7

Conclusion

The preceding chapters considered how artificial intelligence (AI) and related technologies might impact the workforce from a variety of perspectives, examining the state of AI technology and where it might be headed; its potential impact on productivity; how it might impact the demand and supply for different types of labor expertise; the need for and role of AI in education and workforce retraining; and the need and opportunities for collecting data to measure the changing state of the workforce, demand for different types of expertise, and availability of training opportunities for workers.

This chapter first presents the main findings arising from the analysis in the preceding chapters and then presents the study committee’s conclusions about what levers are available to government leaders to influence the impact of AI on the workforce.

FINDINGS

This section presents the primary findings of this study. These findings, and support for them, are discussed in greater detail in earlier chapters.

___________________

¹ General-purpose technologies like the steam engine and electricity have widespread applications and were thus key drivers of economic growth. As discussed in more detail below, AI is advancing exceptionally rapidly, reflecting several key technical breakthroughs.

Generative AI systems released just over the past year—such as GPT-4, Gemini, and related foundation models—exhibit major new AI capabilities. These include the ability to hold meaningful conversations about diverse topics in dozens of languages, automatically summarize the key points discussed in large text documents, perform a variety of problem-solving tasks, write computer programs, write poetry, generate realistic images to match text specifications, interpret and reason about images, pass the high school Advanced Placement (AP) math exam, and pass the Law School Admission Test at the 88th percentile, which is a higher level than average among humans who take and pass this exam. Still, there is a great deal of uncertainty about AI’s future course, suggesting that wide error bands and a range of contingencies should be considered.

Note that passing a competency test is far from sufficient for having the human capabilities necessary to do a job. Note also that current generative AI systems are subject to error, to manipulation to produce false results, and even to hallucination. They also exhibit biased behavior and fail to reason correctly to reach conclusions from given facts. These systems are constantly being adjusted and improved to address these shortcomings and to erect guardrails to prevent them. Although such efforts have reduced these shortcomings, they have not been eliminated and are likely to persist to some degree for some time to come.

A variety of factors have created the environment for the recent acceleration in AI progress, including the increasing volume of online data to train AI systems, improvements to computer hardware and its computational speed, and improvements in AI algorithms. Over recent years, improvements in these three areas have led to remarkably consistent improvements in performance for AI foundation models. Because further increases in all three components in the next few years can be expected, including perhaps orders of magnitude improvement in computational power, it is likely that AI performance will also increase substantially. In response to this rapid AI progress, increasing investments in AI research and development and a burst of new start-up companies in this area create additional avenues for advances. Already, some of the directions in which the field may progress have emerged, including the appearance

of multimodal models that go beyond text to accommodate images, video, sound, and speech; initial experiments to build more personally customized models by giving them access to one’s e-mail; and incorporation of more traditional software such as calculators, search engines, and route planners into generative AI models to increase their capabilities. Furthermore, creative exploration and development of systems and infrastructure around base models, including software development environments and control methodologies, demonstrate impressive leaps in competency. For example, programming environments now enable the construction of collaborative communities of agents, each placing an LLM in a different role, which enables new forms of complex, extended problem-solving sessions. In parallel with such generative models that operate in the knowledge space of the nonphysical world, progress continues as well in robotics, although at a somewhat slower pace. However, there is a possibility that as generative models become more multimodal, they may lead to rapid advances in robotics as well.

Considering all the factors on which productivity effects depend, there is considerable uncertainty regarding the specific size and timing of the increase in productivity resulting from AI, although some estimates suggest as much as a doubling of the rate of growth in productivity from about 1.4 percent per year in the recent past to 3 percent or more in the coming decade. There are important differences in the exposure of different sectors and occupations to AI, so its aggregate productivity effects will depend on how it affects the productivity of different sectors, different occupations, and different firms, and there is likely to be significant heterogeneity across all categories. There are also significant differences in the exposure of different kinds of worker tasks to generative AI. Generative AI can both substitute for labor and complement labor in these tasks. It can also generate new tasks and new activities for labor. All three of these effects can boost labor productivity over time, but the lags, the barriers, and the costs can be significant. Rapid AI deployment could cause considerable disruption in the labor market if workers must move quickly between tasks and occupations. This disruption could weaken AI’s productivity effects if the necessary labor reallocation does not occur rapidly and if displaced workers are not deployed into new tasks with productivity levels at least as high as those in their previous tasks.

