Predictions of Massive Job Losses via AI

No doubt such concerns have been amplified by recent headlines on the potential impact of AI on employment. After all, some predict that robotic automation will eliminate enormous portions of the workforce. Oxford University researchers have predicted as many as 47% of U.S. jobs could be automated away by 2025.¹

There are about 160 million jobs in the entire U.S. workforce, so the Oxford prediction would mean roughly 75 million jobs would simply be gone.² Extrapolated across the G7 industrialized nations, where there are roughly 368 million jobs, this would mean at least 173 million jobs would be eviscerated by the new machine within eight years. Such levels of unemployment occurring at such a pace would lead us to an Elysium-class dystopian future.³

Of course, some jobs will be automated away, but 173 million G7 jobs by 2025? We don't think so. 173 million is such a large number it does not pass the smell test, for that's larger than the entire population of Russia, or the combined workforces of Germany, the UK, France, Italy, Australia, and Canada.

Take a deep breath. The rise of the new machine is not going to lead to full-scale revolution in the streets. A lot of jobs, indeed, millions of jobs over time, will be automated but not at a scale or pace that will create the social dislocation that some are predicting (and many more are fearing).

There are literally dozens of studies on this topic, and in reviewing nearly every one of them over the past three years we have found that the Oxford report takes the most extreme position. Not surprisingly, it has grabbed most of the headlines as it has tapped the nerve of insecurity in our slow-growth, stalled economy.

The consensus among the majority of the studies, though, is far less dramatic, positing a range from around 5% to 15% of jobs being automated away over the coming 10 to 15 years. Based on our analysis, we believe that the middle of this range, about 12%, is the most likely scenario.

This level of job dislocation from AI of course is still significant. 12% is the equivalent of around 19 million jobs in the United States. If one of those jobs is yours, life will undoubtedly be very tough. However, what's often overlooked in examining the big picture of employment levels is the growth of new jobs. We believe that there will be almost 21 million new jobs, about 13% of the current U.S. labor force, directly created as a result of the growth of the new machine. In case these numbers—19 million jobs disappearing and 21 million jobs being created—sound implausible, keep in mind, as a point of reference, that since 2010, during the years of the post–Great Recession recovery, 15 million private sector jobs have been created in the United States.

Our assumption is that in the industrialized world, for nations that embrace the Fourth Industrial Revolution, unemployment rates by 2025 will be roughly what they are today.⁴ This relatively small expected change—plus or minus—in the net unemployment rate will, however, mask huge changes in what work we do and how we do it. Within the overall labor force, there will be massive job transition (often creating skills mismatches), and figuring out “what to do” within this churn is what this book is all about.

The new machine will change the G7 labor force in three distinct ways:

1. Job automation: Roughly 12% of existing jobs are at risk of being taken over by systems of intelligence.
2. Job enhancement: Roughly 75% of existing jobs will be altered or enhanced by the bot. The employment will remain, and these jobs will be delivered with greater output and/or quality.
3. Job creation: 13% net new jobs will be created as the new machine creates new revenue opportunities and/or new job categories.

In our view, any responsible analysis of the impact of the bot on jobs needs to consider job automation, enhancement, and creation together; new machines always give as much as they take away.

We believe that the more fearful predictions are based on research that is incomplete and misinterpreted and is often generated by those who aren't particularly close to work either inside modern corporations or today's technology.

In terms of framing this issue for you and your organization, there are four key considerations that we find to be useful in understanding and explaining what will really happen to jobs within your industry or business.

