Despite what you read in the press, global warming isn’t all bad, and certainly not for everyone. There will be winners and losers, depending on where you live. In my case, it’s a tad too cool around here for my taste, but luckily for me, the average temperature where I live is projected to rise several degrees over the next few decades. Sounds good; hope I live to see it.
Global warming in and of itself isn’t a problem. After all, life on earth has survived numerous cycles of cooling and heating. The real problem with global warming is how quickly it happens. If there isn’t enough time for living things (including us) to adapt, rapid changes in climate, not to mention more volatile weather patterns, can sow havoc. The consequences of catastrophic climate change can reverberate for centuries as species suffer horrific losses of their habitat, leading to mass extinctions.
The impact of technological change on our labor markets works the same way. As long as change is gradual, the markets can respond. Too fast, and it’s chaos. And as with my particular environmental preferences, it creates winners and losers.
The likely accelerating effect of recent advances in artificial intelligence on technological change is going to roil our labor markets in two fundamental ways. The first is the simple truth that most automation replaces workers, so it eliminates jobs. That means fewer places for people to work. This threat is easy to see and measure— employers roll in a robot and walk a worker to the door. But sometimes change is less visible. Each new workstation may eliminate the need for one-fifth of a salesperson, or free Skype calls may allow you to work more productively at home one day a week, deferring the need for that new hire until next quarter.
If this happens slowly, the resulting improvements in productivity and reduced cost eventually create wealth, stimulating job growth that compensates for the losses. The growth may be directly in the newly improved enterprise, as lower prices and better quality increase sales, creating a need to hire more workers. Or it may be in distant parts of the economy where the customers who no longer need to pay as much for some product or service decide to spend the money they saved. If new drilling technologies cause natural gas prices to drop, there’s more left over from your paycheck to save for that sailboat you’ve got your eye on.
But the second threat is much more subtle and difficult to predict. Many technological advances change the rules of the game by permitting businesses to reorganize and reengineer the way they operate. These organizational and process improvements often make obsolete not only jobs but skills. A teller may get laid off when a bank installs ATMs; the improved service creates a need to hire network engineers but not tellers. Even if the bank ultimately expands its total workforce, the tellers remain out of luck. Weavers can eventually learn to operate looms, gardeners to service lawnmowers, and doctors to use computers to select the right antibiotics—once they accept that synthetic intellects are superior to their own professional judgment. But learning the new skills doesn’t happen overnight, and sometimes the redundant workers simply aren’t capable of adapting—that will have to wait for a new generation of workers.
For an example of labor market transformation that we have weathered successfully, consider agriculture. As recently as the early 1800s, farms employed a remarkable 80 percent of U.S. workers.1Consider what this means. Producing food was by far the dominant thing people did for a living, and no doubt this pattern had been typical since the invention of agriculture about five thousand years ago.
But by 1900, that figure had dropped in half, to 40 percent, and today it’s only 1.5 percent, including unpaid family and undocumented workers.2 Basically, we managed to automate nearly everyone out of a job, but instead of causing widespread unemployment, we freed people up for a host of other productive and wealth-producing activities. So over the last two centuries the U.S. economy was able to absorb on average about 1/2 percent loss of agricultural job opportunities each year without any obvious dislocations.
Now imagine that this had happened in two decades instead of two centuries. Your father worked on a farm, and his father before him, as far back as anyone could remember. Then a Henry Ford of farming revolutionized the entire industry in what seemed like a flash. The ground shook with the rumble of shiny new plows, threshers, and harvesters; the air was thick with the smell of diesel. Food prices plummeted, and corporations bought up farmland everywhere with the backing of deep-pocketed Wall Street financiers. Within a few years, your family’s farm was lost to foreclosure, along with every possession except the family Bible.
You and your five brothers and sisters, with an average third-grade education, found your skills of shoeing horses, plowing straight furrows, and baling hay utterly useless, as did all of your neighbors. But you still had to eat. You knew someone who knew someone who operated one of the new machines twelve hours a day in return for three squares, who supposedly got the job in Topeka, so you moved to one of the vast tent cities ringing the major Midwestern cities in the hope of finding work—any kind of work. Before long, you got word that your parents sold the Bible to buy medicine for your youngest sister, but she died of dysentery anyway. Eventually you lost track of the rest of your other siblings.
The 1 percent who still had jobs lived in tiny tract houses and barely got by, but they were nonetheless the envy of the rest—at least they had a solid roof over their heads. Each day, you waited in line outside their gated communities hoping for a chance to wash their clothes or deliver their bag lunches. Rumors spread that the daughters of the storied entrepreneur who changed the world had used his vast fortune to build a fabulous art museum made of crystal in a small town in Arkansas. But all this was before the revolution. After that, things got really bad.
