Only Humans Need Apply

Winners and Losers in the Age of Smart Machines

Thomas H. Davenport, Julia Kirby

• Publisher: Harper Business
• Year: 2016
• Print ISBN:
  9780062438614
• eBook ISBN:
  9780062438607
• Topic: Economics
• Subtopic: Business General
• Index: Economics

An invigorating, thought-provoking, and positive look at the rise of automation that explores how professionals across industries can find sustainable careers in the near future.

Nearly half of all working Americans could risk losing their jobs because of technology. It's not only blue-collar jobs at stake. Millions of educated knowledge workers—writers, paralegals, assistants, medical technicians—are threatened by accelerating advances in artificial intelligence.

The industrial revolution shifted workers from farms to factories. In the first era of automation, machines relieved humans of manually exhausting work. Today, Era Two of automation continues to wash across the entire services-based economy that has replaced jobs in agriculture and manufacturing. Era Three, and the rise of AI, is dawning. Smart computers are demonstrating they are capable of making better decisions than humans. Brilliant technologies can now decide, learn, predict, and even comprehend much faster and more accurately than the human brain, and their progress is accelerating. Where will this leave lawyers, nurses, teachers, and editors?

In Only Humans Need Apply, Thomas Hayes Davenport and Julia Kirby reframe the conversation about automation, arguing that the future of increased productivity and business success isn't either human or machine. It's both. The key is augmentation, utilizing technology to help humans work better, smarter, and faster. Instead of viewing these machines as competitive interlopers, we can see them as partners and collaborators in creative problem solving as we move into the next era. The choice is ours.

This landmark book provides a roadmap for thriving in the age of smart machines, showing professionals and business leaders how to:

• An Optimistic Automation Strategy: Move beyond the fear of job loss. Learn why augmentation—not replacement—is the key to success and how to reframe smart machines as partners, not competitors.
• The Future for Knowledge Workers: Understand the three eras of automation and see why this new wave of AI is different, threatening educated professionals from paralegals to medical technicians.
• Augmentation in Practice: Discover how humans and computers can combine their strengths to achieve outcomes neither could accomplish alone, leading to greater productivity and more fulfilling work.
• Five Actionable Steps: Choose your path forward. Explore five clear strategies—Stepping Up, Stepping Aside, Stepping In, Stepping Narrowly, and Stepping Forward—to build a sustainable and valuable career alongside AI.

1

Are Computers Coming After Your Job?

Even if you have never actually visited the New York Stock Exchange, you’ve probably seen it as a backdrop on financial news shows. It’s a telegenic image, with a series of kiosks for each trading firm, and the company logos of the stocks each firm trades on their walls. Electronic screens with fast-changing prices abound. Traders in bright blue jackets gather around market specialists and wave bits of paper or stick fingers in the air to represent the price they will pay to buy. Often we see them clasping their foreheads on days when stock prices take a nosedive. It’s the picture of capitalism.

Or is it? The last time we visited, in 2014, the visible action was a bit desultory, and we hear that’s the new norm. In 1980 there were 5,500 traders; now there are about 500. A trader could make more than a million dollars a year in the good years; now they struggle to pay back the $40,000 annual cost of a seat on the floor.

During our visit, the few traders we saw who were standing around didn’t seem to have much to do, and did have plenty of time to chat. When we asked why they seemed so relaxed, they explained that the great majority of trading is done on computers in a New Jersey data center. One told us that he no longer works on Mondays or Fridays. Even though the NYSE is one of the last “open outcry” exchanges with human traders, there’s not a lot of outcrying anymore. That’s why it’s so well suited to television broadcasts.

This situation is even further along at other exchanges; almost all equities are traded electronically. The Chicago Mercantile Exchange switched to automated trading of commodities in early 2015. Even bond trading, which has resisted automation because of the complex pricing and trades, is about half-electronic now. Algorithms and digital matching of buyers and sellers have replaced human traders. The result is fast and efficient—so much so that the profit margins from stock trading have been dramatically eroded. Human trading is likely to fully disappear within a few more years.

In addition to being the picture of capitalism, the NYSE trading floor is also the ideal image of automation. Time-lapse photography would show it becoming less populated each year. The jobs ended not with a bang, but with an extended whimper over forty years. Will your job still be around in 2055?