New goods, services, and business models will emerge, and occupations will be significantly reorganized, across firms and industries. Typically, the productivity effects of new technologies can take many years to be realized, but there is reason to believe that the productivity gains of AI may be significantly faster. This reflects the fact that AI systems can often be delivered and implemented on the existing digital infrastructure and information systems built over the past decades. Furthermore, end users do not necessarily need to learn specialized interfacing skills to start using the technology—they can instruct LLMs with English or other natural languages. While further complementary investments can create additional benefits, significant productivity gains have already been documented quite rapidly in some applications, such as customer service and software development. That said, the trajectory of aggregate productivity growth is poorly understood and hence difficult to forecast.

These realities make forecasting labor market implications challenging. However, in attempting to foresee the impact of AI on employment, it is important to recognize that the United States, along with most industrialized countries, will face pronounced labor scarcity over the coming years owing to aging populations, low birth rates, a steeply rising ratio of retirees to workers, and, probably, restricted immigration. Depending on the magnitude of its impact, AI might simply reduce the negative consequences of this labor scarcity. While the net impact on overall employment levels is difficult to forecast, over the next decade AI can be expected to affect substantially the type of work many people do, the skills and expertise required for that work, and the market value of those capabilities. These effects will not simply follow the template of conventional computer-based technologies, which tend to impact routine, codifiable tasks, such as calculation, information storage, information search, and precise repetition of cognitive and physical tasks. In contrast to these routine capabilities, today’s AI can sometimes apply to tasks requiring flexibility, ongoing learning, and a certain element of improvisation—tasks that have largely been beyond the reach of machine capabilities until recent advances in AI. Accordingly, AI tools will likely have the largest near-term consequences for knowledge work and cognitive tasks. A small subset of occupations that could be most affected

include paralegals, customer service agents, software developers, commercial writers, editors, translators, and many managers. This creates a significant risk of worker displacement and skills devaluation. If, hypothetically, AI were to surpass human capabilities rapidly in key tasks required for software development, legal document creation, translation, crafting of business reports and presentations, engineering design, and even drug discovery, this could potentially reduce the earnings capacities of typically highly paid workers who do these tasks for a living—akin to how digital printing displaced skilled typesetters or how electronic telephone switching ultimately extinguished the once-vast telephone operator occupation.

First, AI may usher in new varieties of job tasks (i.e., new work) that demand valuable new expertise. Historically, the contribution of new work has been quantitatively important. As noted previously, research finds that more than 60 percent of the job specialties in which U.S. workers were employed in 2018 did not yet exist in 1940. Research also finds, however, that although technological change displaced workers and created new employment opportunities at about the same rate during the period following World War II through the 1970s, since the 1980s it displaced workers faster than it created new opportunities. A key open question is whether under future AI-driven technological change, the displacement of labor will continue to outpace the creation of new employment opportunities. Second, although AI may displace some forms of expertise, it may complement others, enabling workers to use their expertise more effectively to accomplish a broader variety of valuable tasks, perhaps with less formal training. AI is already lowering barriers to computer software development by serving as a “copilot” for novice (and experienced) programmers to produce code more efficiently, with fewer errors and with potentially less mastery of technical details that can now be handled by AI directly. Looking forward, AI may enable medical professionals such as nurses and nurse practitioners to handle a larger range of diagnostic and treatment tasks with less oversight from doctors. Similarly, workers performing skilled tasks in the trades—plumbing, electrical work, construction, and aircraft maintenance and repair—may be augmented by the guidance and guardrails provided by AI assistants. Although AI-powered robots will not be able to tackle most of these physically dexterous,

flexibility-demanding tasks anytime soon, AI will plausibly enable workers performing these tasks to deploy their expertise more effectively across a larger range of challenges.²

In competitive market conditions, wage growth should reflect productivity growth, but during the past 20–30 years, real wages have grown more slowly than labor productivity in the United States and the other advanced economies. This was true during both the mid-1990s through 2005 period of strong productivity growth as well as the period of slow productivity growth through 2019–2020. In addition, the use of AI to monitor and surveil worker performance to achieve additional labor productivity may erode or enhance job quality, worker satisfaction, and worker commitment.