I’m going to argue that a similarly tectonic shift looms ahead, though doubtlessly less dramatic and more humane. Forged laborers will displace the need for most skilled labor; synthetic intellects will largely supplant the skilled trades of the educated. When initially deployed, many new technologies will substitute directly for workers, getting the job done pretty much the same way. But other innovations will not only idle the workers; they will eliminate the types of jobs that they perform.
For example, consider the way Amazon constantly adapts the stock patterns in its warehouses. If a person were to do the warehouse planning (as in many more traditional companies), products might be organized in a logical and comprehensible way—identical items would be stored next to each other, for example, so when you needed to pick one, you knew where it was. But a synthetic intellect of the sort Amazon has built isn’t subject to this constraint. Like items can be located next to others that are frequently shipped with them, or on any shelf where they fit more compactly. To the human eye, it looks like chaos—products of different sizes and shapes are stacked randomly everywhere—which is why this type of warehouse organization is known as chaotic storage.3 But a synthetic intellect can keep track of everything and direct a worker to exactly the right place to fulfill an order far more efficiently than a human organizer could.
A side effect of introducing this innovation is that it reduces the training and knowledge required of warehouse workers, making them more susceptible to replacement by forged laborers. These employees no longer have to be familiar with the location of products on the shelves; indeed, it would be near impossible to do so in such a haphazard and evolving environment. Having first simplified the skills required to get the job done, Amazon can now replace the workers that roam the warehouse floor picking those orders. This is likely why the company bought the robotics company Kiva Systems, reportedly for $775 million, in 2012.4
This is a single example of a profound shift that synthetic intellects will cause in our world. The need to impose order—not only for warehouses but for just about everything—is driven by the limitations of the human mind. Synthetic intellects suffer no such constraint, and their impact will turn tidiness to turmoil in many aspects of our lives. Our efforts to tame our intellectual and physical domains into manicured gardens will give way to tangled thickets, impenetrable by us.
When most people think about automation, they usually have in mind only the simple replacement of labor or improving workers’ speed or productivity, not the more extensive disruption caused by process reengineering. That’s why some jobs that you might least expect to succumb to automation may nonetheless disappear.
For instance, studies often cite jobs that require good people skills or powers of persuasion as examples of ones unlikely to be automated in the foreseeable future. But this isn’t necessarily the case. (As I noted in chapter 4, the CEO of Rocket Fuel observed that persuasion was a skill that his company’s ad placement service largely replaces.)
The ability to convince you that you look terrific in a particular outfit is certainly the hallmark of a successful salesperson. But why do you need that person when you can ask hundreds of real people? Imagine a clothing store where you are photographed in several different outfits, and the images are immediately (and anonymously, by obscuring your face) posted to a special website where visitors can offer their opinion as to which one makes you look slimmer. Within seconds, you get objective, statistically reliable feedback from impartial strangers, who earn points if you complete a purchase. (This concept is called “crowdsourcing.”) Why put your faith in a salesperson motivated by commission when you can find out for sure?
Reflecting these two different effects of automation on labor (replacing workers and rendering skills obsolete), economists have two different names for the resulting unemployment. The first is “cyclical,” meaning that people are cycling in and out of jobs.5 In bad times, the pool of people who are between jobs may grow, leading to higher unemployment. But historically, as soon as the economy picks up, the idled workers find new jobs. Fewer people are unemployed and for shorter periods of time. This works just like the housing market: in a slow market, there are more houses available and the ones that are take longer to sell. But when the market turns around this excess inventory is quickly absorbed.
I was surprised to learn just how much turnover there is in the U.S. labor market. In 2013, a fairly typical year, 40 percent of workers changed jobs.6 That’s a very fluid market. By contrast, less than 4 percent of homes are sold each year.7 So when we talk about 8 percent unemployment, it doesn’t take long for small changes in the rates of job creation and destruction to soak that up, or conversely to spill more people out of work.
The other type of unemployment is called “structural,” which means that some group of unemployed simply can’t find suitable employment at all. They can send out résumés all day long, but no one wants to hire them, because their skills are a poor match for the available jobs.8 The equivalent in the housing market would be if the types of houses available weren’t suitable for the available buyers. Suddenly couples start having triplets instead of single kids and so need more bedrooms, or people start commuting to work in flying cars that can take off only from flat rooftops, while most houses have pitched roofs.
As you can see from my fanciful examples, the factors that change the desirability of housing don’t usually change very fast, so builders and remodelers have plenty of time to adapt. But this isn’t true for automation because the pace of invention and the rate of adoption can change quickly and unpredictably, shifting the character of whole labor market segments far more rapidly than people can learn new skills—if they can be retrained at all. We’re buffeted about by these fickle winds precisely because they are hard to anticipate and virtually impossible to measure.
