Given that technologies have been automating human work over the past couple of centuries, — why hasn’t automation already wiped out a majority of jobs? Why are there still so many jobs left? The answer isn’t very complicated, although frequently overlooked, explained MIT economist David Autor in a 2015 article “Why Are There Still So Many Jobs? The History and Future of Workplace Automation.”
The answer, he noted, underlines a fundamental economic reality: “tasks that cannot be substituted by automation are generally complemented by it.” Automation does indeed substitute for labor. However, automation also complements labor, raising economic outputs in ways that often lead to higher demand for workers.
Most jobs involve a number of tasks or processes. Some of these tasks are more amenable to automation, while others require judgement, social skills and other human capabilities. But just because some of the tasks have been automated, does not imply that the whole job has disappeared. To the contrary, automating the more routine parts of a job will often increase the productivity and quality of workers by complementing their skills with machines and computers, as well as enabling them to focus on those aspect of the job that most need their attention.
Articles often overstate the extent of machine substitution for human labor “and ignore the strong complementarities between automation and labor that increase productivity, raise earnings, and augment demand for labor.”
But,”when job tasks are automated, does this augment or diminish the value of labor in the tasks that remain?,” asked professor Autor and MIT research scientist Neil Thompson in “Does Automation Replace Experts or Complement Expertise? The Answer is Yes,” a lecture given at the European Economic Association in August of 2024, whose content was subsequently published in “Expertise.”
The answer depends not on how many tasks have been automated, but on which tasks have been automated, that is, how automation changes the expertise required to perform the remaining non-automated tasks. Since the same task may be relatively expert in one occupation and inexpert in another, automation of any given task can reduce the required expertise in some occupations and increase it in others.
Author and Thompson propose a model of occupation expertise that helps predict the impact of changing expertise requirements on wages and employment:
- If mostly higher-skill task are automated: the overall skill requirements for the occupation are lower; the set of qualified workers is higher; and, as a result, the wages are lower.
- If mostly lower-skill tasks are automated: the overall skill requirements for the occupation are higher; the set of qualified workers is lower; and as a result, the wages are higher.
In other words, changing expertise requirements have countervailing effects on wages and employment:
- Automation that raises expertise requirements raises wages but reduces the set of qualified workers.
- Automation that decreases expertise requirements reduces wages but increases the number of qualified, less expert workers.
The authors illustrate their model of occupational expertise with two concrete examples.
Taxi drivers: employment rose; expertise and wages decreased. Taxis, limousines, and other vehicles for hire used to be regulated, their numbers were limited by the cities in which they operated, and the drivers had to obtain a chauffeur license. But, with the advent of Uber and similar ride-hailing e-taxis in the 2010s, the industry has essentially been deregulated. Anyone can now become an e-taxi driver with no need for a taxi medallion or chauffeur license, using their own private car and mobile devices. As a result, the number of taxi drivers has significantly increased, navigation apps have reduced the required expertise, and wages have fallen.
Proofreaders: expertise upgraded, wages rose, employment fell. The main job of a proofreader was to compare a manuscript with a galley proof looking for and correcting human errors during the typesetting phase of publishing, a relatively low skill job that was no longer necessary with the advent of digital proofs. Currently, proofreaders are akin to copy editors, working with the manuscript authors to improve the overall structure of their work, including grammar, spelling, punctuation, and syntax. The expertise now required of proofreaders is significantly higher, leading to lower qualified candidates and higher wages.
Let me summarize the key concepts introduced in the paper.
The Expertise Model of Automation
- An occupation is composed of multiple tasks; automating one set of tasks does not eliminate the need to perform all the others.
- Some tasks require a greater or lesser degree of task-specific expertise and are subject to automation.
- Other tasks are generic and are not subject to automation because they only require core human skills that everyone has like basic common sense and physical dexterity.
- Workers have different levels of expertise; a high-expertise worker can do tasks requiring lower expertise, but a low expertise worker cannot do tasks requiring higher expertise.
- Task-specific expertise commands a wage premium but also serves as a barrier to entry because workers lacking the necessary expertise cannot enter an occupation that requires it.
- Automating a set of tasks in a job shifts the composition of the remaining occupation, including expertise requirements, number of qualified workers, and wages.
The Expertise Measurement Challenge
In order to deploy their model in the real world, the authors needed to come up with a way to measure the expertise required for different tasks that doesn’t rely on subjective judgements and would allow them to quantify the change in the expertise requirements of occupations due to task removal and addition.
To do so, they came up with a novel measure of linguistic complexity as a proxy for task expertise based on the Efficient Coding Hypothesis (ECH), a concept first proposed in 1961 by neuroscientist Horace Barlow. The ECH implies that we can identify words that are primarily used in high skill domains like engineering or medicine to facilitate communications but are infrequency used in everyday, common language. Tasks described with rare or complex words are more likely to require expertise, whereas tasks described using common words are likely to be more routine or inexpert. The ECH has been verified by applying it in almost one thousand languages.
Key Findings
To empirically verify their expertise model of automation, Autor and Thompson identified the job tasks that have been removed from and added to each occupation between 1977 and 2018 by compared their job descriptions in the 1977 Dictionary of Occupational Titles with their job description in the 2018 O*NET data base. They were thus able to measure the evolving expertise requirements of each occupation by comparing the expertise of tasks added and removed in the intervening four decades. Their empirical analysis strongly supports the predictions of the expertise model of automation: automation both replaces and augments expertise.
“Analyzing data on employment and earnings by occupation over four decades, we show that changes in occupational expertise, stemming from both the removal and addition of occupational tasks, strongly predict changes in occupational wages,” wrote Autor and Thompson in conclusion. “Moreover, the expertise requirements of tasks removed from or added to an occupation affect wage levels independently of the quantity of tasks added or removed present. Remarkably, both the removal of expert tasks the addition of inexpert tasks predict relative wage declines in an occupation, while, conversely both the removal of inexpert tasks and the addition of expert tasks predict occupational wage gains.”
“Our model makes the counterintuitive prediction that occupations with increasing expertise requirements see falling employment (alongside rising wages), while occupations with declining expertise requirements see rising employment alongside falling wages. The data robustly confirm this prediction. Crucially, we find the opposite pattern for changes in task quantities. Occupations that gain tasks expand and those that lose tasks contract. This is also opposite to the pattern for wages, where increases in both task quantities and task expertise predict wage increases.”
Finally, let me share this video of a seminar by professor Autor given earlier this year on this topic at the Standford Digital Economy Lab.
