AI and the Productivity Paradox
AI is being successfully applied to tasks that not long ago were viewed as the exclusive domain of humans, matching or surpassing human level performance in more and more domains. But, at the same time, productivity growth has significantly declined over the past decade, and income has continued to stagnate for the majority of Americans. This puzzling contradiction is addressed in Artificial Intelligences and the Modern Productivity Paradox, a working paper recently published by the National Bureau of Economic Research (NBER).
The paper’s authors, - MIT professor Erik Brynjolfsson, MIT PhD candidate Daniel Rock, and University of Chicago professor Chad Syverson, - note that “aggregate labor productivity growth in the U.S. averaged only 1.3% per year from 2005 to 2016, less than half of the 2.8% annual growth rate sustained from 1995 to 2004… What’s more, real median income has stagnated since the late 1990s and non-economic measures of well-being, like life expectancy, have fallen for some groups.”
This productivity puzzle isn’t confined to AI. “Few topics in economics today in most large economies generate as much debate as the productivity puzzle,” said McKinsey in a March, 2017 report. “In the United States, productivity growth has declined sharply since 2004 yet digital technology has been widely apparent during this period… The answer to this puzzle holds the key to future prosperity because now more than ever our economy depends on productivity improvements for long-term economic growth. Economists have proposed competing explanations for declining productivity growth and so far have failed to reach a consensus.”
After considering four potential explanations, the NBER paper concluded that there’s actually no productivity paradox. Given the proper context, there are no inherent inconsistencies between having both transformative technological advances and lagging productivity. Over the past two centuries we’ve learned that there’s generally a significant time lag between the broad acceptance of new technology-based paradigms and the ensuing economic transformation and institutional recomposition. Even after reaching a tipping point of market acceptance, it takes considerable time, - often decades, - for the new technologies and business models to be widely embraced by companies and industries across the economy, and only then will their benefits follow, - including productivity growth. The paper argues that we’re precisely in such an in-between period.
Let me briefly describe the four potential explanations explored in the paper: false hopes, mismeasurements, concentrated distribution, and implementation and restructuring lags.
False hopes. Perhaps our current technologies, advanced and exciting as they might be, aren’t as transformative as the technologies from the period between 1870 and 1970, when we experienced high productivity growth and a rising standard of living. Some have argued, - - most prominently Northwestern University economist Robert Gordon, - that over the past few decades there’s been a fundamental decline in innovation and productivity. “We wanted flying cars - instead we got 140 characters,” is how PayPal cofounder Peter Thiel succinctly described his belief that we’re no longer solving big problems. But, it’s not just flying cars that have eluded us. So has fusion energy, supersonic commercial travel and space exploration.
The NBER paper disagrees, arguing that there’s a compelling case for optimism. “[I]t’s not difficult to construct back-of-the-envelope scenarios in which even a modest number of currently existing technologies could combine to substantially raise productivity growth and societal welfare. Indeed, knowledgeable investors and researchers are betting their money and time on exactly such outcomes.”
Mismeasurement. Another potential explanation is the difficulty of measuring productivity in our services-oriented digital economy. GDP, which factors strongly in measures of productivity, is a relic of an age dominated by manufacturing, where the production of physical goods was easier to measure. But it’s a less reliable measure of services, where there’s much more variation in quality and value. Moreover, how do you measure the value of smartphones, social media, online videos, and a large assortment of websites, which involve relatively little monetary costs yet are now widely used in both our work and personal lives? Traditional metrics don’s adequately account for the digital economy, because so many of its technologies deliver substantial value while accounting for a small share of GDP due to their relatively low price.
The NBER paper cites a number of recent studies that, “each using different methodologies and data, present evidence that mismeasurement is not the primary explanation for the productivity slowdown. After all, while there is convincing evidence that many of the benefits of today’s technologies are not reflected in GDP and therefore productivity statistics, the same was undoubtedly true in earlier eras as well.”