Economists and academics who study labor markets have a natural bias toward the quantifiable. This is understandable, because to credibly sound the alarm, they must have the hard data to back it up. Their work must stand up to objective, independent peer review, which basically means it must be reduced to numbers. But as I learned in business, spreadsheets and financial statements can capture only certain things, while trends that resist reduction to measurement often dominate the outcome. (Indeed, there’s an argument to be made that the troublesome and unpredictable business cycles plaguing our economy are largely driven by the fact that returns are easily quantified, but risks are not.) I can’t count the number of meticulously detailed yet bogus sales projections I’ve seen bamboozle management teams. At work I sometimes felt my most important contribution as a manager was anticipating that which had yet to manifest itself in quantifiable form.
But talking about the overall labor market, unemployment statistics, or the aggregate rate of change obscures the reality of the situation because the landscape of useful skills shifts erratically. The complexity of this web of disappearing labor habitats and evolving job ecosystems resists analysis by traditional mathematical tools, which is why attempts to quantify this whole process tend to bog down in reams of charts and tables or devolve into hand-waving.
Luckily I’m not bound by these same professional constraints, so fasten your seat belt for a quick tour of the future. My approach will be to look at some specific examples, then attempt to reason by analogy to get a broader picture. Let’s start with retail—the largest commercial job market, as determined by the U.S. Bureau of Labor Statistics (BLS).9
The BLS reports that about 10 percent of all U.S. workers are employed in retailing, or approximately 14.5 million people.10 To analyze trends, let’s use salespersons as a proxy for the whole group. The BLS projects that this labor force, which stood at 4.4 million in 2012, will grow by 10 percent to 4.9 million over the next ten years. But this is based on current demographic trends, not a qualitative analysis of what’s actually going on in the industry.
To get a sense of what’s really going to happen, consider the effect on employment of the transition from bricks-and-mortar stores to online retailers. A useful way to analyze this is to use a statistic called revenue per employee. You take the total annual revenue of a company and divide it by the number of employees. It’s a standard measure of how efficient a company is, or at least how labor-efficient.
Average revenue per employee for Amazon (the largest online retailer) over the past five years is around $855,000.11 Compare that to Walmart (the largest bricks-and-mortar retailer), whose revenue per employee is around $213,000—one of the highest of any retailer. This means that for each $1 million in sales, Walmart employs about five people. But for the same amount of sales, Amazon employs slightly more than one person. So for every $1 million in sales that shift from Walmart to Amazon, four jobs are potentially lost.
Now, both companies sell pretty much the same stuff. And Walmart does a good portion of its sales online as well, so the job loss implied by the shift to online sales is understated. And neither company is standing still; both are likely to grow more efficient in the future.
To establish an upper bound on job losses, imagine that magically all retail sales were to suddenly shift from Walmart-like stores to Amazon-like websites. The 10 percent of the labor force (mostly) working in stores would be replaced by 2 percent of the labor force working for online retailers. That’s 8 percent fewer jobs available in the United States, more than the entire 2014 unemployment rate. So are we in big trouble here? Not really. Of course, all sales aren’t going to shift online—your favorite mall isn’t going to close down—and certainly the shift is going to take some time. But how long?
Despite all the hoopla, only 6 percent of U.S. retail sales are currently online. These have been growing consistently at a rate of about 15 percent annually for the past four years.12 If online sales were to continue to grow at that pace for the next twenty years (unlikely), and if all growth in retail sales went to the online segment (also unlikely), they would then account for at most half of all retail sales. That means that retail sales would have roughly doubled over that period, which is pretty much what they did over the previous two decades, but only 10 percent more people would be required to support these sales.13 And that assumes that bricks-and-mortar stores don’t grow at all, which is not plausible.
Meanwhile, what’s going to happen to the labor force? Based on careful demographic projections, the BLS estimates that the labor force will grow only about 12 percent over the next twenty years.14 In other words, a tremendous shift from bricks-and-mortar stores to far more labor-efficient online retailers will likely result in only a 2 percent negative impact on employment over that period. (That is, the 12 percent labor market growth only slightly exceeds the 10 percent more retail workers required.) That’s only .1 percent per year for the economy to absorb, compared to the .5 percent average annual loss of agricultural jobs over the last two centuries. But the story gets better. Surely with a doubling of retail sales, new jobs in all sorts of industries that design, manufacture, and ship these products will more than take up that slack.
Oops, did I include shipping on that list? My mistake. Shipping is a completely different story. In 2012, there were 1.7 million long-haul truck drivers in the United States. These are the people who operate the tractor-trailers and other large cargo-carrying vehicles that frequent the interstate highway system. The BLS projects that the demand for these drivers will grow 11 percent over the next ten years. No way.