Changes in jobs and the expertise they require will increase the demand for continuing education and training. At the same time, AI may lead to a reassessment of what should be taught in primary and secondary schools. If AI makes it easier to access diverse factual knowledge, students might benefit from more emphasis on learning the reasoning skills needed to combine factual knowledge to make final decisions and on assessing the believability of seemingly inconsistent facts. Although computer programming is a widely taught skill today, AI coding tools such as GitHub Copilot are already changing the nature of programming. Some code can now be written purely in natural language rather than a programming language; hence, teaching the skill of coding is likely to take

___________________

² A further possibility is that AI capabilities will help to shift the structure of comparative advantage among nations. AI could, for example, enable the Global North to reduce its dependence on the Global South for labor-intensive service tasks, such as customer support, software development, and back-office operations. Alternatively, AI may increase the ease of outsourcing such tasks to the Global South while boosting the quality of services delivered. Simultaneously, it may enable many more low-income countries to compete in these tasks as machine translation erodes long-standing language barriers. Alternatively, by providing inexpensive access to expertise in medicine, engineering, science, and software development, AI tools may enable some low-income countries to tackle challenges indigenously that would previously have required importing foreign expertise. Because the range of possibilities is vast and multivalent, the committee does not venture an expert opinion on the question of AI and international comparative advantage.

a different form in the future. In addition to reshaping the demand for education, AI is likely to help in the supply of better education. Already there are multiple AI-enhanced online education platforms in use by millions of K–12 students as well as adult continuing education students. Several of these systems use earlier AI methods to customize instruction to individual students and use machine learning to discover which teaching tactics work best for different students. Furthermore, the burst of recent new progress in LLMs has led to a corresponding burst of exploration into new, more powerful conversational approaches to online education that customize even more adeptly to the needs of individual students. Although these new approaches are yet to be proven in practice, there are reasons to be optimistic about the prospects for AI to improve the delivery of education through personal customization.

For years, government agencies such as the U.S. Bureau of Labor Statistics have collected and published data on topics including unemployment rates and numbers of new jobs, and in recent years the types of data available have increased. Nevertheless, these sources are far from providing the kind of real-time picture of the nature and pace of the adoption of AI in the workplace, the changing supply of and demand for different workforce skills, the wages being paid for each skill, continuing education opportunities available to acquire these skills, and where these opportunities are geographically, which would truly support workers seeking to improve their careers. Importantly, much of these data exist already, although they are in the hands of corporations such as LinkedIn, Indeed, Workday, and ADP. Government agencies would have difficulty collecting the volume and quality of real-time workforce data currently held by such corporations. However, new approaches to public–private data partnerships that allow the government to publish to the workforce valuable summaries of these data could make a great difference in the ability of the workforce to adapt to the unpredictable times that lie ahead. As AI expands further, there should be continued tracking of the nature of job displacement, jobs creation, and job transformation, and the implications for employment, education, career development, and the national economy.

These concerns have already led to initial government regulations around AI development and use, and additional future governance initiatives appear likely. Such initiatives to govern the use of AI will also modulate the rate and extent of AI adoption, and its impact on the economy and the workforce.

OPPORTUNITIES TO INFLUENCE HOW ARTIFICIAL INTELLIGENCE WILL IMPACT THE WORKFORCE

As AI technology advances and is broadly adopted by industry, there are numerous opportunities for businesses, nonprofit institutions, worker organizations, colleges and universities, civil society institutions, and government to influence the direction of this development. The impact of AI on the workforce is not preordained but will instead be influenced by what different institutions throughout society choose to do to guide AI’s development and use. The following discussion breaks down these opportunities into the following types:

• Opportunities to influence the rates and direction of the development of AI;
• Opportunities to speed and share the productivity benefits of AI;
• Opportunities to influence the balance among worker augmentation, new work creation, and labor-displacing automation;
• Opportunities to understand the implications of AI for education and assist workers with retraining and continuing education;
• Opportunities for more exact and timely tracking of AI’s impacts on the workforce and the economy; and
• Opportunities to consider more radical potential effects of AI on the workforce.

Opportunities to Influence the Rates and Direction of the Development of Artificial Intelligence

As discussed in Chapter 1, several factors have contributed to the accelerated pace of AI progress in recent years, including (1) the growing volume of available online text, video, and other kinds of data; (2) advances in the computational power of computing hardware; (3) advances in AI algorithms; and (4) the growing dollar investments in new research and development of AI by government and nongovernment organizations. At the same time, there are forces at work that may limit the rate of AI progress, including

(1) increasing secrecy around the exact AI methods used to produce state-of-the-art systems as companies compete for market share; (2) the possibility that large data sets needed to train advanced AI systems might not be widely available because they are held by private organizations that do not want to or cannot share them; (3) the possibility that U.S. research universities that have historically driven AI research and produced the AI experts hired by industry might not be able to afford the expense of developing and experimenting with frontier AI systems; and (4) societal concerns about the risks from AI including risks to safety, fairness, bias, surveillance, and privacy that could influence acceptance and adoption of AI in practice.