Let’s be clear: Humans are problematic as workers. First of all, they’re expensive, and they only get more so. On top of their basic wage, they cost their employers a third again more in payroll taxes, paid time off, health insurance, 401(k) contributions, and other perks. Think that’s all? Ask any facilities manager. Humans need ergonomic workspaces, heat, and light. Plumbing. All this is expensive, but it gets uglier. Ask any corporate counsel if humans like to bring lawsuits. Ask any security officer if embezzlement happens. Ask any inventory managers if they know about shrinkage. Ask any human resource executive what percentage of employees are engaged in their work (the average is 13 percent in the U.S.). But the trouble with human workers is a bigger deal than even that. As we’ll discuss in Chapter 2, technologies get smarter and cheaper all the time, but humans as a group don’t. You can’t simply download preexisting knowledge to a human. Every human starts at square one.

That trading floor is therefore a chilling scene. But at the same time it’s too comforting. It implies that “jobs” remain intact and the only problem is that some can now be taken by machines. That’s a source of solace to all of us who can name the reasons our own jobs can’t be accomplished by machines. But the truth is that jobs are not irreducible. All jobs are really amalgams of tasks, and every job today has some parts that can be effectively automated. The fact that no machine will ever be able to decide, as the executive director of the Pantone Color Institute does, that the design community will embrace “marsala” as 2015’s color of the year, or to predict, as executives must in an acquisition opportunity, whether the top talent of the targeted company will thrive or wilt in the proposed merged culture, or to compose a sentence, as we are doing, that rivals late novelist David Foster Wallace’s in its ability to remain grammatical while becoming remarkably convoluted does not mean that machines can’t take over the large proportions of a knowledge workers’ days that are not devoted to such rarefied tasks.¹

As computer programs focus on the tasks they can do, it’s those pieces of jobs that are taken away. The encroachment happens one task at a time, meaning that a job that is only 10 percent automatable doesn’t go away. It’s just that, now, nine holders of that job can do what used to be the work of ten. This is why, outside The Twilight Zone, you’ve seen virtually no one being summoned into an office and introduced to the computer who will now be doing his job. Instead, they’re just nudged, nudged, nudged toward the door.

And again, as with the manual workers who were tired of the dangerous, dirty, and dull aspects of their day, those nine people who continue to do a job are usually more than happy to see that particular 10 percent of their work go. There are loads of tasks they would rather not spend their time doing. The bane of a lawyer’s existence, for example, is “discovery”—the tedious process of sifting through documents and deposition transcripts in search of nuggets pertaining to a lawsuit. When “e-discovery” and “predictive coding” arrived on the scene, allowing much of this text review to be automated, few shouted their objections. All of us want to have our skills leveraged. In our work, we are all like Sherlock Holmes: We abhor the dull routine of existence.

As part of this, most workers eagerly embrace the machines that save them from the day-in and day-out chores of their jobs that take up time and add nothing to their net knowledge. If it were otherwise, companies’ IT departments wouldn’t be dealing with the scourge of “BYOD”—the growing practice of employees’ bringing their own favorite computers and other devices to the office. People want the extra productivity they get from state-of-the-art tools because it frees up capacity for them to take on more interesting challenges. They want that so much that they are willing to buy the tools for themselves.

So automation of one task after another tends not to be seen as the infiltrating enemy by employees. And neither is it seen as a problem by most customers. When a task can be performed well by a machine, they prefer it, too. Obviously, paying customers appreciate when higher productivity means that prices go down; while some people might cherish paying higher prices to enjoy artisanal products and services, most go for the product that does the job at the lowest price possible. But beyond price, automation often improves quality, reliability, and convenience. When ATMs arrived, customers didn’t complain about the automated option. By now, few could imagine life without them.

So if all of our jobs have parts that are succumbing to automation, which parts will we keep? We might like to think it will be the parts that it took us a long time to learn to do or that we have some special capability to perform. In other words, it will be the same parts that originally gave us the edge over all the other candidates for our jobs. But it isn’t as simple as that. Instead, the parts of our jobs we’ll keep are just the parts that can’t be codified. By that we mean that it can’t be reduced to known contingencies and clear steps. Codified tasks can be specified in rules and algorithms, and hence automated.

This is a theorem we will return to again and again in this book: If work can be codified, it can be automated. And there’s also the corollary: If it can be automated in an economical fashion, it will be. Already we’re seeing a rapid decomposition of jobs and automation of the most codifiable parts—which are sometimes the parts that have required the greatest education and experience.

