Robots at Work: Where Do We Fit?

The robots are coming for our jobs—and sooner than we think. That’s the gist of a number of recent reports by economists and technology researchers. For instance, nearly half of all U.S. jobs could be automated within a decade or two, cautions a study by an Oxford University economist and engineer. Smart machines will replace one in three jobs by 2025, warns technology research firm Gartner. Robots will perform 45 percent of all factory tasks by 2025, up from 10 percent today, blares Bank of America.

We have been down this road before, other economists fire back. Two hundred years ago, English textile workers felt so threatened by power looms that they started smashing machinery. Worries like those of the Luddites also arose when mechanization scythed through farm hands, when automation first threatened factory workers, and when PCs began to eliminate secretarial jobs. Every time, productivity grew, the economy thrived, and employment rose. Why wouldn’t that be the case today?

“What’s new is that algorithms are sensing things and reacting almost as well as a human would,” said W. Brian Arthur, a visiting researcher at the Intelligent Systems Lab at Palo Alto Research Center, whose theories shaped our understanding of the high-tech economy. “We’re living in a world where, for the first time in human history, we can get a lot done, not just in manufacturing but in the service economy, extraordinarily cheaply and automatically.”

Algorithms have already eliminated millions of jobs among factory workers, video store clerks, travel agents, bookkeepers, and secretaries. Middle-skill occupations, which require more schooling or training than high school but less than a four-year college, fell from 60 percent of all U.S. jobs in 1979 to 46 percent in 2012. Similar declines occurred in 16 European economies.

Algorithms running on interconnected computers have reshaped entire industries. Arthur points to the Blockbuster video chain: “It doesn’t employ fewer people, it’s gone. All the travel agents that populated Palo Alto have disappeared.”

Now algorithms are invading the skilled professions. Software is replacing some loan officers, attorneys, and sports and business journalists who write news. IBM is modifying its Jeopardy-winning Watson technology to diagnose diseases and read medical images. And engineers increasingly rely on expert systems to assess designs and simulations.

In the past, when mechanization disrupted farming, laborers took on factory jobs. When factory employment flattened, workers moved into offices. “Today,” Arthur said, “we don’t have a sector that is growing fast enough to mop up those people who get laid off.”

;custompagebreak;

The Second Economy

Five years ago, Arthur coined the phrase “the second economy” to describe a system in which Internet-enabled computers execute business processes once handled by people.

Twenty years ago, for example, when we wanted to travel we called travel agents, who would ask where and when we wanted to go, query some proprietary databases (or even paper catalogs), talk us through our options, and book reservations.

Today, we simply go online. This sets off a conversation among machines. Software gathers information about available flights. It charges our credit card, checks our preferences, reserves our seat, and sees if we qualify for a security clearance or lounge access. It adds our weight and baggage to the flight manifest, and orders additional fuel for the flight.

Everything takes place within seconds, without any human intervention. Similar conversations are happening everywhere in the economy—between RFID tags and scanners at ports and warehouses, between television sets and servers that stream movies, and more.

“We’re undergoing a digital revolution, a transformation of the economy comparable to the Industrial Revolution. The Internet of Things is creating masses of digital sensors, and they are going to generate masses of data,” Arthur said. As new algorithms arise to make sense of the data, they will only strengthen the second economy.

This transformation is already having a profound effect on people’s employment prospects. And its impact is growing because computers are getting smarter, quickly.

Only a few years ago, for example, voice recognition worked in highly structured dialogues, where it expected certain types of responses. Today, Apple’s Siri and its competitors provide (mostly) relevant answers to unstructured questions. Some apps even translate languages on the fly.

Autonomous cars show just how fast AI can evolve. In a 2004 DARPA challenge, the best car chugged only seven miles down a straight road. Three years later, six autonomous vehicles completed a 60-mile circuit through an abandoned military base among moving cars, pedestrians, and street signs. Their performance was not as good as a human. This led Silicon Valley entrepreneur Martin Ford to predict that AI was not likely to replace human truck drivers anytime soon.

