The Economics of Artificial Intelligence

Home > Other > The Economics of Artificial Intelligence > Page 41
The Economics of Artificial Intelligence Page 41

by Ajay Agrawal


  York: Harper & Row.

  236 Daron Acemoglu and Pascual Restrepo

  Syverson, Chad. 2017. “Challenges to Mismeasurement Explanations for the US

  Productivity Slowdown.” Journal of Economic Perspectives 31 (2): 165– 86.

  Tuzel, Selale, and Miao Ben Zhang. 2017. “Economic Stimulus at the Expense of

  Routine- Task Jobs.” Unpublished manuscript, Marshall School of Business, Uni-

  versity of Southern California.

  Zeira, Joseph. 1998. “Workers, Machines, and Economic Growth.” Quarterly Journal

  of Economics 113 (4): 1091– 117.

  9

  Artifi cial Intelligence

  and Economic Growth

  Philippe Aghion, Benjamin F. Jones, and Charles I. Jones

  9.1 Introduction

  This chapter considers the implications of artifi cial intelligence for eco-

  nomic growth. Artifi cial intelligence (AI) can be defi ned as “the capability

  of a machine to imitate intelligent human behavior” or “an agent’s ability to

  achieve goals in a wide range of environments.”1 These defi nitions immedi-

  ately evoke fundamental economic issues. For example, what happens if AI

  allows an ever- increasing number of tasks previously performed by human

  labor to become automated? Artifi cial intelligence may be deployed in the

  ordinary production of goods and services, potentially impacting economic

  growth and income shares. But AI may also change the process by which we

  create new ideas and technologies, helping to solve complex problems and

  scaling creative eff ort. In extreme versions, some observers have argued that

  AI can become rapidly self- improving, leading to “singularities” that feature

  unbounded machine intelligence and/or unbounded economic growth in

  Philippe Aghion is a professor at the Collège de France and at the London School of Economics. Benjamin F. Jones is the Gordon and Llura Gund Family Professor of Entrepreneurship, professor of strategy, and faculty director of the Kellogg Innovation and Entrepreneurship Initiative at Northwestern University, and a research associate of the National Bureau of Economic Research. Charles I. Jones is the STANCO 25 Professor of Economics at the Graduate School of Business at Stanford University and a research associate of the National Bureau of Economic Research.

  We are grateful to Ajay Agrawal, Mohammad Ahmadpoor, Adrien Auclert, Sebastian Di

  Tella, Patrick Francois, Joshua Gans, Avi Goldfarb, Pete Klenow, Hannes Mahlmberg, Pascual Restrepo, Chris Tonetti, Michael Webb, and participants at the NBER Conference on Artifi cial Intelligence for helpful discussion and comments. For acknowledgments, sources of research support, and disclosure of the authors’ material fi nancial relationships, if any, please see http://

  www .nber .org/ chapters/ c14015.ack.

  1. The former defi nition comes from the Merriam- Webster dictionary, while the latter is from Legg and Hutter (2007).

  237

  238 Philippe Aghion, Benjamin F. Jones, and Charles I. Jones fi nite time (Good 1965; Vinge 1993; Kurzweil 2005). Nordhaus (2015) provides a detailed overview and discussion of the prospects for a singularity

  from the standpoint of economics.

  In this chapter, we speculate on how AI may aff ect the growth process.

  Our primary goal is to help shape an agenda for future research. To do so,

  we focus on the following questions:

  • If AI increases automation in the production of goods and services,

  how will it impact economic growth?

  • Can we reconcile the advent of AI with the observed constancy in

  growth rates and capital share over most of the twentieth century?

  Should we expect such constancy to persist in the twenty- fi rst century?

  • Do these answers change when AI and automation are applied to the

  production of new ideas?

  • Can AI drive massive increases in growth rates, or even a singularity, as

  some observers predict? Under what conditions, and are these condi-

  tions plausible?

  • How are the links between AI and economic growth modulated by

  fi rm- level considerations, including market structure and innovation

  incentives? How does AI aff ect the internal organization of fi rms, and

  with what implications?

  In thinking about these questions, we develop two main themes. First,

  we model AI as the latest form in a process of automation that has been

  ongoing for at least 200 years. From the spinning jenny to the steam engine

  to electricity to computer chips, the automation of aspects of production

  has been a key feature of economic growth since the Industrial Revolution.

  This perspective is taken explicitly in two key papers that we build upon:

  Zeira (1998) and Acemoglu and Restrepo (2016). We view AI as a new form

  of automation that may allow additional tasks to be automated that previ-

  ously were thought to be out of reach from automation. These tasks may

  be nonroutine (to use the language of Autor, Levy, and Murnane [2003]),

  like self- driving cars, or they may involve high levels of skill, such as legal

  services, radiology, and some forms of scientifi c lab- based research. An

  advantage of this approach is that it allows us to use historical experience

  on economic growth and automation to discipline our modeling of AI.

