Despite the economic damage wrought by the novel coronavirus over the past year, an analysis published by The Economist in December 2020 argues that the COVID-19 pandemic may have made a boom in productivity more likely to happen because “new technologies are clearly able to do more than has generally been asked of them.” This would be welcome news to observers who have scratched their heads about why supposedly innovative technologies like cloud computing and artificial intelligence have struggled to materially affect topline productivity growth numbers or the rate of overall GDP growth.
Office closures have made firms invest in automation and digitization and make better use of existing resources. Survey data collected by the World Economic Forum during the pandemic show that more than 80% of employers are planning to accelerate the adoption of advanced technologies and provide more opportunities for remote work, while 50% plan to accelerate automation of production tasks. In a way not seen for the last two decades, this turn has the potential to provide sustained productivity growth—the ultimate engine of economic activity in the long run.
To take a step back, in the past decade digital goods and services have been criticized for underdelivering on their enormous economic promise. In spite of the emergence of new technologies capable of diagnosing diseases, understanding speech, or recognizing images, these tools have had startlingly little effect on the disappointingly slow productivity growth rate of advanced economies, critics argue. Indeed, the pace of productivity growth has decelerated in the past two decades—from an average of 2.8% per year in the decade ending in 2005, down to 1.3% per year from 2006 through 2019.
In a recent Stanford HAI and Digital Economy Lab policy brief, we took stock of the latest research and identified four potential reasons why productivity growth is slowing. Besides examining each of these four explanations, we want to sketch out what policymakers can do to reverse this trend, reduce income inequality, and make the United States more competitive. This set of policy actions falls into three broad categories:
- Increasing investments in research and development through direct grants and tax credits.
- Expanding the human capital available to the economy by boosting our education system and expanding immigration of high-skilled labor.
- Removing the legal and regulatory bottlenecks that currently exist to entrepreneurship and business innovation.
Establishing root causes
To begin, why has productivity growth slowed in spite of immense technological innovation? First, we have to consider the possibility that today’s technological advances simply fall short of the promise envisioned by their developers and that they will never fulfill their expected economic promise. Second, economists might be failing to measure properly all the aspects in which technological changes are affecting the economy. Third, the new technologies under consideration may be privately beneficial to the companies that developed them but have fewer positive effects on the broader economy. Lastly and most compellingly from our perspective, transformative technologies take time to diffuse throughout the economy and must be accompanied by appropriate investments and adjustments. They don’t typically translate into improvements in productivity until complementary innovations have been developed.
The argument that tech hype is overblown and will never fulfill supposedly irrational expectations rests on the contested observation that the rate at which innovations are being created is decreasing. This is borne out in some respects, since it is increasingly difficult for researchers to reach the frontiers of their discipline as more specialization is needed per innovation than before. But we do not find as compelling the parallel argument that productivity gains from the adoption of I.T. systems installed early in the 21st century have run their course and that no additional technological improvements to productivity should be expected.
Moreover, as information flows and knowledge-based work increases in importance, accounting for digital goods and services has become an increasingly important part of the economic conversation. While traditional growth accounting handles the case of economic activity like manufacturing pretty well, instances of unmeasured inputs and unmeasured outputs that stem from what are known as intangible or hidden assets—assets like corporate culture or business processes that are not documented on balance sheets, not kept as inventory in a warehouse, and not easily transferable between firms—have upended the mechanics of economic measurement. This raises questions about whether growth accounting is properly capturing the ways in which digital technologies are changing the economy.
The second explanation, that we may be failing to properly measure new sources of economic activity, enjoys broader support than the overblown hype argument. Since the beginning of the productivity slowdown, the way consumers choose to value search engines and social networks demonstrates considerable fluctuation, as has consumer dependence on goods like mapping software and encyclopedias that were not free before they became digital goods. Improper or uncertain measurement must also be seen in conjunction with the fact that prices increasingly being mismeasured.
Improper and uncertain measurement is related to the third hypothesis, that lucrative technologies are increasingly used in zero-sum applications that merely shift wealth around and have fewer positive effects on the economy generally. An example of this can be seen in the misalignment of incentives in tax policy that subsidizes capital at the expense of labor and crowds out investment in labor generative technology. Capital subsidies result in firms developing technologies that are only marginally more efficient than workers but not sufficiently better to incentivize additional investment that could complement workers.
Lastly—and most importantly—slowing productivity growth may be the result of technologies taking time to reach their full economic potential. We find this argument most convincing because of the nature of general-purpose technologies (GPTs) like artificial intelligence—those that are generally pervasive and can improve over time but require complementary investments that are both intangible (e.g. in data collection, employee training) and physical (e.g. computers, 5G towers). In the early stages of GPT-related economic activity, it can appear increased tangible costs are required to achieve the same outputs as in the past—before unmeasured capital service flows and unmeasured costs to create that capital start to balance each other out. This is because of the substantial need for complementary innovations to many intangible assets, especially in the case of fundamental technology advancements such as AI. We have termed this phenomenon the “productivity J-curve.” As we have seen, complementary innovations for productivity enhancing technologies can take years or even decades to create and implement. In the meantime, measured productivity growth can fall below trends since real resources are devoted to investments in these innovations.