• Manual vs. knowledge labor: Researchers—and many of us—still view manual labor and knowledge labor as interchangeable (and therefore “automatable” in the same way). They are not interchangeable, and therefore their substitution by machines is different.
• Jobs vs. tasks: We tend to look at “jobs” holistically, instead of seeing them as being composed of their various tasks, some of which can be automated and others of which will never be automated. In looking at the tasks that make up a job, we quickly see that some jobs indeed will be replaced by the machine, but others will be merely altered and enhanced.
• Technology as job destroyer and creator: Most analyses look at automation and technology solely as a job destroyer, but every major technology shift through history has also led to job creation. Dystopian visions of the future have been largely fueled by ignoring the growth side of the equation.
• Time: In trying to understand the future, one of the most important variables is time. This is often overlooked or extended so far out as to make some predictions meaningless.

Let's look at these four in detail.

Manual vs. Knowledge Labor: As Goes the Factory, So Goes the Office?

For the past several decades, automation in the industrial economy has been grand in scale, yielding product improvements, process efficiencies…and horrid layoffs. Huge factories full of people are now more productive with 80% or 90% fewer people managing fleets of machines. Robots now make our cars, unload ships, assemble any number of products, and even vacuum our floors. When the price was right, the bots took over the work of thousands of physical laborers.

That shift was easy to see, and it was painful in a lot of ways. The three of us as authors were all raised in towns that stood as industrial leaders in the 1960s, but which had become economically depressed by the 1990s. At a human level this transition was nothing short of brutal and provided harsh truths about economics and technology. In fact, the manufacturing robot improved efficiency and boosted quality. Those jobs are gone, and they won't come back. As such, many are drawing a parallel between the blue-collar labor collapse of Detroit and a similar impending white-collar labor collapse in places like London, New York, and Los Angeles.

However, drawing such parallels has limitations, for manual labor and knowledge labor have very different attributes and, as such, the automation of them is substantially different. An old aphorism is instructive in this regard:

If I give you a dollar, then you are a dollar richer and I am a dollar poorer. But if I give you an idea we are both richer, for now we both have the idea and your reaction to it has made it more valuable to me.

Putting a lug nut on an automobile chassis, whether done by hand or machine, is done once and then there's no going back to it. Additionally, this task is the same every time it's conducted. However, knowledge automation is different, for often the atomic unit of “knowledge” can be reused many times and continually “enriched” to become more valuable over time.

As an example of this, think back to the newspapers of a generation ago. These were true one-to-many offers, as all of us received the same newsprint each morning. As such, because those knowledge products arrived in a physical form we thought of them as being similar to that lug nut.

However, now think of the highly personalized news feeds received online today; your local news, weather, sports scores, stock portfolio, and traffic information. All at your request, at any time or place. Think of the permutations involved in generating such customized, curated experiences. In a mid-sized city of 500,000 people, simply delivering the personalized needs across the five basic news variables (news, weather, sports, stocks, traffic) would yield 3.12²⁸ permutations alone! Obviously, it's a mathematical impossibility to think that the newsroom of old could even begin to consider delivering personalized news to everybody in town. Yet today, bots, by recombining all of those units of knowledge work, can do this effortlessly (and we show how they do this in more detail in Chapter 7). Therefore, automation of knowledge assets is not zero-sum, in that it's not just a matter of removing existing labor. It is also allowing for more throughput, and this often manifests itself as a level of mass customization that was impossible before systems of intelligence.

This phenomenon—the codification, recombination, and repurposing—of knowledge assets has broad implications. In the coming pages, we examine how it will impact core processes in your business (within sales, human resources, supply chain management, or finance) to both streamline them and to greatly increase throughput, quality, personalization, and overall capability.

The point is, knowledge work is very different from manual labor. When the bot is applied to knowledge processes, even for the explicit purpose of automation, the underlying knowledge assets become richer and can be used (and reused) in interesting and productive ways. As such, the envelope of work potential (and corresponding output) is usually expanded, thus removing the one-to-one work substitution that occurred with the automation of manual labor.

