Will AI Destroy Jobs? Yes - But That’s Asking the Wrong Question

The looming policy question behind AI progress is whether or not the advancement of AI will destroy people’s jobs. Our modern culture has a deep-seated fear that our technology will destroy us and our work in some way. Everyone wants to know, “will AI destroy work as we know it? And if so, am I safe?”

The question has already been in the zeitgeist for some time. Andrew Yang even ran a presidential campaign premised on universal basic income to offset losing work to machines. But now, as we live through a cambrian explosion of generative AI tools with an overwhelming suite of capabilities, the question is top of mind and promises to remain a key issue in the coming years.

Most of the discussion I see around the issue asks whether or not AI is going to eliminate jobs. That’s the wrong question to ask. Of course AI will eliminate jobs. Every productive technology does. Instead, the more important question is will the transition be painful for our society?

I don’t have a crystal ball to predict the future, but given the state of the current discussion, I want to be more thoughtful in my own thinking. So instead of rampant predictions, I organized my thoughts into two illustrative examples of how the automated process can go: tractors and computers. These simplified case studies can help us think clearer about what the key factors are as we think about the automation of jobs from AI.

Tractors

"Is a tractor bad? Is the power that turns the long furrows wrong? If this tractor were ours, it would be good—not mine, but ours. We could love that tractor then as we have loved this land when it was ours. But this tractor does two things—it turns the land and turns us off the land. There is little difference between this tractor and a tank. The people were driven, intimidated, hurt by both. We must think about this.” - John Steinbeck, Grapes of Wrath

If you read Grapes of Wrath you might remember Steinbeck’s fixation on the role of tractors in displacing Midwest farmer families. (You may also remember his vivid sexual depiction of the machines’ interaction with the land, but that’s neither here nor there). Most people remember the westward Depression-era migration being caused by the Dust Bowl, but it was first kicked into motion by the automation of labor that eliminated the jobs of tenant farmers.

Steinbeck’s book is remembered for demonstrating the traumatic social upheaval as millions of people sought new work out west in California. This labor shift in the midst of the Great Depression left families living in destitution, overwhelmed the social safety nets, and led to intense political conflict and backlash. At one point the LAPD even (unconstitutionally) stationed officers at the state border trying to turn people away who were looking for work. It was an intense, violent time marked by desperate people struggling to survive.

What tractors offer is a case study of the worst case outcome that can happen when people lose their jobs. Families struggle to survive and entire communities can be devastated as their underlying social fabric is torn apart.

The problem with simplified case studies is that it’s impossible to measure the extent farming families’ misery was driven only by technological displacement. Tractors didn’t arrive in a vacuum, as The Great Depression and the Dust Bowl were obviously major factors in making that era uniquely miserable. But the point of this exercise is to illustrate the different paths job replacement can take, and in this case it was deeply traumatic. What are some of the factors that made this the case?

Factor 1: The job loss dramatically shrank the agricultural workforce

Tractors and other farm machinery quickly shifted a significant portion of the labor market away from the land. Compared to 1930 numbers, farm jobs decreased by 10% by 1940 and 25% by 1950. Whereas one third of American workers were farmers in 1910, only one in ten remained in agriculture by 1950. That’s means displacement occurred in a shrinking sector with few immediate work alternatives.

Factor #2: The job losses were geographically concentrated and socially connected

Entire regions of the Midwest were solely based on agriculture. Most communities there had been established when settlers came to farm, and farming remained the main profession of the next generations. The loss of tenant farming jobs eliminated most work for entire communities and created a death spiral of negative network effects.

Factor #3: The tractor did not immediately create new work

If you were a tenant farmer who lost their job just as everyone around you also lost their job, where do you turn? There were few jobs to be had managing or overseeing tractors. Most tenant farmers’ skills were only valuable in a shrinking agricultural sector, and the arrival of tractors didn’t simultaneously lead to new opportunities. Anyone out of work had to find a job in a different part of the economy.

Factor #4: The economy was in the tank

Tractors eliminated jobs amid The Depression, the worst labor market in American history. Eventually all that free labor allowed people to work in other productive areas like manufacturing, which was critical during WWII when the labor market suddenly tightened, but that was over a decade away. Macroeconomic conditions matter.

Computers

Ok, enough Steinbeck. Let’s instead think about Octavia Spencer helping putting Americans on the moon. The movie Hidden Figures gives an example of why automation can be a net positive by creating new opportunities. In the movie, Octavia Spencer’s character leads NASA’s group of black female ‘computers’, the woman who manually double checked all the mathematical calculations. NASA then acquires their first IBM machine, eliminating the need for their department. By the end of the film her character is leading the new department operating the machine along with her old colleagues, the perfect ending to a Disney movie.

When we think of the arrival of computers, we think of optimistic stories like this, or Steve Jobs describing “a bicycle for the mind.” Why is that? It’s because computers almost immediately and directly created more opportunities than they eliminated.

Computers eliminated countless jobs. Think of all the calculations required to pull off feats of engineering, general business accounting, worldwide logistics systems, and rudimentary tasks like typing and sending memos. Those all became obsolete, entire divisions like ‘human computers’ were not longer necessary.

Yet computers stand as a direct contrast to tractors in that there was no parallel social upheaval. There was no real vilification of the technology or novels documenting the suffering of displaced mathematicians. There’s a few obvious reasons for that:

Factor #1: Computers didn’t shrink the knowledge-worker workforce

There’s no charts to show a dramatic drop-off in knowledge work employment like there is for farmworkers. Those sectors continued to grow and require more workers to the point today where as of 2016 over 75% of U.S. jobs required what Brookings describes as medium-to-high digital skills. Software tools “ate the world” while expanding the scope and demand for the skills they were displacing.

