1

Introduction

GOALS FOR THIS REPORT

In 2017, the National Academies of Sciences, Engineering, and Medicine published a comprehensive report exploring the landscape of artificial intelligence (AI) and its implications for work and the workforce.¹ Since then, the effects of AI have expanded at an unprecedented rate, permeating various facets of daily life and significantly altering the workforce terrain. In light of this rapid evolution, the mandate for this follow-on report is clear: to assess the “current and future impact of artificial intelligence on the workforce of the United States across sectors.”² This undertaking is not just an update but a reconceptualization, accounting for the leaps in technology and the consequent ripples throughout the labor market and wider economy.

The charge to the study committee was to focus specifically on the economic, productivity, and workforce dimensions of AI. It is important to acknowledge that AI’s effects are by no means limited to these areas—it has profound implications for democracy, geopolitics, national security, scientific progress, and mental health, among others. Although these spheres are undoubtedly significant, they fall outside the scope of this report. Consequently, this report will concentrate on changes in the technology and capabilities of AI, its adoption and productivity effects, interactions between AI and the

___________________

¹ National Academies of Sciences, Engineering, and Medicine, 2017, Information Technology and the U.S. Workforce: Where Are We and Where Do We Go from Here? The National Academies Press, https://doi.org/10.17226/24649.

² 2021 National Defense Authorization Act, P.L. 116-283, section 5101.

workforce, the implications for education and skill requirements, and the measurement challenges and opportunities in various economic sectors.

WHY IS THIS TOPIC IMPORTANT NOW?

Today, the speed of technological progress is reshaping not just the tools but also the fabric of the workforce and societal structures. AI has emerged as a general-purpose technology with sweeping implications that demand immediate attention and thoughtful analysis. AI stands out among general-purpose technologies owing to its core attribute—a focus on intelligence. This arguably makes it the most general of all general-purpose technologies.

The progression of AI has reached an inflection point with the rise of foundation models such as large language models (LLMs) and multimodal systems, which have begun to be integrated rapidly with a multitude of other technological tools, augmenting their capabilities and applications across industries. The capabilities of these systems have sparked not just excitement but also genuine surprise, leading to their emergent and swift adoption across various sectors.

Although AI has garnered more than its share of hype, the enthusiasm surrounding AI is not misplaced, nor is it purely conjectural. Policy makers, executives, and industry leaders are rightfully eager to understand these advances, as the implications are multifaceted, impacting productivity, the workforce, education, and society at large. The transformative effects can be seen in multiple domains: software development has already witnessed dramatic productivity gains, and the work of paralegals, customer service agents, and others who summarize documents is already being reshaped by AI. Textual monologues are evolving into interactive dialogues; for example, a book might soon serve as a conversational tutor powered by a sophisticated LLM, providing a new avenue to just-in-time, personalized training for the workforce. Entertainment, finance, health care, education, retail, manufacturing, transportation, and many other industries are poised for transformation. These are but a few instances in the litany of ongoing changes propelled by AI.

In contemplating what the future holds, one must approach predictions with humility, acknowledging the lessons of the recent past. The 2017 report did not grasp the full trajectory of AI’s progress—for instance, emergence and adoption of LLMs outpaced expectations, while the road to fully autonomous vehicles has proved lengthier than anticipated. Although it is easy to overestimate the impact of new technologies in

the short term and underestimate it in the long term,³ AI is sure to continue to advance and catalyze change; what remains uncertain is the precise nature and timing of these capabilities.

The magnitude of AI’s impact should be distinguished from the immediacy of that impact. Most general-purpose technologies have historically taken considerable time to integrate fully into society, often owing to the need for intangible complements such as new skills, altered business processes, and co-invention. AI, however, is displaying characteristics that suggest a more accelerated trajectory. The uptake of products like ChatGPT, which soared to reach 100 million users in mere months,⁴ suggests an appetite and readiness for rapid adoption comparable to or greater than that for smartphones, which now connect more than two-thirds of the global population.

