It’s about the size and shape of an office photocopier. With a gentle whirring noise, it traverses the warehouse floor while its two arms raise and lower themselves on scissor lifts, ready for the next task. Each arm has a camera on its knuckle. The left arm eases a cardboard box forward on the shelf; the right arm reaches in and extracts a bottle.1
Like many new robots, this one comes from Japan. The Hitachi corporation showcased it in 2015, with hopes to have it on sale by 2020.2 It’s not the only robot that can pick a bottle off a shelf, but it’s as close as robots have yet come to performing this seemingly simple task as speedily and dexterously as a good old-fashioned human.
Someday, robots like this one might replace warehouse workers altogether. For now, humans and machines are running warehouses together. In Amazon’s depots, the company’s Kiva robots scurry around—not picking things off shelves, but carrying the shelves to humans for them to pick things off.3 By saving the time workers would otherwise spend trudging up and down aisles, Kiva robots can improve efficiency up to fourfold.4
Robots and humans are working side by side in factories, too. Factories have had robots for decades, of course; since 1961, when General Motors installed the first Unimate, a one-armed robot resembling a small tank that was used for tasks such as welding.5 But until recently, they were strictly segregated from the human workers—partly to stop the humans from coming to any harm, and partly to stop them from confusing the robots, whose working conditions had to be strictly controlled.
With some new robots, that’s no longer necessary. A charming example by the name of Baxter can generally avoid bumping into humans or falling over if humans bump into it. Baxter has cartoon eyes that help indicate to human coworkers where it is about to move. And if someone knocks a tool out of Baxter’s hand, it won’t dopily try to continue the job. Historically, industrial robots needed specialist programming; Baxter can learn new tasks from coworkers showing it what to do.6
The world’s robot population is expanding quickly—as of 2016, sales of industrial robots grew about 13 percent a year, which means the robot “birth rate” is almost doubling every five years.7 For years, there’s been a trend to “offshore” manufacturing to emerging markets, where workers are cheaper; now, the trend is to “reshore,” bringing it back again, and robots are part of that.8 Robots are doing more and more things. They’re lettuce-pickers,9 bartenders,10 hospital porters.11 Still, they’re not yet doing as much as we’d once expected. In 1962, a year after the Unimate, the cartoon The Jetsons imagined Rosie, a robot maid, doing all the household chores. Half a century on, where’s Rosie? Despite recent progress, she’s not coming anytime soon.12
That progress is partly thanks to robot hardware—in particular, better and cheaper sensors. In human terms, that’s like improving a robot’s eyes, the touch of its fingertips, or its inner ear—its sense of balance.13 But it’s also thanks to software; in human terms, robots are getting better brains.
And it’s about time. Machine thinking is another area where early expectations were not fulfilled. Attempts to invent artificial intelligence are generally dated to 1956, and a summer workshop at Dartmouth College for scientists with a pioneering interest in “machines that use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.” At the time, machines with humanlike intelligence were often predicted to be about twenty years away. Now, they’re often predicted to be . . . well, about twenty years away.
The futurist philosopher Nick Bostrom has a cynical take on this. Twenty years is “a sweet spot for prognosticators of radical change,” he writes: nearer, and you’d expect to be seeing prototypes by now; further away, and it’s not so attention-grabbing. Besides, says Bostrom, “twenty years may also be close to the typical duration remaining of a forecaster’s career, bounding the reputational risk of a bold prediction.”14
It’s only in the last few years that progress in artificial intelligence has really started to accelerate. Specifically, in what’s known as “narrow AI”—algorithms that can do one task very well, such as playing Go or filtering e-mail spam or recognizing faces in Facebook photos. Processors have gotten faster, data sets bigger, and programmers better at writing algorithms that can learn to improve their own functioning in ways that are sometimes opaque to their human creators.
That capacity for self-improvement causes some thinkers like Bostrom to worry what will happen if and when we create artificial general intelligence—a system that could apply itself to any problem, as humans can. Will it rapidly turn itself into a superintelligence? How would we keep it under control? That’s not an imminent concern, at least; it’s reckoned human-level artificial general intelligence is still about, ooh, twenty years away.
But narrow AI is already transforming the economy. For years, algorithms have been taking over white-collar drudgery in areas such as bookkeeping and customer service. And more prestigious jobs are far from safe. IBM’s Watson, which hit the headlines for beating human champions on Jeopardy!, is already better than doctors at diagnosing lung cancer. Software is getting to be as good as experienced lawyers at predicting what lines of argument are most likely to win a case. Robo-advisers are dispensing investment advice. And algorithms are routinely churning out news reports on the financial markets, sports, and other subjects.15
Some economists reckon robots and AI explain a curious economic trend. Erik Brynjolfsson and Andrew McAfee argue there’s been a “great decoupling” between jobs and productivity—that’s a measure of how efficiently an economy turns inputs, for instance people and capital, into the products and services that we value. Historically, as you’d expect, better productivity meant more jobs and higher wages. But since the turn of this century, that’s not been the case: at least by some measures, U.S. productivity has been improving, but jobs and wages haven’t kept pace.16 Some economists worry that we’re experiencing “secular stagnation,” where even ultralow interest rates have been unable to spark fast economic growth.17
The idea that technology can destroy or degrade some jobs isn’t new—that’s why the Luddites smashed machine looms two hundred years ago. But as we’ve seen, “Luddite” has become a term of mockery because technology has always, eventually, created new jobs to replace the ones it destroyed. Those new jobs have tended to be better jobs—at least on average. But they haven’t always been better, either for the workers or for society as a whole. An example: one dubious benefit of cash machines is that they freed up bank tellers to cross-sell dodgy financial products. What happens this time remains debatable; it’s at least conceivable that some of the jobs humans will be left doing will actually be worse.
That’s because technology seems to be making more progress at thinking than doing: robots’ brains are improving faster than their bodies. Martin Ford, the author of Rise of the Robots, points out that robots can land airplanes and trade shares on Wall Street, but they still can’t clean toilets.18
So perhaps, for a glimpse of the future, we should look not to Rosie the Robot but to another device now being used in warehouses—the Jennifer unit. It’s a headset that tells human workers what to do, down to the smallest detail; if you have to pick nineteen identical items from a shelf, it will tell you to pick five, then five, then five, then four . . . which leads to fewer errors than if you were told “Pick nineteen.”19 If robots beat humans at thinking, but humans beat robots at picking things off shelves, why not control a human body with a robot brain? It may not be a fulfilling career choice, but you can’t deny the logic.