Four Stories about Technology and Inequality

Economic research is typically conducted with mathematical and statistical models.

But the broad direction of economic research is often determined by verbal narratives, which spin off hypotheses that can be tested for their consistency with logic and data. In that spirit, David Autor offers an essay on “The labor market impacts of technological change: From unbridled enthusiasm to qualified optimism to vast uncertainty” (appearing in An Inclusive Future? Technology, New Dynamics, and Policy Challenges, edited by Zia Qureshi, Brookings Institution, May 2022).

Autor describes four (not necessarily exclusive) narratives that economists have used to discuss the question: “[W]hat is the role of technology–digital or otherwise–in determining wages and shaping wage inequality?” He discusses “four paradigms:” ” the education race, the task polarization model, the automation-reinstatement race, and the era of Artificial
Intelligence uncertainty.” He starts with the narrative that was most common up into the 1990s, and continues to have many loyalists today. Autor writes:

Perhaps the most influential conceptual frame for understanding how technology shapes wage inequality originates with a short article published in 1974 by Dutch economist and Nobel Laureate, Jan Tinbergen. Tinbergen was intrigued by the observation that the wages of Dutch workers with post-high school education (which he called ‘third-level’ education) had been rising over the course of many decades despite vast increases in their supply. This pattern is hard to rationalize in a standard competitive setting since it seemingly implies that the demand curve for skilled labor is upward sloping. To interpret these facts, Tinbergen offered a simple but remarkably powerful analogy. Modern economies
face an ongoing race between the demand for and supply of skill, with technological change propelling the demand curve outward and the educational system racing to push the supply curve outward to match it. In this telling, when the demand curve pulls ahead in the race, inequality between more and less-educated workers—college and non-college workers in the contemporary setting–rises, since more-educated workers are becoming relatively scarce. Conversely, when the supply of college-educated workers surges, as occurred during the 1970s, for example, when American men could defer the Vietnam draft by enrolling in college, earnings inequality between college and non-college workers falls. … [T]echnologically advancing countries must keep raising educational attainment cohort by cohort to keep pace with the moving target of rising skill demands. Or, quoting Lewis Carroll’s Red Queen, “it takes all the running you can do, to keep in the same place.”

As David Autor points out, a relatively simple model capturing this race between a greater demand for skilled labor and supply not quite keeping up with that demand does a pretty good job of explaining the rise in the gap between wages of college graduates and those who have only a high school degree over the last five decades or so.

But as David Autor notes, the “education race” narrative has some unanswered question. Must it be true that any advances in technology will always have this effect of requiring more high-skilled labor? Is it possible that some technology might instead have a a greater benefit for low-skilled labor? In addition, does new technology benefit all broad groups of workers, while benefitting some more than others, or might new technology make some broad categories of workers worse off?

The “task polarization” model seeks to address these questions. The idea is to categorize jobs in an economy according to the skills they need. In particular, this narrative emphasizes that “computers accomplish a distinctive subset of tasks, those involving routine codifiable activities that can be fully described by a set of rules and procedures, encoded in software, and carried out by non-sentient machines.” Notice that we are no longer talking about technology in general–say, past technological advances from transportation to electricitity to chemicals or factories–but are now focused on the specific technological change that seems of greatest immediate relevance in recent decades.

The task polarization framework suggests that computerization will have different effects according to the tasks of workers. Technologies based on computerization will have the effect of “displacing the tasks of the middle-skill workers who in many cases previously provided these information-gathering, organizational, and calculation tasks (e.g., sales workers, office workers, administrative support workers, and assembly line production workers).” However, ‘[t]he productivity and earnings power of workers who specialize in abstract reasoning, expert judgment, and interpersonal interactions and leadership rises as the inputs into their work—information access, analysis, and communication—becomes less expensive and more productive. Thus, computerization increases the productivity of better-educated
workers whose jobs rely on information, calculation, problem-solving, and communication, e.g., doctors, architects, researchers, and stock analysts.”

For a third group of workers, computerization has little effect:

However, not all tasks that are hard to automate would be classified as high-skill tasks. Tasks such as waiting tables, cleaning rooms, picking and boxing items, or assisting elderly people to perform acts of daily living, require dexterity, sightedness, simple communications, and common sense, all of which draw on substantial reservoirs of tacit knowledge. Such tasks are commonly found in personal services jobs, e.g., food service, cleaning, security, entertainment, recreation, and personal care. Computerization has generally not substituted for workers in performing such jobs. But neither has it strongly complemented them. Rather, it leaves this work largely untouched, neither automating the central tasks of this job nor augmenting the workers doing it. Moreover, because a large fraction of adults can, with modest training, perform the core tasks of many non-routine manual jobs, such jobs will generally not pay high wages even when demand is rising, except when the labor market is very tight (as is currently the case).