This swift integration of AI is facilitated by its connectivity to platforms, application programming interfaces, and the cloud, alongside plugins that incorporate capabilities of complementary software and overarching software architectures such as LangChain and AutoGPT that employ LLMs as subroutines. It is becoming increasingly clear that AI, much like the Internet, is not simply a tool but also a platform upon which numerous other innovations can be built, adopted, and diffused at a remarkable pace.

The trajectories that AI-enabled futures might take can lead to outcomes of profound benefit or significant disruption. The goal of this report is thus twofold: to responsibly inform about the current state and capabilities of AI as they relate to the workforce and to offer insights that prepare us for the challenges ahead and opportunities that will arise. It also considers how AI is likely to augment human labor, reshape job markets, and influence workforce dynamics.

The future is not preordained; individuals, businesses, nonprofits, colleges and universities, civil society institutions, and government influence it by the choices they make every day, large and small. This moment presents the opportunity to ensure that the awakening of AI augments collective capabilities, enhances human well-being, and constructs a future workforce that is resilient, adaptive, and equipped to meet the challenges of the 21st century.

___________________

³ This observation is often attributed to Roy Amara. See S. Ratcliffe, ed., 2016, “Roy Amara 1925–2007, American Futurologist,” Oxford Essential Quotations, Vol. 1 (4th ed.), Oxford University Press, https://doi.org/10.1093/acref/9780191826719.001.0001.

⁴ J. Porter, 2023, “ChatGPT Continues to Be One of the Fastest-Growing Services Ever,” The Verge, November 6, https://www.theverge.com/2023/11/6/23948386/chatgpt-active-user-count-openai-developer-conference.

HOW TO THINK ABOUT ARTIFICIAL INTELLIGENCE AND ITS IMPACT ON THE WORKFORCE

It is helpful to begin by considering the question of how to think about AI and its future impact on the workforce. Drawing on presentations to the committee (see Appendix B), the study committee formed a set of key assumptions for how to think about AI and the workforce. These assumptions, which also help motivate the committee’s formulation of the topics examined in this report, are as follows:

• AI technology is at an inflection point where recent advances promise to have significant impacts on many parts of the workforce. Technical progress in AI is currently largely driven by recent progress in LLMs such as those that underlie ChatGPT, which was introduced in November 2022. Unlike earlier AI systems, LLMs show novel abilities—for example, to write useful computer software, to pass a variety of graduate exams, and to communicate fluently in many languages. Although today’s AI systems still remain imperfect in many ways (e.g., they can “hallucinate” incorrect answers to factual questions, and they can exhibit biases), many AI experts expect this accelerated rate of progress to continue for some time. Chapter 2 examines the current state of AI, including LLMs; where AI might be headed; and factors that could accelerate and decelerate the current rate of progress.
• Productivity growth is mainly driven by improved technology. Productivity growth, the increase in the amount of output per unit input, is the key to higher living standards. It is mainly a function of improved technologies, especially general-purpose technologies, that affect many sections of the economy, improve rapidly, and catalyze complementary innovations. AI, which seeks to augment intelligence itself, has all the characteristics of an important general-purpose technology.
• Jobs can be thought of as bundles of tasks. One way AI will tend to impact jobs is through its impacts on individual tasks. For example, the job of a physician includes tasks such as (a) diagnosing the patient, (b) generating potential therapies given the diagnosis, and (c) having a conversation with the patient to explain therapy options and jointly choose the way forward. Any given AI system might impact one of these tasks without an equally significant impact on other tasks—for example, an AI system to suggest likely diagnoses might primarily impact the first of these tasks. The impact of AI will not be limited, however, solely to replacement or augmentation of tasks within existing occupations. Widespread adoption of AI may, for example, catalyze fundamental