One year later in 2010, Google announced that its fleet of autonomous cars had logged 100,000 incident-free miles. Today, nearly every large auto company has an autonomous car program. This past May, Nevada granted the first commercial license ever to an autonomous truck.

Ford, meanwhile, appears to have reconsidered the speed of change. He named his most recent book, Rise of the Robots: Technology and the Threat of a Jobless Future.

Getting Smarter Faster

According to Gill Pratt, who spearheaded DARPA’s robot challenge and now heads Toyota’s $1 billion robotics program, two emerging technologies will help robots learn even faster.

The first is cloud robotics. In the past, memory and processor speed limited robots’ ability to learn. Today, robots can upload what they learn into the cloud. Once there, other robots could access instructions for everything from cooking chicken cordon bleu to performing surgery. What one robot knows, every robot can know.

The second is deep learning, an advanced type of machine learning that allows robots to learn things that humans understand tacitly. Robots, for example, have trouble telling tables from chairs. Both may have the same number of legs, similar surface areas, and stand tall or short. Yet humans usually know where to sit and where to place their drinks.

Deep learning tries to overcome this problem with algorithms that sort through vast amounts of data and come to their own conclusions. Google, for example, used deep learning software to scour YouTube cat videos and come to its own conclusions about what defined a cat. It used this algorithm to identify cats twice as well as any other image recognition software. It took Google only 16,000 computers and 10 million videos to learn to do this.

Compared with even the youngest children, who know a cat when they see one, such results may seem pathetic. Yet Google learned enough from its experiment to improve its search engine, slash translation errors, and provide more relevant newsfeeds. Recently, a Google deep learning program, AlphaGo, soundly defeated the reigning European champion of the game Go, a feat that most AI specialists expected to take another decade.

Pratt imagines a world where robots and distributed sensors would send data to the cloud. Deep-learning AI would then analyze the data and use it to make robots and other types of AI software smarter. In this way the combination of cloud robotics and deep learning could yield rapid advances in machine intelligence, and displace many workers in a very short time.

This may already be happening. In 2011, Erik Brynjolfsson and Andrew McAfee of MIT’s Sloan School of Management warned in their book Racing against the Machine about technology’s potential for disruption.

Building on Arthur’s research, they noted that between 1947 and 2000, automation increased productivity, employment, and wages. Since 2000, however, U.S. productivity continued rising, but new job creation slowed and median income actually declined. They see smart machines at work. “It’s not so much a matter of jobs in general as specific types of skills being substituted for by new technologies. People with those skills see falling demand for their labor, so they will have lower incomes and less work unless they develop new skills,” McAfee said.

Typically, the easiest tasks to automate are routine and repetitive, such as classifying information, routing files, or operating a metal press. On the other hand, jobs for restaurant workers, janitors, and home health aides are growing. They require few skills, which keeps wages low, but they involve multiple tasks and human interaction that are difficult to automate.

The true winners in the new economy have specialized skills and often use computers to amplify their knowledge and capabilities. They are the ones, for example, who create software to book the lowest fares or prepare taxes. Their companies may make billions of dollars, but their websites and software put hundreds of thousands of people doing routine jobs out of work.

“Two hundred years ago the Industrial Revolution replaced people and animal power with machines,” Arthur said. “Now we are developing a neural system to go with it. It is a huge and unstoppable transformation.”

;custompagebreak;

The Technology Job Machine

Others are more optimistic. They believe that technology will spur employment, just as it always has.

In 2015, three economists from international management consultant Deloitte defended that view in a paper, “Jobs and People: The Great Job-Creating Machine,” which was short-listed for the Society of Business Economists’ top honor, the Rybczynski Prize. Machines “seem no closer to eliminating the need for human labor than at any time in the last 150 years,” the authors wrote.