  A second theme that emerges in our chapter is that the growth conse-

  quences of automation and AI may be constrained by Baumol’s “cost dis-

  ease.” Baumol (1967) observed that sectors with rapid productivity growth,

  such as agriculture and even manufacturing today, often see their share of

  gross domestic product (GDP) decline while those sectors with relatively

  slow productivity growth—perhaps including many services—experience

  increases. As a consequence, economic growth may be constrained not by

  what we do well but rather by what is essential and yet hard to improve. We

  suggest that combining this feature of growth with automation can yield a

  Artifi cial Intelligence and Economic Growth 239

  rich description of the growth process, including consequences for future

  growth and income distribution. When applied to a model in which AI

  automates the production of goods and services, Baumol’s insight gener-

  ates suffi

  cient conditions under which one can get overall balanced growth

  with a constant capital share that stays well below 100 percent, even with

  near- complete automation. When applied to a model in which AI automates

  the production of ideas, these same considerations can prevent explosive

  growth.2

  The chapter proceeds as follows. Section 9.2 begins by studying the role

  of AI in automating the production of goods and services. In section 9.3,

  we extend AI and automation to the production of new ideas. Section 9.4

  then discusses the possibility that AI could lead to superintelligence or even

  a singularity. In section 9.5, we look at AI and fi rms, with particular atten-

  tion to market structure, organization, reallocation, and wage inequality. In

  section 9.6, we examine sectoral evidence on the evolution of capital shares

  in tandem with automation. Finally, section 9.7 concludes.

  9.2 Artifi cial Intelligence and Automation of Production

  One way of looking at the last 150 years of economic progress is that it

  is driven by automation. The Industrial Revolution used steam and then

  electricity to automate many pr
oduction processes. Relays, transistors, and

  semiconductors continued this trend. Perhaps artifi cial intelligence is the

  next phase of this process rather than a discrete break. It may be a natural

  progression from autopilots, computer- controlled automobile engines,

  and MRI machines to self- driving cars and AI radiology reports. While up

  until recently automation has mainly aff ected routine or low- skilled tasks,

  it appears that AI may increasingly automate nonroutine, cognitive tasks

  performed by high- skill workers.3 An advantage of this perspective is that it

  allows us to use historical experience to inform us about the possible future

  eff ects of AI.

  9.2.1 The Zeira (1998) Model of Automation and Growth

  A clear and elegant model of automation is provided by Zeira (1998). In

  its simplest form, Zeira considers a production function like

  n

  (1)

  Y = AX 1 X 2 . . . X n where

  = 1.

  1

  2

  n

  i

  i=1

  2. In the appendix we show that if some steps in the innovation process require human R&D, AI could possibly slow or even end growth by exacerbating business stealing, which in turn discourages human investments in innovation.

  3. Autor, Levy, and Murnane (2003) discuss the eff ects of traditional software automating routine tasks. Webb et al. (2017) use the text of patent fi lings to study the diff erent tasks that AI, software, and robotics are best positioned to automate.

  240 Philippe Aghion, Benjamin F. Jones, and Charles I. Jones While Zeira thought of the X s as intermediate goods, we follow Acemoglu

  i

  and Autor (2011) and refer to these as tasks; both interpretations have merit,

  and we will go back and forth between these interpretations. Tasks that have

  not yet been automated can be produced one- for- one by labor. Once a task

  is automated, one unit of capital can be used instead:

  L if not automated

  (2)

  X =

  i

  .

  i

  K if automated

  i

  If the aggregate capital K and labor L are assigned to these tasks optimally, the production function can be expressed (up to an unimportant constant) as

  (3)

  Y = A K L 1 ,

  t

  t

  t

  t

  where it is now understood that the exponent refl ects the overall share and

  importance of tasks that have been automated. For the moment, we treat

  as a constant and consider comparative statics that increase the share of

  tasks that get automated.

  Next, embed this setup into a standard neoclassical growth model with

  a constant investment rate; in fact, for the remainder of the chapter this is

  how we will close the capital/ investment side of all our models. The share

  of factor payments going to capital is given by and the long- run growth

  rate of y ≡ Y/ L is

  (4)

  g =

  g ,

  y

  1

  where g is the growth rate of A. An increase in automation will therefore increase the capital share and, because of the multiplier eff ect associated

  with capital accumulation, increase the long- run growth rate.

  Zeira emphasizes that automation has been going on at least since the

  Industrial Revolution, and his elegant model helps us to understand that.

  However, its strong predictions that growth rates and capital shares should

  be rising with automation go against the famous Kaldor (1961) stylized

  facts that growth rates and capital shares are relatively stable over time. In

  particular, this stability is a good characterization of the US economy for

  the bulk of the twenieth century, for example, see Jones (2016). The Zeira

  framework, then, needs to be improved so that it is consistent with historical

  evidence.