Take the job of “physician advisor,” a role important in hospital administration and insurance. In medical settings, physicians see patients and come up with treatment plans for what ails them—but they are expected to do this with an eye to the hospital’s need for sound resource management. Extraneous tests or overnight stays use up limited resources and may not be reimbursed by insurers—and by the way, also take their toll on the patient. The physician advisor is there to review the doctors’ submitted treatment plans and suggest changes if they seem off base in any way. Can you imagine how much knowledge this person needs to have acquired to second-guess highly educated physicians? Beyond that, the role requires diplomacy. A medical newsletter describes the job profile as follows: “[A] skilled physician advisor must learn to manage by influence rather than by authority. This requires a delicate balance between collegiality and firmness relative to the issues at hand. It also requires the ability to provide reasonable alternatives rather than indicating what can’t be done.”²

It sure doesn’t sound like a role a computer could take on. Yet IBM’s Watson and other automated systems are now being used at health insurance companies like Anthem to weigh in as physician advisor. And the point to note is that the most cognitive part of the job—the “ability to provide reasonable alternatives” based on extensive knowledge of similar cases in the past—is the part being automated here. No doctor could possibly hold in memory more prior cases than Watson can. But that is also probably the part in which the doctor takes most pride—certainly it’s what those hard-earned diplomas framed on the wall attest that she can do. Does Watson get rid of the human in the physician advisor job altogether? No, at least not yet. But by supplying that precious knowledge base, it allows the recommendation task to be done more quickly, or by a less credentialed person—perhaps a nurse. Presumably this means the hiring manager for the role now focuses on the other aspects important to success—like that ability to achieve “a delicate balance between collegiality and firmness.” That is undeniably a rare talent, but it’s probably not something anyone explicitly trained for, let alone did a residency in.

We should pause here to mention the threat of “deskilling,” since the physician advisor’s evolution is such a prime example of it. The term, first coined by the Marxist sociologist Harry Braverman, is commonly used to describe both what automation does to jobs and what it does to the labor force. The jobs are deskilled when technologies are introduced that no longer require workers to have formerly necessary skills—meaning that semiskilled or unskilled workers can now hold those jobs. In turn, the labor force is deskilled when, enough machines having taken over a particular task, the skill becomes a “lost art” to people. A simple example courtesy of a 2014 survey of Britons: 40 percent of them admitted to relying completely on autocorrect technology to get their spelling right in daily correspondence—and more than half of those say if they were forced to go without spellcheck, they would “panic.” Yet 90 percent say it is still “absolutely crucial” for children to learn to spell properly.³ For Braverman, and many thinkers since, deskilling is a very dangerous phenomenon. As early as 1974, he was already predicting its inevitable creep into knowledge work, and worrying about the emergence of a “white collar proletariat.”

We do expect deskilling to accelerate as computers take on more knowledge work tasks. Imagine the art of teaching, for example. Today a teacher in an elementary grade performs a number of important educational functions. One is to determine what content students have already mastered and what they still need to learn. Another is to actually transmit the content to the students. A third is to maintain discipline and cultivate a love of learning in the classroom. It’s unlikely that computers will be able to maintain a calm and quiet demeanor among a group of twenty-five or so fourth graders, but many of the other functions that teachers perform can be carried out by computers, and in some cases this is happening today.

Computers are better than many teachers, in fact, at diagnosing what each student needs to learn, and customizing the educational content to the student’s needs. Given traditional classroom sizes, these decisions are just too tailored and time-consuming for many teachers to make effectively. Computers are also good—at least when the educational software is well crafted—at transmitting educational content to students, and knowing when they have mastered it. We could imagine, then, that the human roles in highly computerized schools could be reduced to monitoring and discipline, and being occupied by people who look more like “teachers’ aides” and proctors than professionals with deep knowledge of pedagogy and subject matter. This isn’t a popular idea with teachers’ unions, but that might be the only reason the wholesale shift hasn’t happened already.

Will a Computer Take Your Job?