• changes in the structure of jobs and industries, as occurred during the transition from the artisanal to the industrial eras, when many forms of home-based production were ultimately displaced by the factory system. Major transitions of this sort are disruptive to careers and livelihoods. Even if the longer-term consequences are favorable, the transition is likely to be economically (and perhaps societally) destabilizing. Evidence suggests that it took at least five decades for working-class wages to begin rising again after the advent of the Industrial Revolution.⁵
• AI can affect a work task either by automating it to replace a worker or by assisting the worker in performing the task. Which of these occurs is at least partly a design choice, not a preordained outcome. For example, an AI system to analyze medical radiological images might be used to replace a human at this task; alternatively, it might be used to provide a second opinion to the human who remains responsible for the final decision. Businesses and governments can make choices that will influence which of these future outcomes occurs.
• AI improvements to productivity can result in either a decrease or an increase in total employment, depending on economic factors. As worker productivity increases, certainly fewer workers will be needed to produce the same total output, regardless of whether the improved productivity comes from replacing or assisting workers. However, this does not imply that employment will decrease. It is also possible that improved productivity will lead businesses to decrease prices, resulting in increased demand, which may be large enough to require hiring additional employees despite the increase in productivity per unit output. For example, when jet engines replaced propellers in airplanes, pilot productivity (passenger miles flown per hour of pilot work) increased significantly. However, the result was an overall increase in the number of pilots employed, owing to the resulting increase in demand for airline flights. What is more, AI can lead to the creation of new products and services, which in turn increase overall employment. In general, the impact of AI productivity improvements on overall demand for workers involves a complex interaction among a variety of supply, demand, and price elasticities. Chapter 3 examines the relationships among productivity, AI, and economic growth.
• Because AI involves automating and augmenting expertise, it is useful to analyze its impact in terms of what types of human expertise it makes more and less valuable. For example, the adoption of computers over the past decades for

___________________

⁵ J. Mokyr, C. Vickers, and N.L. Ziebarth, 2015, “The History of Technological Anxiety and the Future of Economic Growth: Is This Time Different?” Journal of Economic Perspectives 29(3):31–50. Mokyr was, in turn, drawing on D. Bythell, 1969, The Handloom Weavers: A Study in the English Cotton Industry During the Industrial Revolution, Cambridge University Press.

• non-AI tasks such as e-mail and office work has led to devaluing expertise at carrying out many middle-skilled, routine-intensive office tasks, while increasing the value of the types of expertise associated with advanced graduate degrees. Given the very different profile of recently developed AI capabilities (e.g., the ability to pass the verbal and quantitative reasoning sections of the Graduate Record Examination used for graduate school admissions), what will be AI’s influence on future demand and supply for various types of human expertise? Chapter 4 examines potential AI impacts on the workforce in terms of the types of human expertise that might become in greater or lesser demand.
• Given the shifting nature of jobs created by the impact of AI, it is important to consider how to adapt approaches to education. During this time of great change, many workers will find continuing education and just-in-time training to be useful in improving their job prospects. A new generation of K–12 students and post-secondary students graduating into the workforce may require a different set of skills and training. This raises the questions of what content should be taught, when in a person’s career to teach it (i.e., K–12, college, continuing education), and how to teach (e.g., how can AI lead to improved educational services). Chapter 5 considers what should be taught to whom as well as the potential of AI systems to provide new modes of personally customized education.
• Given the great uncertainties, both about exactly what AI technical advances might occur in the near future and about how they will impact demand for various types of expertise and workers, it is imperative to improve the tracking of technical progress in AI, its adoption in practice, and its workforce impacts in near real time—and to share this information with the workforce. AI technology is undergoing rapid and difficult-to-predict changes, but these changes are very likely to impact the workforce. A conclusion is that it is possible to help the workforce adapt by better understanding and sharing the changes that are actually occurring in real time over the coming months and years. Chapter 6 considers how to measure AI’s progress and impacts, what kinds of data are already being collected, and the opportunities to collect additional data (including from public–private partnerships) to produce a much more informed picture of the evolving state of AI, jobs, and the workforce.