Don't Confuse Jobs with Tasks

Related to this is the second flaw that most of the doomsday analyses share: they fail to make the crucial distinction between “jobs” and “tasks.” These studies tend to view “jobs” in a binary sense (i.e., “they will be automated away” or “they will not be automated away”). That's far too simplistic; any knowledge job is a collection of tasks. Some of these tasks are ripe for automation, while others never will be. In the vast majority of cases, portions of a job will be impacted or replaced by the bot, while other portions of it will be untouched or even enhanced.

Consider Tamara, an accountant laboring in your company's accounting department. Her job consists of dozens of tasks, some of the primary ones being:

• Documenting financial transactions
• Preparing asset, liability, and account entries
• Preparing tax filings
• Auditing transactions and financial statements
• Recommending corporate policies and procedures
• Reconciling financial discrepancies
• Creating profit and loss statements
• Creating balance sheets
• Providing strategic counsel
• Pursuing data integrity

Some of these tasks will be automated away with the use of the new machine or made dramatically more efficient by them, but not all of them. Tamara's job will change, but it will not go away completely. To imagine that the entire profession of accounting will in short order disappear and be replaced by software is to fall prey to the trap of over-extrapolating how and how fast technology changes things in the real world.

One of the best analyses of machine-based job displacement has been undertaken by Forrester Research, which has taken the “task-based” vs. “job-based” approach. In many cases, “robotic process automation” will eliminate only portions of a job, most often the ones humans find to be drudgery. Thus, in many cases 20% of the routine—and highly boring—portions of a job go to the machine.

Forrester puts it like this:

The greatest change to the workforce…will be in job transformation—that is, occupations in which 25% or more tasks are automated, leading to redeployment and responsibility-shifting on the part of the worker. In 1992, cable TV technicians had a relatively simple job: connecting coaxial cable to a pole and into households. Since then, their job tasks have expanded to include Internet service, wireless routers for Internet, voice-over-IP (VoIP) telephony, and even home security installation. With each new task, the overall composition of the job transformed, even requiring the technician to log into your PC to activate services—a new skill set for sure. Similarly, we'll see jobs transformed across all categories….⁵

Figure 3.1 highlights the percentage of job tasks that will be impacted by the machine, not necessarily replaced by the machine.

For example, the findings are not claiming that 92% of management, business, and finance jobs will disappear by 2022. Instead, they are asserting that 92% of the “workflows, processes, and metrics” in those job types will be changed by the new machine. In cases where a specific job has too many tasks “cannibalized” by AI, the job can, in fact, disappear completely. This accounts for the 12% of jobs that will be automated away.

Such a task-based analysis (by vocation) starts to paint a much more realistic picture of the machine-generated transformation of work, and it has three important elements:

• Time: Forrester's findings still show a significant near-term impact, but they are not suggesting the end of employment as we know it during the next few years. Also, each worker can see the steady impact of automation in his or her chosen field. Seeing the transition provides both a personal roadmap and some time to adapt or to develop new skills.
• Eliminating rote work: Automating 50% of a job may not be a bad thing. As we explore in the coming pages, in many cases this automation is focused on rote, non–value added activities (e.g., the manual grading of homework by a school teacher). When workers are liberated from such activities, how can they then reinvest their time?
• Performance growth: Job performance can then be enhanced. The worker can double down on the more human elements of the job, double output, or greatly increase quality of delivery. For example, automation won't make a teacher disappear; instead it can make that teacher much more effective.

Don't Overlook the Job-Growth Story

This atomic-level view of a job and its associated tasks has provided a foundation for our AHEAD model. Will some jobs be automated away by AI? Yes, of course. But far more will be enhanced, and in time millions more new jobs will be discovered, driving future employment. Our confidence in these predictions is not based just on the capability of the new machine in the present; we have seen this movie before—automation is really the story of business.

Today, we are all great beneficiaries of industrial-age automation. The creature comforts many of us enjoy—the cars we drive, the TVs we watch, the computers we use, the appliances in our kitchen, the clothes we wear, the flights we take, the food we eat, and the entertainment we enjoy—are all delivered at a price/performance ratio unimaginable just a few generations ago. After all, look no further than your 60-inch flat-screen high-definition TV, which, in real terms, costs one-third the price of the old 19-inch RCA that graced your parents' den.