Given these considerations, opportunities for government leaders to influence the direction, robustness, and speed of AI development include the following:

• Support basic research in AI.
• Support research into standards and guardrails that could promote adoption of AI in business environments.
• Provide incentives, standards, and/or regulations to encourage sharing and transparency regarding the data used to train advanced AI models, enabling a level playing field where new companies can enter the market and contribute to progress while preserving privacy.
• Support AI research toward specific applications deemed to be of high societal priority, such as education and training, but where market forces appear to be insufficient to drive progress, although the benefits to society may be great.
• Fund initiatives such as the National AI Research Resource and the Microelectronics Commons that can provide hubs for computational resources and talent needed to keep U.S. universities vital players in the development of frontier AI methods and advances in AI safety, as well as evolve new research models.

Opportunities to Speed and Share the Productivity Benefits of Artificial Intelligence

Faster productivity growth would not only raise living standards but also make it easier to address a variety of national challenges including the budget deficit, poverty, health care, the environment, and national security. Rapid improvements in the technical capabilities of AI open the door to higher productivity growth, but they do not automatically translate into productivity gains. Typically, complementary investments are needed to translate technological advances into productivity gains, and there are a variety of bottlenecks and barriers that need to be addressed. Understanding the nature of these

complements and barriers should be a key goal of policy makers. There are several opportunities for policy makers:

• Address the uncertainty about the most effective applications of AI. Government can help provide, fund, and disseminate data and statistics on what works and what does not. In part, this can be done by better documenting AI adoption across American firms and the variation in their performance through expanded investment in efforts by national statistical agencies and in public–private partnerships to collect and share information about the effects of AI on productivity in the economy. These data and statistics should help to provide policy makers with the knowledge to assess the benefits and risks accurately and motivate the speed of adoption, implementation, and productivity gains.
• Support research into the effectiveness of policies that could enable labor mobility between occupations, firms, and geographical regions, and help workers take advantage of new job opportunities. Potential policies include retraining programs to help workers develop the new skills that will be needed and portable benefits and sensible occupational licensing rules that would make it easier for workers to move from job to job.
• Support research into areas that contribute to regulatory uncertainty—including product liability, copyright, privacy, and bias—and that complicate efforts of decision makers to assess benefits and risks, speed adoption and implementation, and drive productivity gains.
• Support research to identify and assess the potential for AI technologies to create new harms either inadvertently or through abuse and help policy makers understand and consider the associated trade-offs and work with the private sector to develop sensible guardrails.
• Support research to understand the implications for market concentration in AI, such as winner-take-most dynamics, and options for ensuring that consumer product markets are competitive while still enabling the benefits of scale and scope.
• Increase efforts to identify which specific occupational tasks are affected by AI, as well as which old and new skills and expertise will therefore be in greater or less demand.
• Support AI research that speeds scientific discovery, which is a key contributor to productivity growth.

Opportunities to Influence the Balance Among Worker Augmentation, New Work Creation, and Labor-Displacing Automation

The degree to which AI will assist current workers to perform their work with higher quality, versus substitute for workers and automate their tasks, will depend upon a complex set of economic and technical factors. However, the mixture of automation, augmentation, and new task creation that takes shape in the years ahead will not be determined merely by the technologies themselves but by the incentives and institutions in which they are created and deployed. Some nations already use AI to surveil their populations heavily, squelch viewpoints that depart from official narratives, and identify (and subsequently punish) dissidents—and they are rapidly exporting these capabilities to like-minded autocracies. In other settings, the same underlying AI technologies are used to advance medical drug discovery (including the development of COVID-19 vaccines), enable real-time translation of spoken languages, and provide free online tutoring in frontier educational subjects. The potential effects of AI on the work of the future depend critically on what objectives individuals, corporations, educational institutions, worker representatives, and governments pursue and what investments they make.

A crucial contention of the current report is that by providing relevant information, guidance, and digital guardrails, AI can enable workers who possess expert judgment to perform a broader range of expert tasks—in effect, allowing less-expert workers to perform more expert tasks. These beneficiaries of AI augmentation need not be exclusively or even primarily workers with college and post-college degrees. They can and should include the vast number of workers whose jobs require judgment, problem solving, and decision making, such as modern craft workers, teachers, health technicians at all levels, designers, contractors, software developers, customer support workers, skilled repair persons, and workers of innumerable jobs that demand clear written communications.