Supercharging productivity growth
Taking the above analysis into account allows us to develop the following recommendations policymakers should take to enhance productivity growth. In order to ensure that the economic gains from integrating these hard-to-measure intangible assets are not consumed entirely by the wealthy and privileged, we propose a set of interventions across three broad issue areas that are designed to share prosperity among the entire population.
First, to address inadequate research and development (R&D) efforts, boost levels of spending in both public and private R&D by authorizing large, government-directed research projects, government grants through the National Science Foundation or the National Institutes of Health, and through tax credits for private businesses. Fundamental science is often best carried out by government, academia, or nonprofits while marketable applications of that basic research are often optimized through private development. Thus, the federal government should adopt a diversified approach in building this program in order to reduce overall risk and fund early stage or large-scale projects that the private sector either would not be able to pursue or would not want to pursue. This will increase the likelihood of positive effects from at least one avenue.
The second category of policy actions we recommend involve increasing U.S. human capital. This can be accomplished through shoring up our education system and encouraging high-skilled immigration. Boosting the attractiveness of the United States to high-skilled immigrants is the most simple and important action the country could take today to increase growth. This includes immigrants and refugees who do not have university degrees. Those who come to America tend to be entrepreneurial and ambitious and represent some of the most talented and tenacious individuals in their home countries. Immigration also has the added benefit of expanding market size and providing opportunities for entrepreneurs to serve specialized markets.
The United States should also boost funding and support for schools and universities, including by potentially funding new universities, updating the land-grant process used to create institutions like the University of California system, or by allocating appropriately sized endowments to be administered by the states. In order to better prepare children for college, the United States should do more to improve the quality of primary- and secondary-school instruction through better accountability for teachers, extending the length of school days and the school year, offering optional weekend classes, and providing one-on-one math tutoring. The goal here is to not only produce more STEM PhDs in the United States, but to promote the training of scientists abroad as well, since R&D conducted abroad is likely to produce positive spillovers in this country.
Our third category of policy interventions involve eliminating bottlenecks to innovation in the legal, regulatory, and tax realms. In order to reduce adjustment costs and lags to the benefits of new technologies, policymakers should pursue legislation to eliminate or weaken the non-compete clauses that prevent skilled engineers from bringing their talents and insights to competitors. Policymakers should further enact intellectual property reforms that push more technologies and artistic concepts into the public domain. Besides investing in infrastructure and public goods, the United States should also reinvigorate antitrust enforcement, including by empowering the Federal Trade Commission to subpoena information needed for better understanding and regulating monopolies.
Rather than focusing on breaking up digital platforms—which might destroy productivity-enhancing network effects—the federal government should promote standards that enable easier market entry and interoperability among competitors. Where this is impossible, regulators should focus on tax policy, regulation, and collective bargaining tools to ensure the benefits from these platforms are more widely distributed. Decoupling healthcare from employment and reforming occupational licensing will help make it easier for people to start a new business and boost entrepreneurship. Lastly, the United States should correct the subsidy it currently provides to capital-intensive automation over the invention of new tasks for labor. Washington should also collaborate with other countries on corporate tax reform in order to prevent a race to the bottom with respect to corporate tax havens in the international contest to attract capital.
Pursuing these policies will reward firms for creating new jobs instead of economizing on labor costs and will ensure that the innovation provided by GPTs accelerates productivity growth across the entire economy. This in turn will help expand wages, reduce income inequality, and ensure that more equitable growth is enjoyed across the country. Addressing the productivity paradox will contribute to the speedy integration of scalable machine intelligence into the global economy and ensure that its integration reflects our fundamental values about the dignity of human work. In sum, we are optimistic that the coming decade will be one of higher productivity growth as firms continue to adjust their business practices because of the COVID-19 pandemic and as policymakers take the reins in making a plan for equitable growth a reality.
Erik Brynjolfsson is the Jerry Yang and Akiko Yamazaki Professor and Senior Fellow at the Stanford Institute for Human-Centered AI (HAI), and Director of the Stanford Digital Economy Lab. He also is the Ralph Landau Senior Fellow at the Stanford Institute for Economic Policy Research (SIEPR), Professor by Courtesy at the Stanford Graduate School of Business and Stanford Department of Economics, and a Research Associate at the National Bureau of Economic Research (NBER).
Seth G. Benzell is an Assistant Professor at the Argyros School of Business and Economics at Chapman University. He is a Fellow of the Stanford Digital Economy Lab, part of the Stanford Institute for Human-Centered Artificial Intelligence (HAI).
Daniel Rock is an Assistant Professor of Operations, Information, and Decisions at the Wharton School of the University of Pennsylvania and a Digital Fellow at both the Stanford HAI Digital Economy Lab and the MIT Initiative on the Digital Economy.
This post is adapted from the Stanford HAI Digital Economy Lab’s policy brief, “Building Solutions to the Modern Productivity Paradox”