Artificial Intelligence and the Future of Work (2025) / Chapter Skim
Currently Skimming:
Pages 29-64
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 29... ...
Trained LLMs are found to implicitly have much of this kind of 11 commonsense knowledge and, as illustrated in the preceding examples, much more. To be precise, one technical caveat must be mentioned here: Although the core transformer on which GPT-4 is based has been trained as described earlier (to predict Artificial Intelligence 29
|
From page 31... ...
This produced Artificial Intelligence 31
|
From page 33... ...
In other cases, computer vision can lead to higher quality or complete deployments of tasks previously performed by people alone -- for example, Artificial Intelligence 33
|
From page 34... ...
This application is under development, and although AI-based driver assistance is in fielded use, fully autonomous driving in arbitrary environments has remained elusive owing to the broad variety of rare cases that must be handled in order to achieve 100 percent autonomous 34 ARTIFICIAL INTELLIGENCE AND THE FUTURE OF WORK
|
From page 35... ...
Artificial Intelligence 35
|
From page 36... ...
For example, the advent of deep neural network algorithms around 2010 led to major advances in computer vision and speech recognition over the subsequent decade. More recently, the development of the transformer architecture and the method of training these systems to predict the next word in a sequence was also an important step forward.
|
From page 37... ...
Partially offsetting this has been the availability of open-source models, some of which were released by the technology companies. Artificial Intelligence 37
|
From page 38... ...
. However, there might also be regulations of more general AI technologies such as LLMs (e.g., to require that they report on their factual accuracy in various topics, or whether they can inadvertently output part of the text or images they were 38 ARTIFICIAL INTELLIGENCE AND THE FUTURE OF WORK
|
From page 39... ...
. Even in cases where AI assists and increases productivity of workers, it can be unclear whether demand for those workers will increase or decrease; improved productivity can lead to reduced prices and therefore increased demand, so that the final impacts depend also on elasticities in the demand for the goods that these Artificial Intelligence 39
|
From page 41... ...
In turn, improvements in technology, especially general-purpose technology, are key to better productivity growth. The most promising general-purpose technology of the present era is artificial intelligence (AI)
|
From page 42... ...
Adjustments in productivity metrics, although imperfect, can be made for capital quality, labor quality, capacity utilization, intangibles, and other factors. a For instance, the Congressional Budget Office projects that if productivity growth ends up being 0.5 percentage points higher than its baseline, the projected debt/GDP ratio for the United States would be about 40 percent lower by 2052.
|
From page 43... ...
Will wage growth continue to lag productivity growth as it has over the past 20–30 years during periods of both strong productivity growth and slow productivity growth? Artificial Intelligence and Productivity 43
|
From page 44... ...
Last, although the chapter focuses on AI's effects on productivity, it ends with a brief discussion of how AI may affect other measures of human well-being such as social progress and happiness as well as how it poses significant risks that could undermine human well-being, including risks to privacy, risks of discrimination and bias, risks to democracy and political stability, ethical risks, national security risks, risks of military arms races driven by new AI weapons, and even existential risks. In the words of Ian Bremmer and Mustafa Suleyman, "The decentralized nature of AI development and the core characteristics of the technology, such as open-source proliferation, increase the likelihood that it will be weaponized by cybercriminals, state-sponsored actors, and lone wolves."2 ARTIFICIAL INTELLIGENCE: A GENERAL-PURPOSE TECHNOLOGY Although adoption of AI so far is limited, AI is a general-purpose technology, much like the steam engine and electricity.
|
From page 45... ...
6 N.N. Maslej, ed., 2023, "Artificial Intelligence Index Report 2023," Stanford University Human-Centered Artificial Intelligence, https://aiindex.stanford.edu/report.
|
From page 46... ...
FIGURE 3-2 The slowdown in labor productivity primarily reflects slower total factor productivity growth. SOURCE: Based on data from the Federal Reserve Bank of San Francisco, n.d., "Total Factor Productivity," https://www.frbsf.
|
From page 47... ...
The Great Recession was sparked by a financial crisis that left many firms facing constraints on their investments in physical, intangible, and human capital. During the recovery, the decline in the growth of capital intensity per worker explains about one-third of the slowdown in labor productivity growth.
|
From page 48... ...
producing and using industries -- "high-tech" sectors -- accounted for the surge in productivity growth between 1995 and 2004, and they led the significant decline thereafter. Byrne and colleagues provide compelling evidence of the role of these industries.9 Figure 3-4 shows patterns of growth rates of labor productivity in subperiods from 1990 to 2019 for high-tech and other industries.
|
From page 49... ...
