When people think about automation, they write a scary story in their heads. A business hums along, employing lots of people. A machine capability is developed that does what employees do, for less money. Cost-cutting bosses lay off all their employees. It’s all very sudden and dramatic and confrontational. Things that make it easy to get worked up about.
I think the real story is different. Machines don’t replace humans. Machine-dependent businesses replace human-dependent businesses. To the extent that automation causes job losses, that’s how.
Think about travel agents. If the ability to communicate instantly over the internet to hotels and transportation companies around the world had only been available to travel agents and not to travelers, it might have enhanced their jobs, rather than replacing them. They would have been able to use time saved on international phone calls on client service, either taking on more clients or doing more thinking and research on behalf of existing clients.
But instant communication with hotels in Hungary and bus companies in Beijing wasn’t just offered as a tool to travel agents; it formed the basis Expedia, Orbitz, and Travelocity, examples of a new business model in the travel industry with low labor concentration that served travel agents’ erstwhile customers. By offering a more convenient and transparent user experience, they gobbled up travel agents’ marketshare. Employment of travel agents collapsed.
Retail is another example. Retail jobs have remained stagnant over the last few years as a direct result of information technology. But it’s not because JC Penney decided to close physical branches and fire workers and invest in e-commerce instead; it’s because legacy brands have been out-competed by Amazon and others that use information technology as the centerpiece of a new business model -- and employ far fewer people.
ATMs did completely substitute for the core tasks that tellers had been relied on to perform -- routine dispensing of cash. But adding ATMs didn’t change banks’ structure and business model, and they didn’t lead to the emergence of new IT-based services that competed for banking customers. As James Bessen has written, although ATMs reduced the number of tellers needed per bank branch, this made it cheaper for banks to open new branches. Banks used this efficiency boost to expand, and since their business still required some tellers to handle special transactions and serve in customer engagement roles, overall teller employment has increased as ATMs have been adopted.
Automation is confusing. There are so many contradictory theories and examples demonstrating why robots will or won’t take all of our jobs. At different times I’ve felt myself leaning in either direction. There’s really no safe conclusion than to throw up one’s hands and say, “only time will tell.”
But this theory of automation might be consistently applicable across the economy: when thinking about whether a new machine capability will lead to job losses, the critical variable is whether the new capability is introduced as a solution used to enhance the efficiency of existing business models, or used as the foundation of entirely new business models that employ fewer people than their legacy competitors. From my research so far, in the former case, jobs are usually enhanced more than they are replaced. But in the latter case, job losses ensue. ATMs didn’t undermine the business of banks. But travel booking sites did undermine the entire business model of travel agents. Savings garnered from ATMs went to banks, so they could reinvest and expand. But savings garnered from travel sites went entirely to consumers.
I don’t think this means that we’re doomed to staggeringly high unemployment. As Ezra Klein has written, humans are really good at reinventing our idea of economically valuable labor. Even as retail jobs have remained stagnant and hundreds of thousands of travel agents have had to find other work, unemployment has remained low. Hundreds of thousands of jobs are created in industries like home care, whose core tasks will not be performable by machines for a long time.
But automation can still threaten widespread prosperity. When travel agents started disappearing, the businesses that replaced them didn’t employ nearly as many people. Most of the value lost by travel agents was transferred to consumers who traveled frequently. In the case of retail, value has transferred from retail workers to people who like to shop online. In both cases, the result of human-dependent businesses being replaced is modest gains for people who I hypothesize are pretty well off -- and wild gains for the very few people who build and maintain the businesses that did the replacing. New jobs that are created in other sectors either don’t pay well or require skills that few people have.
As more data gets produced, as more and different kinds of sensors are built to capture that data, as algorithms to analyze that data become more sophisticated, and as devices to act on what algorithms decide expand through our lives, we’re going to see algorithm-centric businesses replace human-centric ones in many, many more sectors of the economy. That’s what I’m worried about.
I wrote this theory off the top of my head after thinking about it this weekend. Meaning, there could very well be counter-examples to this hypothesis that prove me wrong, and because I’m busy, I haven’t had the time to think about what they might be. So if you can think of any cases where automation has caused unemployment WITHOUT the automating technology serving as the foundation of a new business model, where robots have directly replaced humans within the same business model, I’d love to hear it.