Concentrated distribution. Another possibility is that the productivity benefits of new technologies are being enjoyed by a relatively small fraction of the economy. The digital economy has given rise to the global superstar company, primarily driven by economies of scale, generally achieved through platforms and network effects. The result is that a small number of companies have become category kings dominating the rest of their competitors in their particular market, - the Facebooks, Googles, Twitters, Ubers and AirBnbs. “As successive waves of innovation expand the definition of what is possible, the most sophisticated users have pulled far ahead of everyone else in the race to keep up with technology and devise the most effective business uses for it,” noted a December, 2015 McKinsey report.
“Although this evidence is important, it is not dispositive,” adds the NBER paper. “The aggregate effects of industry concentration are still under debate, and the mere fact that a technology’s gains aren’t evenly distributed is no guarantee that resources will be dissipated in trying to capture them - especially that there would be enough waste to erase noticeable aggregate benefits.”
Implementation and Technology Lags. The paper’s authors essentially explains the paradox away. We’re living in a time of transformative technology advances, - the Internet, smartphones, cloud, IoT, AI - whose deployment and impact on economic and productivity growth are still lagging. Moreover, the more transformative the technologies, the longer it takes for them to be embraced by companies and industries across the economy.
In her 2002 influential book, Technological Revolutions and Financial Capital, economic historian Carlota Perez wrote that since the advent of the Industrial revolution, we’ve had a technological revolution roughly every 60 years, - our present digital technology revolution being the 5th in that span.
Each technology revolution is composed of two different periods, each lasting around 30 years. The installation period is the time when new technologies emerge from the lab into the marketplace, entrepreneurs start many new businesses based on these new technologies, and venture capitalists encourage experimentation with new business models and speculation in new money-making schemes. Inevitably, this all leads to the kind of financial bubble and crash we’re quite familiar with from our recent experiences.
After the crash, comes the deployment period, which she views as a time of institutional recomposition. The now well accepted technologies and economic paradigms become the norm; infrastructures and industries start getting better defined and more stable; and production capital drives long-term economic and productivity growth by spreading and multiplying the successful business models.
In a recent roundtable discussion Perez was asked to comment on the differences between our present technology revolution and previous ones. We’re now over 45 years into the digital revolution, and over 15 years since the crash of the dot-com bubble, - the longest such periods according to her theory. Perez replied that our digital revolution is the deepest transformation of everyday life, as well as the most global. “Even after 40 years, the information and communications technology (ICT) revolution is far from complete. It hasn’t fully changed our way of life, as previous technological revolutions had done.
We’re still in a long in-between period, where multiple, overlapping technologies are continuing to emerge from R&D labs into the marketplace, but because of their profound transformative nature, their full deployment is still ahead of us.
The NBER paper further argues that general purpose technologies (GPT), like AI, are the most transformative, precisely due to their broad potential. Their deployment time-lags are longer because attaining their full benefits requires a number of complementary co-inventions and investments, including additional technologies, applications, processes, business models, and regulatory policies.
“The steam engine, electricity, the internal combustion engine, and computers are each examples of important general purpose technologies. Each of them increased productivity not only directly but also by spurring important complementary innovations.” AI has the potential to become such an important GPT as it’s improved upon over time, but we’re still in the very early stages of AI’s deployment. It’s only been in the last few years that complementary innovations like machine learning have taken AI from the lab to early adopters in the marketplace. Beyond the relatively small number of early adopters, AI is still in the experimental stage. Considerable innovations and investments are required for its wider deployment in robotics, self-driving cars, truly intelligent personal assistants, and smart healthcare applications.
“This explanation resolves the paradox by acknowledging that its two seemingly contradictory parts are not actually in conflict. Rather, both parts are in some sense natural manifestations of the same underlying phenomenon of building and implementing a new technology… While each of the first three explanations for the paradox might have a role in describing its source, the explanations also face serious questions in their ability to describe key parts of the data. We find the fourth - the implementation and restructuring lags story - to be the most compelling.”