent of current jobs in the United States and the rest of the developed world will be eliminated during the next ten to twenty years, with many more “at risk.”

A survey conducted of Davos celebrants of the Fourth Industrial Revolution gives a more focused sense of the remarkable speed of the rise of the machines in the immediate future. The Davos researchers asked the relevant executives at three hundred big corporations employing 15 million people about the impact of automation on their companies’ work and workers. In 2018, machines were doing 29 percent of all the work at all those companies combined, the “total task hours”; just four years later, in 2022, the executives collectively forecast, it would be up to 42 percent. They expected that even the parts of jobs “that have thus far remained overwhelmingly human,” such as “communicating and interacting,” “managing and advising,” and “reasoning and decisionmaking,” would go from an average of 20 percent automated to 29 percent automated by 2022. No wonder the executives at half those companies also said they were planning to shrink their full-time workforces during those same four years. The Davos report euphemistically admitted that the survivors’ jobs would keep getting iffier because, on top of accelerating automation, the companies were making a “significant shift in the quality, location, format and permanency of new roles” and would “expand their use of contractors” and “workers in more flexible arrangements.”

The hopeful spin by some experts, even some scholars, has been that the machines will be our
helpmates,
not our replacements. Mostly. For the time being. But Martin Ford, the Silicon Valley investor, says beware of assurances that the “jobs of the future will involve collaborating with the machines,” because “if you find yourself working with, or under the direction of, a smart software system, it’s probably a pretty good bet that you are also training the software to ultimately replace you.” The authors of
What to Do When Machines Do Everything
—three executives at the huge digital services and consulting firm Cognizant, whose whole business is about enabling corporations to shrink their workforces—absurdly promise that while some jobs will “be ‘automated away’ in the coming years…for the vast majority of professions, the new machine will actually enhance and protect employment.”

Walmart, which employs more Americans by far than any other company, leans hard on that enhance-and-protect line. “Every hero needs a sidekick,” said its cute 2019 press release headlined #SquadGoals, “and some of the best have been automated. Think R2D2, Optimus Prime and Robot from
Lost in Space
.” In about half of its several thousand American supercenters Walmart had just installed three thousand actual autonomous robot workers—one breed called the FAST Unloader takes merchandise off trucks and sorts it, another perpetually roams the store neatening shelves and notifying the FAST Unloader which items are missing, and a third cleans the floors. “Just like Will Robinson and Luke Skywalker, having the right kind of support helps our associates succeed at their jobs.” Of course, the new era of viral disease might eliminate lots of associates’ jobs more quickly, because robots don’t get sick, or sicken customers easily.

Walmart’s main competitor, Amazon, has been introducing machines to replace its tens of thousands of warehouse workers who pack orders. Humans can pack two or three a minute, but the current robots can do ten or twelve a minute. Like Walmart, Amazon accentuates the positive: because the job is crappy, turnover is high, so instead of laying off packers to bring in more robots, the bosses were saying, pre-pandemic, they’ll just “one day refrain from refilling packing roles.” Similarly, a small Massachusetts manufacturer of medical and aerospace parts, Micron Products, suggests it did a favor for the employees it recently replaced with robots. Yes, buying the robots made economic sense, because they cost only what the twenty-dollar-an-hour assembly-line workers were paid every seven months, but—win-win—the head of business development says “this was the worst job in the plant. They”—the people replaced—“had to work at a furious pace.”

Fast-food restaurants employ about 4 million Americans, almost half again as many as they did a decade ago. A sleek, groovy new hamburger place called Creator, in San Francisco’s SoMa neighborhood, has very, very few employees, although Creator’s messaging is all hospitably first-person-plural. “Food is communal,” the website says. “We love hanging out at the store, creating burgers. Our space is your space; we want you to feel at home eating here with people you care about.” The six-dollar hamburgers, quite good, are made entirely by a robotic system that has been in development for years. In fact, the restaurant is actually just the public showroom of a Google Ventures–funded technology company whose founder and CEO forthrightly admits that its mission is not “to make employees more efficient” at the fast-food chains to which they want to sell their technology, but rather “to completely obviate them.”

Among those fast-food workers are some of the 3.4 million American cashiers, who are officially distinct from the 4.1 million American retail sales workers—and obviously the great majority of all of them are replaceable sooner rather than later by e-commerce and improved self-checkout machines, known in the industry as “semi-attended customer-activated terminals.” Starting now, retail chains will have a public health argument for replacing workers behind the counters with machines. The most common American job, however, has been driver—the 4 or 5 million FedEx and UPS and tractor-trailer and bus drivers, and the maybe 2 million taxi and Uber and Lyft drivers. During this decade, autonomous vehicles will begin making the 6 or 7 million (potentially infectious) people doing those jobs redundant as well.