While you may regard highway driving as requiring greater skill and more experience than navigating local streets, exactly the opposite is true when it comes to autonomous driving technology, a wonderful hybrid of synthetic intellects and forged laborers. Highways are well maintained, contain fewer random moving obstructions (such as pedestrians and bicycles), and are far more predictable than your local neighborhood streets. The technology to operate self-driving trucks is available today and can be retrofitted to existing fleets at very reasonable costs. Trucks outfitted with such technology can “see” in all directions instead of mostly just straight ahead, drive in complete darkness or blackout conditions, and instantly share information about road conditions, nearby risks, and their own intentions. (Basically, they can rely on detailed 3D radar, called Lidar, in conjunction with detailed maps and GPS, and so have no need for headlights.)
What’s more, their reaction time is close to zero. As a result, self-driving trucks can safely caravan with only inches of space between them (called “platooning” in the literature), reducing road congestion and resulting in 15 percent or more fuel savings.15 Delivery is quicker because they can operate around the clock without rest stops. They don’t get tired, drunk, sick, distracted, or bored; they don’t doze off, talk on the phone, or go on strike for better wages and working conditions. And how many of the 273,000 large-truck accidents taking 3,800 lives and costing over $4.4 billion (in 2011 alone) could be avoided in the future?16 May I point out that this single innovation could save more lives annually than were lost in the September 11th World Trade Center disaster?
Such systems aren’t futuristic pipe dreams; they are already being tested on real highways and other venues. To quote from a recent press release: “Rio Tinto is rolling out a fleet of 150 automated trucks at our Pilbara iron ore operations, the world’s first major deployment of an autonomous truck fleet. Since the two-year trial began, the autonomous trucks have operated every day, 24 hours a day, and have moved more than 42 million tonnes of material in approximately 145,000 cycles. They have travelled more than 450,000 kilometres. We control the trucks from our Operations Centre in Perth, 1,500 kilometres away. The trucks follow pre-defined courses, and GPS systems navigate autonomously from loading units to dump locations.”17
You don’t have to be much of a futurist to see what’s coming. Nearly 2 million long-haul truck drivers in ten years? I suspect that the BLS is way off base on this one—more likely closer to zero. But that’s only one of the applications for autonomous driving. How many of the more than 5.7 million licensed U.S. commercial drivers (2012) will lose their jobs as a result of variations of this technology?18 I wouldn’t recommend this career option to my kids.
So just based on the aggregate numbers, drivers are probably going to be losing their jobs in droves, but not retail employees. However, there’s a twist—the raw numbers obscure a deeper truth. The real issue is not just the overall number of jobs available but the skills required to perform them.
Here’s where things get qualitative, so permit me to paint some pictures. There’s a big difference between the skills required to sell people things in stores and (for example) those required to maintain an online retailing website. It’s not that easy for some kindly grandmother to go from pointing out the location of the shoe department at Walmart to monitoring product reviews at Amazon. A truck driver who may or may not have completed high school and whose main familiarity with computers is watching Netflix may not be well suited for many other jobs, particularly since a wide array of other blue-collar professions are likely to be succumbing to automation as well. Robotic devices that can see and operate in natural environments are about to decimate all sorts of labor markets. Forged laborers, in short, are approaching from all directions. I’ll describe a few here.
Agricultural workers. Projects are under way that threaten the livelihoods of the remaining 2 to 3 million U.S. farmworkers.19 In 2010 the European Union started funding the Clever Robots for Crops program (cleverly abbreviated as CROPS). As the project leader explains, “An agricultural robot must be equipped with intelligence so as to be able to robustly operate in the unstructured, dynamic and hostile agricultural environment.”20
Agrobot, a Spanish company opening an office in Oxnard, California, makes a commercial robot that harvests strawberries.21 It identifies only the fruit ripe enough for picking. The good news is they’re hiring, but only if you’ve got an engineering degree. I doubt that’s much comfort to Elvia Lopez, a kindly thirty-one-year-old Mexican immigrant who picks strawberries in Santa Maria, California (who was profiled in the Los Angeles Times).22 Agrobot isn’t alone in tackling this opportunity. A Japanese competitor claims that its technology can reduce strawberry picking time by 40 percent.23
Blue River Technologies, a Silicon Valley venture-funded startup headed by a Stanford graduate, is developing robots that can weed. To quote from their marketing materials: “We are creating systems that can distinguish crops from weeds in order to kill the weeds without harming the crops or the environment. Our systems use cameras, computer vision, and machine learning algorithms.”24
Note that the coming army of mechanical farmworkers doesn’t have to be faster than the workers they replace because, like autonomous vehicles, they can work in the dark and so aren’t limited to operating in daylight.
