Artificial Intelligence, Automation, and the Economy
Technology is not destiny; economic incentives and public policy can play a significant role in shaping the direction and effects of technological change. Given appropriate attention and the right policy and institutional responses, advanced automation can be compatible with productivity, high levels of employment, and more broadly shared prosperity. In the past, the U.S. economy has adapted to new production patterns and maintained high levels of employment alongside rising productivity as more productive workers have had more incentive to work and more highly paid workers have spent more, supporting this work. But, some shocks have left a growing share of workers out of the labor force. The report on Artificial Intelligence, Automation, and the Economy advocates strategies to educate and prepare new workers to enter the workforce, cushion workers who lose jobs, keep them attached to the labor force, and combat inequality. Most of these strategies would be important regardless of AI-driven automation, but all take on even greater importance to the degree that AI is making major changes to the economy.
Accelerating artificial intelligence (AI) capabilities will enable automation of some tasks that have long required human labor. These transformations will open up new opportunities for individuals, the economy, and society, but they have the potential to disrupt the current livelihoods of millions of Americans. Whether AI leads to unemployment and increases in inequality over the long-run depends not only on the technology itself but also on the institutions and policies that are in place. This report examines the expected impact of AI-driven automation on the economy, and describes broad strategies that could increase the benefits of AI and mitigate its costs.
Economics of AI-Driven Automation
Technological progress is the main driver of growth of GDP per capita, allowing output to increase faster than labor and capital. One of the main ways that technology increases productivity is by decreasing the number of labor hours needed to create a unit of output. Labor productivity increases generally translate into increases in average wages, giving workers the opportunity to cut back on work hours and to afford more goods and services. Living standards and leisure hours could both increase, although to the degree that inequality increases—as it has in recent decades—it offsets some of those gains.
AI should be welcomed for its potential economic benefits. Those economic benefits, however, will not necessarily be evenly distributed across society. For example, the 19th century was characterized by technological change that raised the productivity of lower-skilled workers relative to that of higher-skilled workers. Highly-skilled artisans who controlled and executed full production processes saw their livelihoods threatened by the rise of mass production technologies. Ultimately, many skilled crafts were replaced by the combination of machines and lower-skilled labor. Output per hour rose while inequality declined, driving up average living standards, but the labor of some high-skill workers was no longer as valuable in the market.
In contrast, technological change tended to work in a different direction throughout the late 20th century. The advent of computers and the Internet raised the relative productivity of higher-skilled workers. Routine-intensive occupations that focused on predictable, easily-programmable tasks—such as switchboard operators, filing clerks, travel agents, and assembly line workers—were particularly vulnerable to replacement by new technologies. Some occupations were virtually eliminated and demand for others reduced. Research suggests that technological innovation over this period increased the productivity of those engaged in abstract thinking, creative tasks, and problem-solving and was therefore at least partially responsible for the substantial growth in jobs employing such traits. Shifting demand towards more skilled labor raised the relative pay of this group, contributing to rising inequality. At the same time, a slowdown in the rate of improvement in education, and institutional changes such as the reduction in unionization and decline in the minimum wage, also contributed to inequality—underscoring that technological changes do not uniquely determine outcomes.
Today, it may be challenging to predict exactly which jobs will be most immediately affected by AI-driven automation. Because AI is not a single technology, but rather a collection of technologies that are applied to specific tasks, the effects of AI will be felt unevenly through the economy. Some tasks will be more easily automated than others, and some jobs will be affected more than others—both negatively and positively. Some jobs may be automated away, while for others, AI-driven automation will make many workers more productive and increase demand for certain skills. Finally, new jobs are likely to be directly created in areas such as the development and supervision of AI as well as indirectly created in a range of areas throughout the economy as higher incomes lead to expanded demand.
Recent research suggests that the effects of AI on the labor market in the near term will continue the trend that computerization and communication innovations have driven in recent decades. Researchers’ estimates on the scale of threatened jobs over the next decade or two range from 9 to 47 percent. For context, every 3 months about 6 percent of jobs in the economy are destroyed by shrinking or closing businesses, while a slightly larger percentage of jobs are added—resulting in rising employment and a roughly constant unemployment rate. The economy has repeatedly proven itself capable of handling this scale of change, although it would depend on how rapidly the changes happen and how concentrated the losses are in specific occupations that are hard to shift from.
Research consistently finds that the jobs that are threatened by automation are highly concentrated among lower-paid, lower-skilled, and less-educated workers. This means that automation will continue to put downward pressure on demand for this group, putting downward pressure on wages and upward pressure on inequality. In the longer-run, there may be different or larger effects. One possibility is superstar-biased technological change, where the benefits of technology accrue to an even smaller portion of society than just highly-skilled workers. The winner-take-most nature of information technology markets means that only a few may come to dominate markets. If labor productivity increases do not translate into wage increases, then the large economic gains brought about by AI could accrue to a select few. Instead of broadly shared prosperity for workers and consumers, this might push towards reduced competition and increased wealth inequality.