This beneficial deployment of AI to augment labor, complement expertise, and create new forms of valuable work is a possibility but is not an inevitable outcome. Achieving these ends requires intentional design that seeks to complement and expand the applicability of expertise rather than primarily displacing it. Policies that could be applied to further these objectives include the following:

• Support research on human-complementary AI—research into the design of AI systems that augment, rather than replace, human workers, resulting in human–AI teams that produce higher-quality outputs than either could alone.³

___________________

³ Much as the Defense Advanced Research Projects Agency made investments and held competitions to spur the development of dexterous robots and self-driving cars, the federal government could invest and foster competition in pairing AI tools with human expertise. See D. Acemoglu, D. Autor, and S. Johnson, 2023, “How AI Can Become Pro-Worker,” Center for Economic Policy Research Policy Insight 123, October 4, https://cepr.org/voxeu/columns/how-ai-can-become-pro-worker.

• Support research into best practices for fostering inclusive AI adoption within firms and organizations.
• Support research into how the relative costs of capital and labor affect business decisions about AI adoption for automation and for augmentation.
• Support research to explore how to provide individuals ways to control and be compensated for the use of their creative works, their likenesses, and their other personal attributes.
• Build AI expertise within the federal government to support effective investment, oversight, and regulation across all mission areas including transportation, labor, health care, education, environmental protection, public safety, and national security.
• Provide technology certification through which appropriate investments are incentivized by offering advice on the quality of human-complementary technology and whether it is good enough for use in publicly funded sectors such as education and health care.⁴

Opportunities to Understand the Implications of Artificial Intelligence for Education and Assist Workers with Retraining and Continuing Education

Although the exact direction and timing of future AI advances and their adoption are difficult to predict, it is much easier to predict that AI adoption will result in changing the nature of many jobs, yielding a shift in the demand for various types of current expertise, and creating demands for new kinds of expertise not previously considered. These changes in labor demand will increase the need for retraining programs for many workers as they navigate the changing jobs landscape. Opportunities to assist workers in their efforts to adapt to these changes include the following:

• Support research on effective continuing education approaches, especially short-term programs that teach specific skills in high and growing demand. Community colleges and sectoral programs can be important in delivering such programs.
• Support research into the nature, types, and delivery methods for continuing education and training programs to foster workforce flexibility.
• Support research on how AI, augmented reality, and other technologies can be used to improve education.
• Support research into how standards and certification for training programs can help community colleges and other educational institutions more

___________________

⁴ D. Acemoglu, D. Autor, and S. Johnson, 2023, “How AI Can Become Pro-Worker,” Center for Economic Policy Research Policy Insight 123, October 4, https://cepr.org/voxeu/columns/how-ai-can-become-pro-worker.

• effectively signal graduates’ skills to employers and improve the match of new graduates to in-demand job opportunities.
• Support appropriate organizations to develop, maintain, and disseminate a “career roadmap” that would enable workers to understand the shifting demand for different types of skills and workers (and improve matching) as well as the continuing education opportunities available to them to acquire high-demand skills that will advance their career.
• Support research into new education objectives for all levels of education, including K–12, in order to provide the current and next generation with the knowledge and skills needed to take full advantage of future AI capabilities.

Opportunities for More Exact and Timely Tracking of Artificial Intelligence’s Impacts on the Workforce and the Economy

Given the great uncertainty about exactly which AI technologies will become available when, as well as uncertainties around factors that will influence their adoption, together with the likelihood that AI will have a broad and profound impact on the workforce, it is essential to improve capabilities to observe and track in near real time the impacts AI is having on the economy, especially on the supply and demand for different types of worker expertise.

During the COVID-19 pandemic, the United States experienced significant supply chain challenges in areas ranging from masks to semiconductors, and efforts to gain greater real-time situational awareness capitalized on emerging private-sector data sources and methodologies that used machine learning. This effort underscored the challenge of integrating public- and private-sector data, particularly firm-level data that provide vital insights but include strategically sensitive information. During the Great Depression, a similar recognition of the need for greater situational awareness of the state of the economy sparked a comprehensive effort to advance new methodologies and build an infrastructure for the collection of employment data (and the development of new areas of academic research⁵).