All of these goods and services are the direct result of automation. However, any mention of the word “automation” in 2017 is frequently met with a negative—and sometimes viscerally negative—reaction. Many seem to forget that throughout history, automation has provided a net benefit to society. In the process of automating our work and our society, through generation after generation, three positive things have repeatedly occurred:

1. A new abundance has been created; sales of products and services produced by automation—now vastly more affordable and of higher quality—skyrocket.
2. With the new abundance, overall employment rises, even when there is less labor input per unit.
3. Society gains a net benefit, with higher living standards created by newly affordable products and services.

Each time a new form of automation is introduced, there is consternation and anxiety. After all, in the moment we often cannot yet see the new abundance, the growth of overall employment, and the net societal benefit, but we certainly recognize the initial job losses. This process of automation, initially cursed and then ultimately lauded, has repeated itself with great consistency.

Think of the steamboat, the locomotive, and the assembly line. With the introduction of each new technology, vested interests were threatened and business as usual was upended. This was the story of the aforementioned Luddites. In the context of the moment, their arguments had merit. Yet in the context of history, as we recognize how the loom clothed the world, established a foundation for global trade, initiated the growth of a large middle class, and launched various related industries, the Luddites were wrong.

Automation is a deep and unstoppable force. Automation of your core processes is a key first step for you and your organization to gain a deep operational understanding of the new machine, and to unlock its potential for future bounty. Don't allow Luddite thinking to remove those opportunities.

This may sound like the 30,000-foot theory, but it has a very practical application. As a manager, in considering automating jobs within your company, these four considerations should be useful. After all, if one is afraid to automate some internal jobs away, then, in time, all jobs at the company may be jeopardized (given that the company will become noncompetitive on cost). Additionally, the jobs that can potentially be enhanced by automation won't come to fruition, and new offers with their associated jobs won't come to market.

The new machine will be painful for some to accept, but this shift is inevitable; if we manage it wisely, it will ultimately be positive for our companies and our societies.

The Pace of This Transition

Our prediction is that AI will impact nearly 100% of knowledge jobs, while completely eliminating approximately 12% of them. But the key question is “When?”

AI will eat existing jobs in a “slowly, slowly, suddenly” manner. Certain tasks will quietly and increasingly become automated and will then hit a potential tipping point that will fundamentally impact the very nature of certain jobs (such as the 50% cannibalization point in the Forrester model). This transition will follow the pattern of technology adoption outlined by Bill Gates in that “We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten.”

Consequently, it's easy to believe both sides of the job dislocation argument. In taking a short-term view (of the next three years) one can think “No way our accounting department will be automated away.” Yet, in understanding the capabilities of AI platforms, one can take a 15-year view and think, “No way we will have more than a handful of people processing accounts receivables by 2030.”

The key for setting a realistic timeframe lies in (a) the task-based view of work and (b) the value of the remaining human element. In looking at these two variables we can start to make solid predictions as to how quickly the bot will eat a certain profession.

Getting AHEAD in a Time of Churn

In concluding this examination of the job-destroying nature of the new machine, the perspective we've outlined provides a basis for the remainder of this book. In the coming chapters we explore the practical applications of these dynamics and what they will mean for you and your organization. In Chapter 7 we examine automation in depth, looking at specific processes, functions, and jobs within your company that are particularly ripe to be taken over by the new machine. In reading this chapter, you may think, “Hmm, Tamara in accounting is in trouble if she doesn't respond quickly.” In Chapter 9 we outline the jobs that will be protected and enhanced. In Chapters 10 and 11 we look to net-new job creation through the new abundance and a process of invention and discovery.

However, before we get to determining the future of work, we need to take a deeper look at the new machine that's driving all of this change.