Historically and across countries, however, there has been a strong relationship between productivity and wages—and with more AI the most plausible outcome will be a combination of higher wages and more opportunities for leisure for a wide range of workers. But the degree that this materializes depends not just on the nature of technological change but importantly on the policy and institutional choices that are made about how to prepare workers for AI and to handle its impacts on the labor market.
Responding to the economic effects of AI-driven automation will be a significant policy challenge for the next Administration and its successors. AI has already begun to transform the American workplace, change the types of jobs available, and reshape the skills that workers need in order to thrive. All Americans should have the opportunity to participate in addressing these challenges, whether as students, workers, managers, technical leaders, or simply as citizens with a voice in the policy debate.
AI raises many new policy questions, which should be continued topics for discussion and consideration by future Administrations, Congress, the private sector, academia, and the public. Continued engagement among government, industry, technical and policy experts, and the public should play an important role in moving the Nation toward policies that create broadly shared prosperity, unlock the creative potential of American companies and workers, and ensure America’s continued leadership in the creation and use of AI.
Accelerating AI capabilities will enable automation of some tasks that have long required human labor. Rather than relying on closely-tailored rules explicitly crafted by programmers, modern AI programs can learn from patterns in whatever data they encounter and develop their own rules for how to interpret new information. This means that AI can solve problems and learn with very little human input. In addition, advances in robotics are expanding machines’ abilities to interact with and shape the physical world. Combined, AI and robotics will give rise to smarter machines that can perform more sophisticated functions than ever before and erode some of the advantages that humans have exercised. This will permit automation of many tasks now performed by human workers and could change the shape of the labor market and human activity.
Critically, technology alone will not determine the economic outcomes in terms of growth, inequality or employment. The advanced economies all have had access to similar levels of technology but have had very different outcomes along all of these dimensions because they have had different institutions and policies. But understanding the technological forces is critical to shaping the continued evolution of these policies.
At times, new technologies have raised the productivity and increased employment opportunities for workers with little education, and other times for workers with more. To illustrate the diversity of potential impacts and provide a framework for understanding today, this section discusses historical examples of how innovations affected workers in different ways.
The 19th century was characterized by technological change that raised the productivity of lower-skilled workers and reduced the relative productivity of certain higher-skilled workers. This kind of innovation has been called unskill-biased technical change. Highly-skilled artisans who controlled and executed full production processes saw their livelihoods threatened by the rise of mass production technologies that used assembly lines with interchangeable parts and lower-skilled workers. In reaction, some English textile weavers participated in the Luddite Riots of the early 1800s by destroying looms and machinery that threatened to undercut their highly-skilled, highly-paid jobs with lower-wage roles. Ultimately, the protesters’ fears came true, and many skilled crafts were replaced by the combination of machines and lower-skill labor. There were also new opportunities for less-skilled workers and output per hour rose. As a result, average living standards could rise, but certain high-skill workers were no longer as valuable in the market.
Technological change tended to work in a different direction throughout the late 20th century. The advent of computers and the internet raised the relative productivity of higher-skilled workers, an example of skill-biased technical change. Routine-intensive occupations that focused on predictable, easily-programmable tasks—such as switchboard operators, filing clerks, travel agents, and assembly line workers—have been particularly vulnerable to replacement by new technologies. Some entire occupations were virtually eliminated and demand for others reduced. In routine jobs is largely responsible for recent low labor demand for less educated workers.
Like these past waves of technological advancements, AI-driven automation is setting off labor-market disruption and adjustment. Economic theory suggests that there must be gains from innovations, or they would not be adopted. Market forces alone, however, will not ensure that the financial benefits from innovations are broadly shared.
Today, it may be challenging to predict exactly which jobs will be most immediately affected by AI-driven automation. Because AI is not a single technology, but rather a collection of technologies that are applied to specific tasks, the effects of AI will be felt unevenly through the economy. Some work tasks will be more easily automated than others, and some jobs will be affected more than others.
Some specific predictions are possible based on the current trajectory of AI technology. For example, driving jobs and housecleaning jobs are both jobs that require relatively less education to perform. Advancements in computer vision and related technologies have made the feasibility of fully automated vehicles (AVs), which do not require a human driver, appear more likely, potentially displacing some workers in driving-dominant professions. AVs rely upon, among other things, capabilities of navigating complex environments, analyzing dynamic surroundings, and optimization. Seemingly similar capabilities are required of a household-cleaning robot, for which the operational mandate is less specific (i.e. “clean the house,” as opposed to the objective of navigating to a specific destination while following a set of given rules and preserving safety). And yet the technology that would enable a robot to navigate and clean a space as effectively as a human counterpart appears farther off. In the near to medium term, at least, drivers will probably be impacted more by automation than will housecleaners.
Nevertheless, humans still maintain a comparative advantage over AI and robotics in many areas. While AI detects patterns and creates predictions, it still cannot replicate social or general intelligence, creativity, or human judgment. Of course, many of the occupations that use these types of skills are high-skilled occupations, and likely require higher levels of education. Further, given the current dexterity limits of the robotics that would be needed to implement mass AI-driven automation, occupations that require manual dexterity will also likely remain in demand in the near term.