As David Autor points out, an array of empirical research in labor markets of high income countries support the overall prediction of the task-based model that computerization technologies will tend to have a polarizing effect on the income distribution: benefit those with high school levels, injure the job and wage prospects of those with intermediate skill levels, and have little effect on lower-paid workers. As one example from the US labor market, Autor notes: “Acemoglu and Restrepo (2021) estimate that 50 to 70 percent of the increase in earnings inequality between education, sex, race, and age groups during 1980 through 2016—and the entirety of the fall in real wages of men without high school—are due to the adverse effects of automation on worker groups that were initially more specialized in routine task-intensive work.”

Notice that the task-based approach does not contradict the education race approach, but instead digs down into the particular effects of computerization. But the task-based approach continues to leave some questions unanswered. Must technology have this kind of effect on wage inequality, or is this a peculiarity of computerization technology in particular. Also, the task-based approach in its simplest form seems to argue that the tasks of a given job or worker are fixed and unchanging, when we all know that tasks in a given job and indeed the jobs themselves can evolve over time. One cannot answer the question of whether technology affects wages without some insight into these new jobs.

David Autor offers a vivid illustration of how jobs evolve over time. The blue bars show how the US workforce was divided up by sector in 1940. The neighboring green/pink bars show how the US workforce was divided by sector in 2018, with the green bar showing how many people were doing the same job categories as in 1940, and the pink bar showing how many people were doing jobs in categories that did not even exist in 1940. Thus, you can see that the share of US workers in farming/mining fell from about 18% to less than 2% during this time period–and half of the workers in farming/mining jobs in 2018 didn’t have jobs that didn’t exist in 1940.

As David Autor is quick to acknowledge, economists don’t have great theories of how certain new job categories are created rather than others. But we can say something about past patterns:

Autor et al. (2021b) estimate that more than 60 percent of employment in 2018 was found in job titles that did not exist in 1940 … The introduction of new work, however, is not uniform across skill groups. Between 1940 and 1980, most new work that employed non-college workers was found in construction, transportation, production, clerical, and sales jobs–which are squarely middle-skill occupations. In the subsequent four decades (1980–2018), however, the locus of new work creation for non-college workers shifted away from these middle-tier occupations and towards traditionally lower-paid personal services. Conversely, new work creation employing college-educated workers became increasingly concentrated in professional, technical, and managerial occupations. In combination, these patterns indicate that new work creation has polarized, mirroring (and in part driving) the aggregate polarization of employment …

Again, it is an open question whether these patterns of new work and tasks must follow this pattern moving forward, or whether it might be possible for new work and tasks to focus more on middle-skill occupations.

The last of David Autor’s four paradigms is “the present era of artificial Intelligence uncertainty,” and at this stage, it’s more about questions than answers. Instead of substituting for routine tasks, like computerization, the emerging artificial intelligence technologies may be able to replace certain kinds of expert judgement. Remember, artificial intelligence doesn’t need to be perfect at these tasks to be useful: it just needs to be more consistent or accurate than at6 least some of the humans currently doing these tasks.

At present, discussions of artificial intelligence tend to rely on “perhaps” and “possibly.” For example, one possibility is that middle-skill workers, equipped with artificial intelligence, can be empowered to become more productive, while some high-skill workers will find the value of their knowledge and expertise will be eroded. Perhaps artificial intelligence will substitute for many workers and the middle- and high-skill level, but will have a hard time substituting for low-skill personal service jobs, and thus will make those workers relatively better off. It is quite unclear what kinds of new jobs and tasks might be facilitated and encouraged in an economy with greatly improved artificial intelligence capabilities.

David Autor summarizes his thinking in this way:

What these observations imply is that the work of the future is not an empty set—not even remotely. In Autor et al. (2022), we write that “No compelling historical or contemporary evidence suggests that technological advances are driving us toward a jobless future. On the contrary, we anticipate that in the next two decades, industrialized countries will have more job openings than workers to fill them, and that robotics and automation will play an increasingly crucial role in closing these gaps. Nevertheless, the impact of robotics and automation on workers will not be benign. These technologies, in concert with economic incentives, policy choices, and institutional forces, will alter the set of jobs available and the skills they demand.” It is that adaptation that creates both challenge and opportunity. The problem that industrialized countries face in the immediate decades ahead is not a shortfall in the quantity of jobs. It is rather that many of the jobs may be of low quality, use only generic human capacities, and provide little opportunity for skills acquisition, specialization, and rising lifecycle productivity. This is not a new problem, however. It has been unfolding over four decades. And in general, the U.S. has adapted to it poorly.