Collecting and transparently disseminating this information to the workforce is one of the best ways to support workers in making their own career and continuing education decisions during a time of rapid change. This is a task over which government leaders can have significant control, including the following:

___________________

⁵ For example, D. Card, 2011, “Origins of the Unemployment Rate: The Lasting Legacy of Measurement without Theory,” American Economic Review 101(3):552–557, https://doi.org/10.1257/aer.101.3.552.

• Improve and expand existing data collection efforts by government agencies, including high-frequency, real-time tracking of the use of AI by businesses and workers and the impact on the workforce.
• Create new public–private data partnerships in which privately held data about skills supply and demand, and wages currently paid for these positions as well as continuing education opportunities and their link to getting a better job are collected, with summaries made available in real time to members of the workforce to support their efforts to improve their livelihoods.
• Explore the development of an independent, not-for-profit, government-chartered entity to create the infrastructure, protocols, and expertise needed to support public–private data sharing and integrated analysis. The goal would be both to support the above goal of keeping the workforce well informed about risks and opportunities and to provide greater situational awareness of the general state of the economy by building an infrastructure for data collection and development of new methodologies for assessing the contributions of goods and services, especially digital goods and services that are poorly accounted for in the current national accounts.
• Measure and mitigate disparate impacts of new technologies on underrepresented groups or communities as well as the global impact of differences in AI adoption between high-income and low-income countries.
• Measure and characterize the heterogeneity in patterns of AI adoption across economic sectors, across firms within sectors, and across geographical regions, along with their heterogeneous impacts on productivity and the workforce.

Opportunities to Consider More Radical Potential Effects of Artificial Intelligence on the Workforce

Recent capabilities of AI, especially those relating to LLMs and their multimodal counterparts, have advanced faster than many experts expected. Accordingly, it is possible that more radical improvements will emerge in the coming decade as funding increases, the models get larger, hardware improves, more data are used, and research continues. The effects on the workforce could be commensurately greater than anything seen so far. Consideration of these possibilities suggests several actions, including the following:

• Dramatically expand the research needed to inform policies and strategies that can assess the social and economic implications of emerging AI technologies. Build the instructional capabilities and scale training and educational programs accordingly.

• Undertake scenario planning to consider more extreme cases, even if they currently appear unlikely, such as AI’s being able to do a large majority of cognitive, emotional, and/or physical tasks currently done by the workforce, and the implications for wages, employment, and income distribution.
• Develop a comprehensive and timely dashboard of key economic and social indicators that reflect potential changes in the economy that might be expected if AI becomes substantially more pervasive and powerful, including the extent of AI use in the workforce, capital–labor substitution in various tasks, resources devoted to training and deploying AI models, and productivity gains in particular tasks and activities as well as in the aggregate.
• Undertake modeling and simulation exercises to consider the potential effects on employment, capital allocation, government budgets, household incomes, and other metrics of sharply more capable models.
• Explore policy options and pilot programs that could be appropriate to implement or scale up if AI becomes radically more capable.

THE ROAD AHEAD

In the 7 years since the previous National Academies’ report on automation and the workforce,⁶ and even in the time since this committee began work on this study, AI capabilities have improved tremendously. Moreover, it is reasonable to expect even greater advances in the years to come. Improved intelligence is key to solving many of the nation’s thorniest problems and unlocking greater prosperity and well-being. As a result, AI, the technology of intelligence, ranks among the most general of all general-purpose technologies.

Although technical progress and business transformation will continue, it is impossible to predict exactly the nature of the coming changes and all their effects on the economy and society. Accordingly, it makes sense to build in the ability for rapid data gathering and analysis to track these changes, and to build as flexible an approach as possible for reacting to the changes observed. Fighter pilots are most likely to be successful when they have a fast and accurate OODA loop—the ability to observe, orient, decide, and act—and the same applies for individuals, organizations, and even societies. In practice, this means increased capacity for research not only in AI but also in social science so AI’s implications can be understood better.

___________________

⁶ 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.

Rather than trying to predict any specific future path, it is important to have the flexibility to sense and respond rapidly to opportunities and challenges and be prepared for a variety of scenarios and possibilities.

Most importantly, individual people, businesses, nonprofits, colleges and universities, institutions of civil society, and governments each have agency about the type of society to which they belong. AI is a tool that is meant to be directed by humans. As it becomes more powerful, it will afford greater ability to shape the world in accordance with society’s values and goals. All elements of society can and should think carefully about what ends they seek to achieve with this tool.

This page intentionally left blank.

Appendixes

This page intentionally left blank.