There are four categories of jobs that might experience direct AI-driven growth in the future. Employment in areas where humans engage with existing AI technologies, develop new AI technologies, supervise AI technologies in practice, and facilitate societal shifts that accompany new AI technologies will likely grow. Current limits on manual dexterity of robots and constraints on generative intelligence and creativity of AI technologies likely mean that employment requiring manual dexterity, creativity, social interactions and intelligence, and general knowledge will thrive. Below are descriptions and potential examples of future employment for each category.
Engagement. Humans will likely be needed to actively engage with AI technologies throughout the process of completing a task. Many industry professionals refer to a large swath of AI technologies as “Augmented Intelligence,” stressing the technology’s role as assisting and expanding the productivity of individuals rather than replacing human work. Thus, based on the biased-technical change framework, demand for labor will likely increase the most in the areas where humans complement AI-automation technologies. For example, AI technology such as IBM’s Watson may improve early detection of some cancers or other illnesses, but a human healthcare professional is needed to work with patients to understand and translate patients’ symptoms, inform patients of treatment options, and guide patients through treatment plans. Shipping companies may also partner workers who pickup and deliver goods over the last 100 feet with AI-enabled autonomous vehicles that move workers efficiently from site to site. In such cases, AI augments what a human is able to do and allows individuals to either be more effective in their specialty task or to operate on a larger scale.
Development. In the initial stages of AI, development jobs are crucial and span multiple industries and skill levels. Most intuitively, there may be a great need for highly-skilled software developers and engineers to put these capacities into practical use in the world. To a certain extent, however, AI is only as good as the data behind it, so there will likely be increased demand for jobs in generating, collecting, and managing relevant data to feed into AI training processes. Applications of AI can range from high-skill tasks such as recognizing cancer in x-ray images to lower-skill tasks such as recognizing text in images. Finally, to an increasing degree, development may include those specializing in the liberal arts and social sciences, such as philosophers with frameworks for ethical evaluations and sociologists investigating the impact of technology on specific populations, who can give input as the new technologies grapple with more social complexities and moral dilemmas.
Supervision. This category encompasses all roles related to the monitoring, licensing, and repair of AI. For example, after the automated vehicle development phase, the need for human registration and testing of such technology to ensure safety and quality control on the roads will still likely exist. As a widespread new technology, AV will require regular repair and maintenance, which may expand mechanic and technician jobs in this space as well. Real-time supervision will also be required in exceptional, marginal, or high-stakes cases, especially those involving morality, ethics, and social intelligence that AI may lack. This might take the form of quality control of recommendations made by AI or online moderation when sensitive subjects are discussed. The capacity for AI-enabled machines to learn is one of the most exciting aspects of the technology, but it may also require supervision to ensure that AI does not diverge from originally intended uses. As machines get smarter and have improved ability to make practical predictions about the environment, the value of human judgment will increase because it will be the preferred way to resolve competing priorities.
Response to Paradigm Shifts. The technological innovation surrounding AI will likely reshape features of built environment. In the case of AVs, dramatic shifts in the design of infrastructure and traffic laws—which are currently built with the safety and convenience of human drivers in mind—may be needed. The advent of self-driving cars may result in higher demand for urban planners and designers to create a new blueprint for the way the everyday travel landscape is built and used. Paradigm shifts in adjacent fields such as cybersecurity—demanding, for instance, new methods of detecting fraudulent transactions and messages—may also necessitate new occupations and more employment.
AI-driven automation stands to transform the economy over the coming years and decades. The challenge for policymakers will be to update, strengthen, and adapt policies to respond to the economic effects of AI.
Although it is difficult to predict these economic effects precisely with a high degree of confidence, the economic analysis in the previous chapter suggests that policymakers should prepare for five primary economic effects:
There is substantial uncertainty about how strongly these effects will be felt, and how rapidly they will arrive. It is possible that AI will not have large, new effects on the economy, such that the coming years are subject to the same basic workforce trends seen in recent decades—some which are positive, and others which are worrisome and may require policy changes. At the other end of the range of possibilities, the economy might potentially experience a larger shock, with accelerating changes in the job market, and significantly more workers in need of assistance and retraining as their skills are no longer valued in the job market. Given presently available evidence, it is not possible to make specific predictions, so policymakers must be prepared for a range of potential outcomes. At a minimum, some occupations such as drivers and cashiers are likely to face displacement from or restructuring of their current jobs, leading millions of Americans to experience economic hardship in the short-run absent new policies.
Because the effects of AI-driven automation will likely be felt across the whole economy, and the areas of greatest impact may be difficult to predict, policy responses must be targeted to the whole economy. In addition, the economic effects of AI-driven automation may be difficult to separate from those of other factors such as other technological changes, globalization, reduction in market competition and worker bargaining power, and the effects of past public policy choices. Even if it is not possible to determine how much of the current transformation of the economy is caused by each of these factors, the policy challenges raised by the disruptions remain, and require a broad policy response.
(Link: https://www.whitehouse.gov/sites/whitehouse.gov/files/images/EMBARGOED%20AI%20Economy%20Report.pdf)