Warehouse workers. Beyond the picking and packing of orders, as I’ve described above, there’s the loading and unloading of packages. This is done by human workers now because it takes human judgment to decide how to grasp and stack randomly shaped boxes in delivery vehicles and shipping containers. But another Silicon Valley startup, Industrial Perception, Inc., is changing all that. Its robots can peer into a truck, select an item, then pick it up. As it quipped on its website (before the site went dark after the company was acquired by Google in 2013), Industrial Perception is “providing robots with the skills they’ll need to succeed in the economy of tomorrow.”25
Sex workers. You’d think prostitution might be a job requiring a human touch. It may be illegal in most of the United States, but sex toys aren’t. And they are about to take an entirely new form. Companies like New Jersey–based TrueCompanion are developing full-sized interactive sex dolls in both female and male versions (named Roxxxy and Rocky).26 As the company founder, Douglas Hines, who previously worked in AI at Bell Labs, said in an interview in 2010, “Artificial intelligence is the underpinning of the whole project.” According to the company, “Roxxxy can carry on a discussion and expresses her love to you. She can talk, listen, and feel your touch.”27
Other projects cooking in AI labs around the world are almost too numerous to mention. They are aimed at tasks like folding laundry, rinsing dishes, then loading them in a dishwasher, bagging groceries, and fetching coffee (one robot even navigates the elevators).28
My examples so far may seem to offer some comfort to readers employed in more cerebral endeavors, but this relief would be misplaced. The coming wave of synthetic intellects is going to devastate many of their professions just as surely as forged laborers are going to replace manual laborers. Automation is blind to the color of your collar.
Let’s start with the practice of law. The American Bar Association estimates that there were 1.2 million licensed attorneys in the United States in 2010, roughly three-quarters of them in private practice.29
There’s been much hand-wringing over the challenging economics of getting a professional law degree. It used to be that getting into law school was a great accomplishment, not to mention a ticket on the partner track to the good life. But no more. Applications have been falling year after year as a more practical generation is waking up to economic reality. The Law School Admissions Council reports that applications in 2014 were down nearly 30 percent over just the previous two years, returning to levels last seen in 1977.30 New graduates can be saddled with debt of more than $150,000, while the average graduate’s starting salary in 2011 was only $60,000, down nearly 17 percent from just two years earlier.31 But they were the lucky ones. In 2009, an astounding 35 percent of newly minted law school graduates failed to find work that required them to pass the bar exam.32
There are, of course, many factors affecting job opportunities for attorneys, but automation is certainly among them. And the problems are just getting started. To date, the use of computers in the legal profession has been largely focused on the storage and management of legal documents. This reduces billable hours because you don’t have to start from scratch when drafting contracts and briefs. But a new crop of legal-tech entrepreneurs is working to greatly reduce or eliminate the need for lawyers for the most common transactions. In specialty after specialty, innovators are finding that most productive work is sufficiently rote to permit delegation to synthetic intellects. Common commercial contracts, from leases to loans to licenses to incorporation papers to purchase agreements, are well structured enough to allow a first draft, if not a final one, to be written by a computer program.
Consider the legal-tech startup FairDocument.33 By focusing on estate planning, a well-defined and fairly routine area of law, the company is able to “interview” clients on its website and prepare initial draft documents. Potential clients answer some initial questions, then attorneys bid to get their business. Most of the time, if the case is relatively straightforward, attorneys opt for the standard recommended bid of $995 for an estate plan prepared through Fair-Document, for a service that might otherwise typically cost $3,500 to $5,000.
You might think this simply reduces the lawyer’s pay, but attorneys still come out ahead because of what happens next. Instead of conducting the usual phone or in-person interview to educate the new client and collect the needed information, then spending several hours drafting documents, the attorneys let FairDocument walk the client through a lengthy, structured online consultation, explaining the required concepts and collecting the client’s particulars. The software then delivers an initial draft to the lawyer, calling out areas that are likely to require his or her additional judgment or attention. Jason Brewster, the company’s CEO, estimates that FairDocument reduces the time required to complete a straightforward estate plan from several hours to as little as fifteen to thirty minutes, not to mention that his company is doing the prospecting for new clients and delivering them to the attorneys.
A more sophisticated example of synthetic intellects encroaching on legal expertise is the startup Judicata.34 The company uses machine learning and natural language processing techniques to convert ordinary text—such as legal principles or specific cases— into structured information that can be used for finding relevant case law. For instance, it could find all cases in which a male Hispanic gay employee successfully sued for wrongful termination by reading the actual text of court decisions, saving countless hours in a law library or using a more traditional electronic search tool.
Other startups are tackling the time-consuming process of early case assessment, discovery processing, document review, document production, and internal investigations.35 Some do actual legal research and provide advice on case strategy, answering questions like, “How often has a judge ruled in favor of defendants on motions to transfer or motions for summary judgment?” and “What mistakes have tripped up others around similar IP [intellectual property]?”36
Some firms are even considering moving the machines from the back room to the nameplate. Consider the cleverly named law firm of Robot, Robot, and Hwang. Yes it’s a joke, but the firm is real. The junior partner, Tim Hwang, has an undergraduate degree from Harvard and a J.D. (law degree) from the University of California at Berkeley. To quote from its website, the firm “attempts to marshal a universe of thinking from the world of technology, startup, and computational science to bear on the often staid and conservative world of legal practice.”37
Despite efforts by the legal profession to protect its members’ livelihoods, an increasing number of startups are bypassing restraints on how and by whom law can be practiced by offering what amounts to automated legal advice presented in different forms over the Internet. For instance, they may employ a small staff of attorneys to “review” documents for correctness before they are released to a client. But most of these startups offer a different format, whereby individual practitioners can be introduced to clients, establish a working (and billing) relationship, then perform their duties with extensive automated support provided by the company.