  Acemoglu and Restrepo (2016) provide one approach to solving this prob-

  lem. Their rich environment allows for a constant elasticity of substitution

  (CES) production function and endogenizes the number of tasks as well as

  automation. In particular, they suppose that research can take two diff erent

  directions: discovering how to automate an existing task or discovering new

  tasks that can be used in production. In their setting, a refl ects the fraction of tasks that have been automated. This leads them to emphasize one possible

  Artifi cial Intelligence and Economic Growth 241

  resolution to the empirical shortcoming of Zeira: perhaps we are inventing

  new tasks just as quickly as we are automating old tasks. The fraction of

  tasks that are automated could be constant, leading to a stable capital share

  and a stable growth rate.

  Several other important contributions to this rapidly expanding literature

  should also be noted. Peretto and Seater (2013) explicitly consider a research

  technology that allows fi rms to change the exponent in a Cobb- Douglas

  production function. While they do not emphasize the link to the Zeira

  model, with hindsight the connections to that approach to automation are

  interesting. The model of Hemous and Olsen (2016) is closely related to

  what follows in the next subsection. They focus on CES production instead

  of Cobb- Douglas, as we do below, but emphasize the implications of their

  framework for wage inequality between high- skill and low- skill workers.

  Agrawal, McHale, and Oettl (2017) incorporate artifi cial intelligence and

  the “recombinant growth” of Weitzman (1998) into an innovation- based

  growth model to show how AI can speed up growth along a transition path.

  The next section takes a complementary approach, building on this lit-

  erature and using the insights of Zeira and automation to understand the

  structural change associated with Baumol’s cost disease.

  9.2.2 Automation and Baumol’s Cost Disease

  The share of agriculture in GDP or employment is falling toward zero.

  The same is true for manufacturing in many countries of the world. Maybe

  automation increases the capital share in these sectors and also interacts

  with nonhomotheticities in production or consumption to drive the GDP

  shares toward zero. The aggregate capital share is then a balance of a rising

  capital share in agriculture/ manufacturing/ automated goods with a declin-

  ing GDP share of these goods in the economy.

  Looking toward the future, 3D printing techniques and nanotechnology

  that allow production to start at the molecular or even atomic level could

  someday automate all manufacturing. Could AI do the same thing in many

  service sectors? What would economic growth look like in such a world?

  This section expands on the Zeira (1998) and Acemoglu and Restrepo

  (2016) models to develop a framework that is consistent with the large struc-

  tural changes in the economy. Baumol (1967) observed that rapid productiv-

  ity growth in some sectors relative to others could result in a “cost disease”

  in which the slow- growing sectors become increasingly important in the

  economy. We explore the possibility that automation is the force behind

  these changes.4

  4. The growth literature on this structural transformation emphasizes a range of possible mechanisms, see Kongsamut, Rebelo, and Xie (2001), Ngai and Pissarides (20
07), Herrendorf, Rogerson, and Valentinyi (2014), Boppart (2014), and Comin, Lashkari, and Mestieri (2015).

  The approach we take next has a reduced form that is similar to one of the special cases in Alvarez- Cuadrado, Long, and Poschke (2017).

  242 Philippe Aghion, Benjamin F. Jones, and Charles I. Jones Model

  Gross domestic product is a CES combination of goods with an elasticity

  of substitution less than one:

  1/

  1

  (5)

  Y = A

  X di

  where > 0,

  t

  t

  it

  0

  where A = A egt captures standard technological change, which we take to t

  0

  be exogenous for now. Having the elasticity of substitution less than one

  means that tasks are gross complements. Intuitively, this is a “weak link”

  production function, where GDP is in some sense limited by the output of

  the weakest links. Here, these will be the tasks performed by labor, and this

  structure is the source of the Baumol eff ect.

  As in Zeira, another part of technical change is the automation of produc-

  tion. Goods that have not yet been automated can be produced one- for- one

  by labor. When a good has been automated, one unit of capital can be used

  instead:

  L if not automated

  (6)

  X =

  it

  .

  it

  K if automated

  it

  This division is stark to keep the model simple. An alternative would be to

  say that goods are produced with a Cobb- Douglas combination of capital

  and labor, and when a good is automated, it is produced with a higher expo-

  nent on capital.5

  The remainder of the model is neoclassical:

  (7)

  Y = C + I ,

  t

  t

  t

  (8)

  K = I

  K ,

  t

  t

  t

  1

  (9)

  K di = K ,

  it

  t

  0

  1

  (10)

  L di = L.

  it

  0

  We assume a fi xed endowment of labor for simplicity.

  Let be the fraction of goods that that have been automated as of date

  t

  t. Here, and throughout the chapter, we assume that capital and labor are

  allocated symmetrically across tasks. Therefore, K / units of capital are

 

‹ Prev