Factor #2: Job losses were distributed across geography and industry

There was no single geographic hub of computational workers like there was in farming. No cities of switchboard operators were devastated in the way a factory town is when the town plant is shuttered. Jobs replaced by computers were spread across sectors and jobs, so any disruption or job loss was diluted across the broader economy. This meant there were no negative network effects or communities that were irrevocably

Factor #3: Computers immediately and directly created new opportunities

The most important difference is that the arrival of computers immediately created new work directly adjacent to the jobs it eliminated. At a macro level, the process of creative destruction occurred seamlessly, more or less like in the movie Hidden Figures where the people formally doing calculations by hand were hired to manage the work of the computers. The new jobs created by computers were often analogous to the roles that were eliminated. Engineers, accountants, and writers didn’t have to change professions, they simply had to update the tools they used to do their work.

Factor #4: The economy was HOT (or at least not cold)

The implementation of computers took several decades, but in general the years since computers arrival have been defined by economic growth and relatively low unemployment (barring the Great Recession and the recession in the 80’s). A lot of that growth was driven by new technology like computers, but it also was an era of remarkable peace and international cooperation compared to the rest of human history. A rising tide lifts all boats, or in this case, a hot labor market dulls the negative social consequences of technologically-driven upheaval (I’ll workshop that …)

Looking Ahead to AI

So where does this leave us with AI? It’s impossible to predict, but those 4 factors provide a clear sense of what factors to look for when we think about job loss. While we can’t predict future macroeconomic conditions all the other factors are promising. From my understanding as of today, AI should (1) lead to greater productivity and growth in knowledge work, (2) be distributed throughout a number of sectors, and (3) create new jobs (it already has), all of which bodes well.

Of course, there’s also the black swan scenario where advanced AI leads to total destruction as the machines try to turn us all into paperclips. But it’s impossible to know how great a risk that really is. So I’ll set that aside for now for the purposes of this essay.

Given my inability to predict the future (as evidenced by my annual March Madness bracket), I don’t claim to know that the AI transition will be smooth after a cursory glance at the factors I listed out. So instead, below are some of the key things I’ll be watching to determine how painful the AI transition will be.

Reasons to not freak out:

1. Implementation could take forever or we might resist it. Just because the technology exists to automate workers doesn’t mean companies will do it. Companies had the ability to hire remotely for years but didn’t choose to do so until they were forced to by COVID. Similarly, it took forever to translate a lot of analog tasks to go digital. Most employees work in large, established companies, for the government, or in highly regulated industries, and those groups often resist change in how the fundamentally operate. Just because generative AI creates the capability of eliminating junior employees across departments doesn’t mean they will.

2. It’s easier to absorb knowledge workers back into the economy. The most remarkable piece of recent generative AI breakthroughs is that they have jumped the value chain and are doing the work of white collar workers and creatives. While it’s startling to see AI’s capabilities leapfrog into higher value work, that jump may also bode well for job displacement. Software engineers, writers, business analysts, and graphic designers aren’t tenant farmers. They have valuable skills in our modern economy, and they’ll find a way to use them. Over 120,000 tech workers were laid off in 2022, with more this year, yet that hasn’t led to major social upheaval. Why wouldn’t it be the same for AI?

Reasons to be concerned:

1. The implementation of AI tools could occur quickly. As opposed to the cultural factors I listed above suggesting companies will resist change, it’s possible “The Great Implementation” happens much quicker. Transitioning companies from analogue to digital tools took decades. Now that companies are already digital, it’s possible they more seamlessly integrate autonomous productivity into their workflows much easier. Microsoft and Google are already integrating generative AI into the tools every company uses for work.
   
   For example, a corporation like Coca-Cola now has decades of digital records of their past marketing campaigns, financial models, and operating procedures. In a matter of days they can train models to replicate past work, and then in a matter of months can (maybe?) figure out how to automate a large portion of their workflow. What if the implementation of AI is such a competitive cost advantage that companies’ only chance to remain competitive is to implement them as quickly as possible?

2. Job loss could be devastate specific knowledge worker sectors. Take tools like Adobe’s Project Shasta, software that immediately takes any audio and radically improves the quality. What if, in a matter of months, tools like this eliminate a large portion (25%?) of the work currently done by audio engineers? Or what about the latest version of MidJourney that can now do a convincing impression of work that used to require professional photographers and editors? Now do the same thing for graphic designers, copywriters, marketers, and any other task that new AI tools are effective at. Yes, those jobs will still exist and be transformed, but what if companies are able to maintain current output while eliminating 40%, or even 20% of their headcount in a few short months? Such a loss of work could have sharp, negative effects. Work that previously required an artisanal touch could be commodified, slashing wages, and making an entire segment of the creative class obsolete. I can imagine a sharp social backlash if people are suddenly told that the skills they’ve invested years of schooling and training on are no longer valuable.

We’ll see …

These is just my initial attempt to organize my thoughts, but I’m skeptical of my own conclusions. As Ezra Klein’s recent essay articulated, it’s hard to wrap our brains around where this is all heading. We’re conjuring up technological forces that we don’t know how to predict or control. I guess I leave this at the end just to emphasize my own uncertainty and deep curiosity at where this is all heading. We’ll see! If you have strong thoughts on where I’m wrong or what you’re looking at, please shoot me a line on twitter. I’d love to hear other people’s thoughts.