By offering attorneys the option to work at home, thus avoiding the expense of an office, and by reducing costs through the substitution of sophisticated computer systems for skilled paralegals, these virtual law offices provide an attractive option for practitioners seeking more independence and control over their work. Needless to say, this is an excellent opportunity for graduates who are unable to land an entry-level position at a traditional firm, but it’s also attractive to experienced partners who are tired of office politics or of handing over a large proportion of their billings to their firms. These trends are driving down the cost of high-quality legal assistance while improving access for millions of potential clients.
Law schools aren’t standing still, though. For instance, a recent course, Legal Informatics, at Stanford University is cotaught by law school and computer science department faculties. The course description says, in part, “What role will lawyers play when customized advice is dispensed over the Internet as easily as cappuccino from a vending machine? Register for a preview of what your job will be like in five years.”
If it’s less attractive to become a lawyer, how about a doctor? The days of the “country doc” are long gone, but information technology is also transforming the character of medical practitioners in surprising ways.
The main shift is a growing recognition that the medical arts are not arts at all but a science that is better driven by statistics and data than intuition and judgment. In bygone eras, it was at least plausible that someone could absorb a reasonable proportion of the world’s medical knowledge and apply it to cases as they are presented. But over the past half century or so, as it became clear that the avalanche of research, clinical trials, and increased understanding of how our bodies (and minds) work was beyond the comprehension of a single individual, the field fractured into a myriad of specialties and practices. Today, your “primary care physician” is more of a travel agent to the land of specialists than a caregiver, except for the simplest of ailments.
But the hidden costs of this divide-and-conquer approach to medical care are about to become painstakingly clear. Coordinating the activities of multiple practitioners into a coherent plan of action is becoming increasingly difficult, for two reasons. First, no one has the complete picture, and, even if they do, they often lack the detailed knowledge required to formulate the best plan of action. Second, specialists tend to treat the specific conditions or body parts that they are trained for, with inadequate regard for the side effects or interactions with other treatments the patient may be receiving. For me, the practice of medicine today conjures the image of a Hieronymus Bosch painting, with tiny, pitchfork-wielding devils inflicting their own unique forms of pain.
As a patient, you would ideally prefer to be treated by a superdoc who is expert in all the specialties and is up to date on all of the latest medical information and best practices. But of course no such human exists.
Enter IBM’s Watson program. Fresh off its Jeopardy! victory over champions Brad Rutter and Ken Jennings, Watson was immediately redeployed to tackle this new challenge. In 2011, IBM and WellPoint, the nation’s largest healthcare benefits manager, entered into a collaboration to apply Watson technology to help improve patient care. The announcement says, “Watson can sift through an equivalent of about one million books or roughly 200 million pages of data, and analyze this information and provide precise responses in less than three seconds. Using this extraordinary capability WellPoint is expected to enable Watson to allow physicians to easily coordinate medical data programmed into Watson with specified patient factors, to help identify the most likely diagnosis and treatment options in complex cases. Watson is expected to serve as a powerful tool in the physician’s decision making process.”38 As with its original foray into AI fifty years ago, IBM is still cautious not to ruffle the feathers of the people whose rice bowls they are breaking, but one person’s decision process support tool is another’s ticket to the unemployment line.
No one likes the idea that his or her field is simply too big and fast moving to master. And doctors in particular aren’t likely to graciously concede control of their patients’ treatment to synthetic intellects. But eventually, when outcomes demonstrate that this is the better option, patients will demand to see the attentive robot, not the overworked doctor, for a fraction of the fee, just as many people would now rather have an ATM than a human teller count out their cash.
Doctors and lawyers are not the only professionals with much to worry about—there are plenty of others. For instance, today’s highly trained commercial pilots should rarely fly planes, particularly if they have the best interests of their passengers in mind. Pilot error is by far the leading cause of fatal crashes, hovering around 50 percent for the past fifty years despite dramatic improvements in air safety.39 By comparison, mechanical failures are blamed in only 20 percent of events. Planes’ autopilots are now so sophisticated that pilots are required to use them in certain conditions rather than flying the planes themselves.40 You may feel some comfort knowing that a trained pilot is standing by in the cockpit in case of a problem, but not if he’s feeling blue and decides to crash the plane, as has happened in at least three documented commercial flight accidents in the last twenty-five years that killed everyone on board.41
It should be obvious that technologies are capable of replacing teachers and professors in a wide variety of settings. The current buzzword for this is the flipped classroom—students watch lectures and learn the material online at home, then do their homework at school with the help of teachers and teaching assistants. Teachers may no longer need to prepare or deliver lectures, reducing them to what could be called “learning coaches.” The diminished skill set required is sure to transform the profession and create yet more challenges for our already beleaguered teachers.