EXPLANATIONS FOR THE SLOWDOWN IN PRODUCTIVIT Y GROW TH Economists differ on explanations for the significant, unexpected, and persistent slow down in labor productivity growth and TFP growth, which occurred not just in the United States but in the other advanced economies after 2006. As noted in the preceding section, all of these economies were hit by the Great Recession and an anemic recovery that slowed investment and contributed to slower productivity growth.
|
From page 50... ...
Miranda, 2020, "Changing Business Dynamism and Productivity: Shocks Versus Responsiveness," American Economic Review 110(12) :3952–3990, https://doi.
|
From page 51... ...
and increases in political and 18 McKinsey Global Institute, 2023, "An Approach to Boosting US Labor Productivity," May 25, https://www. mckinsey.com/mgi/our-research/an-approach-to-boosting-us-labor-productivity.
|
From page 52... ...
Miranda, 2020, "Changing Business Dynamism and Productivity: Shocks Versus Responsiveness," American Economic Review 110(12) :3952–3990, https://doi.
|
From page 53... ...
gov/data/datasets/time-series/econ/bds/bds-datasets.html. Artificial Intelligence and Productivity 53
|
From page 54... ...
Economic theory suggests that over time, more productive firms grow while less productive firms are replaced or are driven by competition to improve their performance. Such productivity-enhancing reallocation has been an important contributor to productivity growth over time.28 Relatedly, the innovative process itself is closely tied to the pace of reallocation, with young firms playing an outsized role in major innovations.29,30 Unfortunately, as shown above, during the productivity slowdown there has been a decline in the pace of business dynamism and entrepreneurship in the United States.
|
From page 55... ...
They argue that these patterns are consistent with large incumbent firms having strategic reasons to slow innovation so as not to cannibalize their existing products and market shares. Their findings reinforce concerns about the potentially adverse implica tions for innovation and productivity growth of both increasing concentration of large incumbents in many sectors, especially mega firms in high-tech sectors, and decreasing entrepreneurship.
|
From page 56... ...
New AI firms are likely to be classified in one of these two industries. EFFECTS OF ARTIFICIAL INTELLIGENCE ON PRODUCTIVIT Y Overall Adoption Is Limited But Growing Rapidly AI adoption in most firms is still low, but it has been gradually permeating economic activity over several years -- for example, with the technology powering smartphones, in autonomous-driving features on cars, for digital retail sales via platforms like Amazon, for 34 J.C.
|
From page 57... ...
Acemoglu and colleagues also report that firms adopting AI have higher productivity and lower labor shares than similar firms, a result that is consistent with automation being a major application for AI.40 Of AI adopters, 15 percent reported an increase in employment and 6 percent reported a decrease, while 41 percent reported an increase in skill demand and none reported a decrease. Although AI has affected specific applications and firms, to date the deployment of AI has been too small to have had a detectable effect on aggregate productivity growth or on productivity growth by industry.
|
From page 58... ...
The finance industry was hit hard by the 2007–2008 global financial crisis and the restructuring that followed, with negative consequences for its productivity growth. A Framework for Thinking About the Effects of AI on Aggregate Productivity How much will AI affect aggregate productivity?
|
From page 59... ...
Over time, the productive sectors may require less labor and fall in cost. If demand does not grow commensurately, then the sectors with rapid productivity growth will shrink while the more stagnant sectors will become increasingly important.
|
From page 61... ...
w25695. Artificial Intelligence and Productivity 61
|
From page 62... ...
Factor 1: Share of the Economy Potentially Affected by AI A major reason AI may significantly boost labor productivity growth is that it has poten tial applications in so many parts of the economy. Generative AI along with other types of AI and robotics have the potential to affect activities that today encompass a majority 52 See, for example, D
|
From page 63... ...
It is important to note that to date much of the investment in generative AI is concentrated in a handful of highly digitized tech giants and platform companies along with venture capital–financed firms in the United States. Unlike other technological advances in recent decades that automated many routine physical and cognitive tasks done by humans, generative AI systems will mostly affect cognitive work -- both routine tasks and nonroutine tasks.
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.