Is it starting to feel to you like your highfalutin knowledge worker job might not be so invulnerable? To read the signs a little more, let’s consult the radiology profession. Radiologists are another highly educated group whose jobs are being deconstructed, with the parts they trained longest and hardest to do being the very ones that are automated. And note here that, not long ago, we would not have said the ability to study an X-ray or MRI and render a diagnosis could be codified. This is, after all, the profession that loves its “Aunt Minnie.” That’s a term, reportedly first used by a Cincinnati radiologist named Ben Felson in the 1940s, that honors tacit knowledge. As a radiologist gains practical experience, some diagnoses become possible at a glance, because the same image has come up so many times before. In Felson’s words, the radiologist is presented with “a case with radiologic findings so specific and compelling that no realistic differential diagnosis exists.” Over time, the radiologist learns: If it looks like your Aunt Minnie, then it’s your Aunt Minnie.

Radiology isn’t just a specialty requiring a long education; it has been one of the highest-paid medical specialties in the United States. The explosion of imaging technologies over the past couple of decades made doctors who can read such images the “cash cows” of hospitals and medical practices. But in recent years their numbers have been decreasing and incomes falling. It’s instructive to look at the three-step process that brought this about. First, the image-reading work was outsourced and offshored to radiologists overseas, because that allowed a greater volume of images to be processed. This could only happen when the images were digitized and could be sent across an ocean in an instant. Second, the discovery of the much lower cost of those offshore physicians’ time caused more work to flow their way. And all that shifting of work off-site from hospitals forced administrators to codify it more, in order to monitor the quality of work being done remotely. Finally, that thorough codification has made it more possible to take the ultimate step, to automation.

There are already technologies that can read CT scans and MRIs and seize upon the likely lesions that may mean cancer. They highlight the suspicious spots with prominent brackets so that any doctor or nurse can see the problem. Looking ahead, as the prices of imaging devices continue to fall, the day will come when every family doctor’s office has one—thoroughly deskilling the interpretation of radiologic findings. Aunt Minnie is rolling over in her grave.

Not surprisingly, the number of medical students applying for radiology internships in the United States has been dropping steadily over the past several years. But again, there are still parts of what is today a radiologist’s job in a hospital setting that no machine can perform. There is an art to getting a nervous patient properly positioned for imaging, for example (though this task is often performed by technicians, not radiologists). “Interventional” radiologists, moreover, must be able to read images in real time as they direct minimally invasive instruments through tissue. That is a skill that is still a very long way from being automated. It’s digital, but still in the sense that involves fingers.

The process we’re describing, of machines taking the high-end cognitive parts of work and turning people into a sort of human user interface, is occurring across many professional realms. Actual decision-making roles have been ceded to computers—and they are doing pretty well in those roles, despite some occasional hiccups. “Program trading” (also known as high-frequency, algorithmic, or quantitative trading) of equities and fixed-income investments, for example, is widespread on Wall Street and around the financial world. It’s one of the reasons why the New York Stock Exchange is so quiet today. Decisions about which stocks and bonds to buy for what price used to be made by human traders but are now largely made by computer. Likewise, decisions that used to be made by human pricing analysts are now arrived at automatically. What price should an organization charge for perishable goods like airplane seats and hotel rooms? That depends—and depends on more factors than the human mind can process in time to make the sale. With thousands of flights per day and hundreds of prices per flight, the result is literally millions of airline price changes per year; one analysis found that the lowest-price ticket on one flight changed 139 times based on seat availability and demand.

It goes on. Who now makes the business decisions of whether to give someone a mortgage or credit card, the premium to charge for an insurance policy, or which ad to show to a media consumer? All require chewing through massive amounts of data, intense analytics, and strict adherence to rules. Very few humans need apply for these jobs. At the level of setting the rules and writing the code that automates that decision-making, a few people still have very important roles. But on a day-to-day basis, these relatively structured and quantitative tasks are today no longer performed by wetware.

Ten Reasons to Look over Your Shoulder

We’ve always loved the line in H. G. Wells’s classic The War of the Worlds when the narrator rues the fact he did not react sooner to the arrival of an “intelligence greater than man’s”—in his case, Martians landing on earth. Co...

Table of contents

• Dedication
• Contents
• Introduction
• 1. Are Computers Coming After Your Job?
• 2. Just How Smart Are Smart Machines?
• 3. Don’t Automate, Augment
• 4. Stepping Up
• 5. Stepping Aside
• 6. Stepping In
• 7. Stepping Narrowly
• 8. Stepping Forward
• 9. How You’ll Manage Augmentation
• 10. Utopia or Dystopia? How Society Must Adapt to Smart Machines
• Acknowledgments
• Notes
• Index
• About the Authors
• Also by Thomas H. Davenport and Julia Kirby
• Copyright
• About the Publisher