But these are anecdotal examples. Just how many professions will be newly subject to automation in the next decade or so? It’s a tough question to answer, but a group of researchers at Oxford University has bravely taken a quantitative approach, matching near-term technologies to job skills required for 702 occupations detailed by the BLS. The researchers’ remarkably detailed and insightful analysis reports that 47 percent of total U.S. employment is at high risk of significant automation across a wide variety of blue-collar and white-collar professions. They specifically call out “advances in the fields of ML [machine learning], including Data Mining, Machine Vision, Computational Statistics and other sub-fields of Artificial Intelligence, as well as MR [mobile robotics]” as key drivers of these trends.42 So an incredible half the workforce, give or take, is in danger of replacement by a machine in the near future.
What’s to be done with all these surplus workers with obsolete skills? We need to teach old dogs new tricks—but not just any tricks, tricks that employers will pay them to perform. And the only people who know for sure what tricks these are, are the employers themselves.
With respect to professional training, we are making two mistakes. The first is relying mainly on traditional schools to decide what they should teach students. Our accredited educational institutions are not known for their responsiveness to economic trends, and since the administrators developing the curricula are not customarily out in the field keeping up to date on what novel skills will be most valuable in the economy, they couldn’t do it if they wanted to. It’s a mystery to me why my kids had to learn penmanship, calculus, and French in high school rather than more practical skills like typing, statistical estimation, and Chinese. (Reading and writing make sense, though.)
Of course, not all educational decisions should be driven by employment prospects. Learning and training are not the same thing. There’s plenty of value in raising well-rounded, historically aware, articulate, thoughtful citizens. But beyond a core of basics—which in my opinion doesn’t extend to memorizing rows of the periodic table or performing partial differentials—the subject matter should be aimed at equipping students with useful and marketable skills. We should focus on vocational training, not vacational training.
The second mistake is the tacit assumption that first you go to school, and when you are done, you go get a job. This made sense when jobs and skills changed on a generational timescale, but it does not in today’s fast-moving labor markets. These two phases of life need to be strongly interleaved, or at least the opportunity for new skill acquisition must be explicit and omnipresent.
The path to addressing both of these issues is through enlightened economic policy. The obvious question regarding the retraining of workers with obsolete skills is who’s going to pay. The equally obvious answer is the ones who stand to benefit the most: the workers themselves. But how can down-on-their-luck unemployed workers find and pay for training that matches their abilities and is of real value to employers?
Just as we have special types of loans that are intended to encourage and support homeownership, we need to develop a system of vocational training loans that bear a similar relationship to the targeted asset—in this case, employment rather than houses. When you get a mortgage, the government or the bank that originates it doesn’t pay off the loan—you do. If a problem arises, such as your house burns down or you simply can’t make the payments, you can walk away and lose only your down payment, because most mortgages are “‘nonrecourse” loans, meaning that the only security provided to the lender in the event of default is the property itself. Now, abandoning a house because you can’t (or won’t) make the monthly payments is certainly painful. You, in addition to the lender, have every incentive to be sure you are a solid credit risk. Both parties also have an incentive to ensure that the property is worth the price being paid (or worth, at least, enough to cover the loan with some margin of safety). This is why lenders require an appraisal supporting the current market value of the house before funding a mortgage.
And similar principles can apply to a vocational training loan. For simplicity, let’s call this a job mortgage. Here’s at least one way this might work, though there are many variations that could be equally or more effective. To get a job mortgage, you have to secure the sponsorship of a potential employer—perhaps the one you are already working for—just as you apply for a mortgage on a particular property. But in this case, the employer isn’t promising to hire you, and you aren’t promising to take that particular job, though there’s a reasonable expectation that if all goes well this is likely to happen. In effect, you are applying for a future job, and the employer is issuing a good-faith letter of intent that it has (or will have) a real need to hire someone like you for the position within some reasonable time period.
Because the employer, who is presumably having trouble finding workers with appropriate skills, can issue only as many of these sponsorships as it has jobs available, there is a natural limit on the number of people who are able to secure job mortgages. Employers who fulfill their promises can get a tax break (say, relief on payroll taxes for the position for the first six months), which incents them to participate in the program. On the flip side, penalties can be assessed on employers who issue these letters of intent willy-nilly and don’t ultimately follow through, within some statistical bounds. A simple way to accomplish this is to require employers to post a modest “bond,” which is released only when the position is filled. Employers would also be required to certify that a particular course of training—possibly even one that they themselves provide—will target the needed skills.
