Will AI Make Jobs Obsolete?

Martin Ford’s recent book, Rise of the Robots: Technology and the Threat of a Jobless Future, explores the impact of automation, artificial intelligence, and smart machines on our economy and society.  Ford is the founder of a Silicon Valley-based software development firm, with twenty-five years of experience in computer design and software development. The book’s early chapters describe recent developments in information technology, with emphasis on the synergy between hardware improvement and algorithm design. Later chapters explore the implications for specific businesses, industries, and sectors of society.

Chapter 1 focuses on recent advances in computer vision and the shop floor revolution they portend. A firm called Industrial Perception, Inc. is currently marketing a robot capable of restacking an irregular heap of boxes of varying size, shape, color, and orientation. A human assigned such a task would immediately know where to begin. But robots, lacking applicable experience, must pause to formulate a plan. As a result, writes Ford, Industrial Perception’s machine appears sluggish and hesitant at times. However, the company estimates that upgrades already in the pipeline will permit the transfer of a box per second, far faster than the six seconds apiece required by human workers. The electric car company Tesla uses 160 surprisingly versatile robots to assemble some 400 cars per week in its Fremont, California factory. Each robot is able to exchange one tool for another before starting a new task. The same robot that installs the seats, for instance, is able to retool before applying adhesive and dropping the windshield into place.

Operative, vision-equipped robots, including those used by Tesla, currently see in just two dimensions. But modern technological developments indicate the possibility for more advanced vision. In November 2006, Nintendo introduced its Wii video game console, complete with a wireless wand enabling players to exert control by waving the wand to return an approaching tennis ball, for instance. Sony soon entered the market with a competing product, likewise reliant on two-dimensional vision. Microsoft, on the other hand, gambled with an add-on to the Xbox 360 game console called Kinect, which sees in three dimensions.

Using technology developed by a small Israeli company called PrimeSense, Kinect deploys a sort of sonar that operates at the speed of light. Bathing its field of vision in flashing infrared, Kinect measures the length of time required for each reflected flash to return to (roughly) its point of origin. Players need no magic wand to interact with Kinect; they need only move about in its field of vision. Remarkably, Kinect makes a sophisticated machine vision system—which might previously have cost tens or even hundreds of thousands of dollars—available in a lightweight consumer device priced at $150!

Researchers in robotics were quick to realize the potential for such technology to revolutionize their field. Within weeks of Kinect’s introduction, says Ford, a number of potential competitors had hacked into the system and posted videos on YouTube demonstrating the ability of their own retro-engineered devices to see in three dimensions.

Time alone will reveal the practical impact of acute machine vision, which includes the ability to see in three dimensions. But Parkdale Mills, a textile factory in Gaffney, South Carolina, may offer a glimpse into the future. Employing a mere 140 workers, Parkdale now ships quantities that would have required 2,000 employees as recently as 1980. The impact of machine vision on agriculture could be equally dramatic. Ford points out that experimental robots are already pruning grapevines in France using machine vision augmented by algorithms that decide which branches to eliminate. And a Japanese machine is able to recognize ripe strawberries by color and pick one every eight seconds, working continuously even after dark!

In Chapter 3, Ford cites a study by Martin Grötschel of the Zuse Institute Berlin. The study asserts that a certain complex production planning problem would have taken 82 years to solve in 1982, using what was then state-of-the-art hardware and software. As of 2003, the same problem could be solved in about a minute, an improvement of roughly 43 million! And since computer speed increased “only” a thousand-fold during that time, Ford writes that the remaining factor of 43,000 must be due to algorithm improvement. This suggests that improved algorithm design may pose an even greater threat to mankind’s long-term job prospects than do faster computers.

Chapter 4 reproduces a three-paragraph newspaper account of a postseason baseball game played in October 2009. Though not the most eloquent prose, it is eminently readable, grammatically correct, and an accurate description of the event in question. The account is remarkable only because the author is a computer program! The software responsible, called Stats Monkey, was created by a team of students and professionals at Northwestern University’s Intelligent Information Laboratory. Its purpose is to automate the process of sports reporting by transforming a stream of numerical data into a compelling narrative, complete with items of human interest.

To realize the commercial potential of their design, the creators of Stats Monkey founded a company called Narrative Science, Inc. They raised venture capital, hired a team of experienced software developers, and set out to construct a far more ambitious system called Quill. A variety of media outlets, including Forbes, now employ the system to compose articles on sports, business, and politics. Co-founder Kristian Hammond estimates that 90 percent of all news articles will be written algorithmically within fifteen years. In addition, Ford notes that one of the company’s earliest backers was In-Q-Tel, the venture capital arm of the Central Intelligence Agency. Presumably the company’s software is already being modified to transform the ever-growing deluge of raw intelligence data into easy-to-digest narrative reports.

The Narrative Science experience illustrates Ford’s claim that a wide variety of white-collar jobs face clear and present danger of automation. For further evidence, one might invoke Google’s introduction of an online language translation tool. The project development team began by assembling millions of pages previously translated into multiple languages, beginning with official documents prepared by the United Nations. Later on they exploited the web, where proprietary search engines located a host of additional pages suitable for analysis by the company’s machine-learning algorithms. In 2005, Google’s natural language translator was the clear winner of the National Institute of Standards and Technology’s annual machine translation contest, besting a host of rival systems constructed by professional linguists. Google’s system is not yet competitive with the product of skilled human translators, but it offers two-way translation between upwards of 500 language pairs, meaning that pretty much anyone can freely and all but instantaneously obtain a rough translation of just about any document in thirty or so different languages. While expensive human translators may never become extinct, call for their services seems certain to diminish.

Rise of the Robots is written, as the subtitle indicates, to warn of an impending shortage of jobs. Ford quotes a 2013 study by Carl Frey and Michael Osborne of Oxford University (UK), which found that occupations accounting for nearly half of total U.S. employment may be vulnerable to automation within the next 20 years. He also points to the emergence of companies like YouTube which, at the time of its $1.65 billion acquisition by Google, had been in business for less than two years and employed only sixty-five people. That’s $25 million per employee!

In Ford’s eyes, the prevalence of soaring per-employee valuations—which apply to a number of other high-tech start-ups as well as YouTube—constitute a direct challenge to the conventional economic wisdom whereby advanced technologies must, in the process of destroying “old-economy” jobs, automatically create “new-economy” substitutes requiring greater skill and offering better pay. He takes little comfort from the fact that things have often worked out that way in the past, since they have not done so lately and probably will not in the future. Indeed, Ford presents mounting evidence that “this time is different,” in that the next new economy promises to be of the jobless variety. Accordingly, he believes that nothing short of a radical reorganization of society—what political scientists would call a new “social contract”—will enable the ordinary citizen to earn a living wage.

The sort of reorganization Ford has in mind is outlined in the final chapter, entitled “Toward a New Economic Paradigm.” One need not agree with his conclusions or specific reform proposals to be impressed with the wealth of evidence he has marshaled in support of his central thesis. Rise of the Robots is well worth reading by anyone involved even peripherally in the information technology revolution, or curious to know where it seems to be leading. 

James Case writes from Baltimore, Maryland.