That debate over whether to blame automation or cheap labor for eliminating U.S. jobs and suppressing wages is continuing to become moot, because robots are replacing foreign workers as well, both here and abroad. Starting in the 1990s, for instance, American customer service operators were replaced by cheaper ones in India and the Philippines and elsewhere—and now the humans abroad and in the United States are being replaced by AI chatbots. The Taiwan-based company Foxconn is by far the largest manufacturer of consumer electronics on Earth—TVs, gaming consoles, iPhones, iPads, Kindles—in factories all over the developing world. And its chairman said at Davos in 2019 that Foxconn intended to replace 80 percent of its 1 million employees with robots during the 2020s.

Back during the period when America massively offshored manufacturing and other jobs to Foxconn and other foreign companies, we were also massively onshoring workers from foreign countries to the United States by means of immigration. We more than doubled the foreign-born fraction of our workforce between 1980 and 2007, when it stopped increasing.

Take agriculture. Three-quarters of the million-plus people paid to work on U.S. farms were born in Mexico and countries to its south. Now that the influx of new immigrants has mostly stopped and seems highly unlikely to increase anytime soon, there’s upward pressure on farmworkers’ wages, which means the economics argue more strongly for replacing farmworkers with machines. At the California company Taylor Farms, the largest supplier of cut vegetables in the United States (and the world), robots already harvest most of the lettuce, cabbage, and celery, because they can do it twice as fast as people. Soon robots will be picking fruit, a more delicate operation. For instance, a robot made by a company in Florida that can harvest eight acres of strawberries in a day, the work of thirty people, is just about good to go. And if farmers and growers buying robots make themselves feel better because they’re relieving people of tedious, backbreaking labor, they’ve got a point.

Machines want to do all our work. We’re letting them. We will keep letting them. And the right turn we made forty years ago has made our political economy and government particularly ill-equipped to deal with the transformation.

Most jobs probably won’t become superfluous for a couple of decades, but that’s not the distant future unless you consider the 1990s or early 2000s the distant past. Of his coinage
useless class
for the eventual majority of humankind, all those whom AI will make economically obsolete, Harari says he

chose this very upsetting term…to highlight the fact that we are talking about useless from the viewpoint of the economic and political system, not from a moral viewpoint….I’m aware that these kinds of forecasts have been around…from the beginning of the [first] industrial revolution and they never came true so far. It’s basically the boy who cried wolf. But in the original story of the boy who cried wolf, in the end, the wolf actually comes, and I think that is true this time.

It will be a terrifically difficult transition to navigate, and essential to avoid turning people whose economic boats have already stopped rising, the majority of Americans, into a pathetic or contemptible useless class. Not only are the changes required economic
and
political
and
cultural, but some are urgent and some aren’t. Two basic goals will seem contradictory: we want everyone with talent or passion for their work to keep working, and all employees to be treated with fairness and respect
now,
but for the long term we need to start making self-respect and usefulness more independent of employment, to educate and enable and encourage Americans to be and feel engaged and useful and respected regardless of how they receive their fair share of the national wealth.

One of the main challenges will be changing what Harari calls the moral viewpoint. We need to think of his scary wolf, AI and robots, not
necessarily
as a terrifying predator. Instead, they can be like the gray wolves that we tamed thousands of years ago and turned into humans’ best friend—dogs. Technological unemployment and its approaching endgame are indeed an existential threat, but they’re also a potentially grand existential opportunity. And taking advantage will first require a shift by the United States to some kind of economic democracy, taking the power away from big business and the rich to write all the rules only to serve themselves.

There will be political and cultural problems at every step of the way, no question, but we need to keep our eye on the prize: it’s all about solving the one overriding problem—what economists call
the economic problem,
how people decide how to use the available resources to survive and, beyond mere survival, to enjoy life.

In that 1930 essay about the future, as the Great Depression was just descending, Keynes asked his readers to imagine “that a hundred years hence we are all of us, on the average, eight times better off in the economic sense than we are today.” America’s GDP per person at that time was $8,221 in today’s dollars, and at the end of 2019 it was $62,853—exactly 7.6 times better off in the economic sense, right on track to meet Keynes’s eight-times-better-off deadline, with a decade left to go. So around 2030, he speculated, “the economic problem may be solved, or be at least within sight of solution.” But once that ancient problem is solved, and people can work less and less to live comfortably, the next problem arises.