“The problem is that, while it is easy to point to jobs lost due to technology, it is not as easy to identify jobs created by technology,” Alex Cole, one of the authors, said.

Cole and his fellow researchers found those jobs by analyzing labor data going back to 1871. Some of their findings were not surprising. Machines replaced muscle on farms and in factories, while employment grew among people who create, implement, and maintain technology.

Employment also rose for those with specialized knowledge. For example, Britain’s 1871 census recorded only 28,000 nurses. They held low-skilled positions closer to domestic service than medicine. With better training, the value of nurses increased and their numbers swelled to 300,000 in 2014.

Demand also surged for service and caring jobs like hairdressers and bar staff. Cole attributes this to technology.

Why? The ability of technology to raise productivity slashed the cost of many necessities, Cole said. In 1950, for example, food made up 35 percent of what the average Briton spent on essential goods and services. By 2014, food’s share had fallen to 11 percent. Meanwhile, the real cost of U.K. cars fell by half over the past 25 years, while the U.S. cost of televisions plummeted 98 percent since 1950.

Technology-driven price decreases give consumers more money to spend. Over the past 20 years, they increasingly spent it on health (think nurses), education (more teachers and teacher’s aides), and services that were once considered a luxury, Cole said. In 1871, for example, there was only one hairdresser per 1,800 people; today there are more than six. Because Britons can afford to go out more frequently, the number of bar staff has quadrupled since 1951. Other researchers have reached similar conclusions.

Nevertheless, these studies can be misleading: If smart machines are truly game changers, then past trends and historical data say little about the future. For that reason, leading MIT labor economist David Autor looks instead at the inherent limitations of smart machines in an aptly named 2015 paper, "Why Are There Still So Many Jobs?"

For example, why do robots install car windshields in factories, but humans repair them when they break, Autor asks. The answer is that robots require controlled environments, while humans, who are more flexible, can cope with unstructured tasks. That same adaptability is essential for medical technicians, plumbers, electricians, and many other middle-skill jobs.

Autor is also underwhelmed by machine learning, finding it cumbersome and often riddled with surprising errors. Humans leave computers scratching their digital heads when it comes to identifying cats or not sitting on tables. Even IBM’s Watson, which trounced the world’s best Jeopardy player, got one question spectacularly wrong, he noted. Asked which U.S. city named two airports for a military hero and a battle, it named a city in Canada—Toronto. No one wants a self-driving car to make a similar mistake.

Jobs on the Rebound?

Although they disagree about the long-term impacts, nearly all experts agree that smart machines are creating a very different job market. Autor and like-minded optimists expect pressures on middle skill jobs to eventually reverse because these jobs combine not only knowledge, but also adaptability, problem solving, common sense, and the ability to communicate with other people.

“Many of the tasks currently bundled into these jobs cannot readily be unbundled—with machines performing the middle-skill tasks and workers performing only a low-skill residual—without a substantial drop in quality,” Autor has written.

Businesses are already pairing human flexibility with mechanical precision. In 2012, for example, Amazon bought warehouse robot developer Kiva Systems. Kiva’s robots work alongside people, bringing shelves of parts to workers. The robots even use laser pointers to show workers which parts to pick, but only humans are flexible enough to rapidly manipulate and wrap parts for shipment

Arthur is not as sanguine. The growth of the second economy is great for people plugged into the system or working with robots. Meanwhile, those without the right skills are finding it harder to secure good, full-time jobs.

“It’s not just that we’ve lost jobs, but that the middle class has steadily lost hope that life would become better. In America, it was always taken for granted that the next generation would do better. Many people cannot take that for granted anymore,” Arthur said.

“It’s going to be a large social problem over the next 20 or 30 years,” Arthur said. “We’ll solve it, because as human beings we always solve our problems with new institutions and new types of arrangements. But until that day, it is not going to be easy.”

Alan S. Brown is associate editor at Mechanical Engineering magazine.

Today, we don’t have a sector that is growing fast enough to mop up those people who get laid off.