Enrollments at the training institutions will be naturally sized to the available pool of jobs in the market, because they are largely reliant on these loans to fund their programs. They are also incented to stay keenly focused on the relevant skills; otherwise the employer(s) will not approve the training as satisfactory for their needs. As a consequence, there is no need for formal government accreditation of these programs—the system is, in effect, self-regulating.
For the potential employee, the key is that the loan is paid back only out of earned income—it is secured by future paychecks. Payments would be limited to some percentage of earnings, say 25 percent of net pay, in a similar spirit to the way mortgage lenders enforce a payment-to-income ratio for their loans. And there needs to be certain built-in relief in case of problems. For example, monthly payments are capped or deferred if the net take-home pay falls below 150 percent of the poverty level established by the government. Because the loan is paid solely from earned income, payments are effectively suspended (though interest may accrue) if the worker is unemployed for any reason, and the loan is automatically reamortized.
What if the training isn’t successful, the job isn’t available (and no other suitable employment is available either), or the trainee simply decides not to work? As in the case of a home mortgage, the trainee is still responsible for repaying a portion of the loan, say, 20 percent (a typical down payment on a house), regardless of earned income, after a grace period to account for unemployment. That’s the way mortgage risk is managed today, and it works just fine.
There are many details that need to be fleshed out, but the basic idea is to create a new type of financial instrument—the job mortgage—through regulations and policies to substitute for or supplement the largely broken current system of student loans, which in many cases saddles innocent victims with debts they can ill afford for inadequate training at for-profit colleges created for the primary purpose of collecting government loan money, with little accountability for the results. Most existing efforts to address this problem have focused on cajoling colleges to do a better job through ratings and other inducements, though recently the government has taken a stronger stand on abusive practices by for-profit institutions, requiring certain graduation rates and employment prospects.43 But by creating the proper economic incentives for employers, lenders, and trainers through appropriate public policies, we can render the process of skill acquisition and retraining both practical and humane, not to mention much more effective than it is today.
The concept of a job mortgage is a modernized free-market version of the historical apprenticeship or internship model. The main advantage is that it partially decouples the training from the specific employer or position, to great advantage for both the employer and the worker. Instead of the implicit indentured servitude of low-paid trainee positions, which effectively forces companies to operate their own mini-education business and workers to remain at unwanted or inappropriate jobs, people can apply their newly acquired skills where they are most highly valued while employers can draw from an expanded pool of highly skilled workers. This is nothing more than the lubricating role that money is supposed to play in the economy. It’s a mystery to me why we seem to treat occupational skills differently than other assets, like some sort of medieval barter system, at great cost to society. If major-league sports figures can securitize their future earnings, why can’t the average person do the same?44
My specific concept of a job mortgage might be new, but the basic approach certainly isn’t. Milton Friedman, leader of the Chicago school of economics, wrote an essay entitled “The Role of Government in Education” in 1955, in which he draws a distinction between “general education for citizenship” and “vocational or professional education.” He recommends that the latter should be subject to analysis as an investment, similar to physical assets, and that government policies be put in place to facilitate investment (as opposed to subsidies) in such training. As he put it, “The individual would agree in return to pay to the government in each future year x percent of his earnings in excess of y dollars for each $1,000 that he gets in this way… . An alternative, and a highly desirable one if it is feasible, is to stimulate private arrangements directed toward the same end.”45
And indeed, today there are some tentative private-sector steps in that direction.46 For example, Chicago-based Education Equity, Inc., is making income-linked loans to students entering certain approved programs, albeit on a small scale so far.47
With this perspective, let’s return briefly to the challenges my former employee Emmie Nastor has encountered. In my view, he was utterly failed by our educational system. His B.S. in business administration would seem to be of little practical value, at least in terms of providing the skills that employers in the local job market are actually prepared to pay for. Reportedly, less than half of San Francisco State University’s students graduate within six years, and less than half of those secure full-time employment within six months of graduating.48 (I can find no official statistics on this subject published by the school.) Nonetheless, the university has little incentive other than goodwill to monitor this, much less fix it, as long as students are willing to attend and pay (or take out loans to pay). If to fill its classrooms the university had to meet the collective expectations of local employers, who have every motivation and desire to see the school turn out qualified candidates, it stands to reason that the system would quickly come into equilibrium.
Global warming is a bear, but we aren’t bears. Most animals don’t have the native intelligence to think their way out of habitat changes, but we do. The accelerating evolution of our labor ecosystem, propelled by continual technological advances, compels us to take a fresh look at the way we prepare our young and even ourselves for productive and fruitful lives.
Excess workers and obsolete skills are a by-product of accelerating economic progress just as surely as greenhouse gases are, and the potential damage to our global labor ecosystem deserves a level of attention comparable to that of climate change. The engines of prosperity, fueled by innovation, are beautiful things to behold—unless you happen to be standing by the tailpipe. Recycling our natural resourcefulness along with our natural resources will surely convey benefits for all.