The Economics of Artificial Intelligence
Page 31
sance Man’: Is Innovation Getting Harder.” Review of Economics and Statistics
76 (1): 283– 317.
174 Ajay Agrawal, John McHale, and Alexander Oettl
Jones, Charles. 1995. “R&D-Based Models of Economic Growth.” Journal of Po-
litical Economy 103 (4): 759– 84.
———. 2005. “Growth and Ideas.” Handbook of Economic Growth, vol. 1B, edited
by Phillipe Aghion and Steven Durlauf. Amsterdam: Elsevier.
Kauff man, Stuart. 1993. The Origins of Order. Oxford: Oxford University Press.
LeCun, Yann, Yoshua Bengio, and Geoff rey Hinton. 2015. “Deep Learning.” Nature
521:436– 44.
Levinthal, Daniel. 1997. “Adaptation on Rugged Landscapes.” Management Science
43:934– 50.
Mamoshina, Polina, Armando Vieira, Evgeny Putin, and Alex Zhavoronkov. 2016.
“Applications of Deep Learning in Biomedicine.” Molecular Pharmaceutics
13:1445– 54.
Mokyr, Joel. 2002. The Gifts of Athena: Historical Origins of the Knowledge Economy.
Princeton, NJ: Princeton University Press.
———. 2017. “The Past and Future of Innovation: Some Lessons from Economic
History.” Paper presented at the NBER Conference on Research Issues in Artifi -
cial Intelligence, Toronto, Sept. 2017.
Mullainathan, Sendhil, and Jann Spiess. 2017. “Machine Learning: An Applied
Econometric Approach.” Journal of Economic Perspectives 31 (2): 87– 106.
National Science and Technology Council. 2011. “Materials Genome Initiative for
Global Competitiveness.” Washington, DC.
Nelson, Richard, and Sidney Winter. 1982. An Evolutionary Theory of Economic
Change. Cambridge, MA: Harvard University Press.
Nielsen, Michael. 2012. Reinventing Discovery: The New Era of Networked Science.
Princeton, NJ: Princeton University Press.
Nordhaus, William. 2015. “Are We Approaching an Economic Singularity? Informa-
tion Technology and the Future of Economic Growth.” NBER Working Paper
no. 21547, Cambridge, MA.
Rivkin, Jan. 2000. “Imitation of Complex Strategies.” Management Science 46:824– 44.
Romer, Paul. 1990. “Endogenous Technical Change.” Journal of Political Economy
94: S71– 102.
———. 1993. “Two Strategies for Economic Development: Using and Producing
Ideas.” In Proceedings of the World Bank Annual Conference on Development Eco-
nomics, 1992. Washington, DC: The World Bank.
———. 2008. “Economic Growth.” In The Concise Encyclopaedia of Economics,
Library of Economic Liberty. http:// www .econlib .org/ library/ Enc/ Economic
Growth .html.
Rosenberg, Nathan. 1998. “Chemical Engineering as a General Purpose Tech-
nology.” In General Purpose Technologies and Economic Growth, edited by Elhanan Helpman. Cambridge, MA: MIT Press.
Weitzman, Martin. 1998. “Recombinant Growth.” Quarterly Journal of Economics
113:331– 60.
Zeira, Joseph. 1998. “Workers, Machines, and Economic Growth.” Quarterly Journal
of Economics 113 (4): 1091– 117.
6
Artifi cial Intelligence
as the Next GPT
A Political- Economy Perspective
Manuel Trajtenberg
6.1 Introduction
Artifi cial intelligence (AI) and related technologies are being heralded
as “the next big thing,” one that promises to revolutionize many areas of
economic activity and thus to have a profound impact on economic growth.
However, the rise of AI coincides with a recent wave of pessimism in terms
of productivity growth, expressed forcefully by prominent economists such
as Larry Summers (2016), and more thoroughly by Robert Gordon (2016).
Side by side with the gloom, the new “technology enthusiasts” envision
a not- too- distant future in which AI will displace most ( all?) human occu-
pations while unleashing tremendous gains in productivity. This view poses
once again disturbing questions about the future of employment, the distri-
butional consequences of mass displacement, and so forth.
Nobody holds the crystal ball, hence rather than arguing about the inscru-
table future, it is at least as important to inquire into what we can learn from
history regarding episodes like this, that is, the appearance of a major new
technology that is posed to have profound economic implications. Of course,
the future is never a replay of the past, but it may provide a useful benchmark
against which to assess the unfolding of the new technology.
Mokyr (2017) sounds a cautionary note in that regard: ever since the
dawn of the Industrial Revolution in the late eighteenth century, both the
pessimists and the enthusiasts have almost invariably been proven wrong.
Manuel Trajtenberg is a professor in the Eitan Berglas School of Economics at Tel Aviv University and a research associate of the National Bureau of Economic Research.
This is a follow-up to my discussion at this conference of Joel Mokyr’s paper “The Past and the Future of Innovation: some lessons from Economic History.” For acknowledgments, sources of research support, and disclosure of the author’s material fi nancial relationships, if any, please see http:// www .nber .org/ chapters/ c14025.ack.
175
176 Manuel Trajtenberg
Moreover, Mokyr dismantles with solid historical and present day evidence
Gordon’s claim that technological advance is bound to slow down in a deter-
ministic fashion (in particular, the claim that “all low- hanging scientifi c and
technological fruit has already been picked”).
However, nothing can be taken for granted—as Mokyr skillfully describes,
institutions (including government policies) may play a key role enabling or
retarding innovation. This is precisely the focus here: given that AI is poised
to emerge as a powerful technological force, I discuss ways to mitigate the
almost unavoidable ensuing disruption, and enhance AI’s vast benign poten-
tial. This is particularly important in present times, in view of political-
economic considerations that were mostly absent in previous historical epi-
sodes associated with the arrival of new general purpose technologies.
6.2 Is This Time Diff erent? The Political Economy
of Technological Disruptions
The presumption here, well argued in other papers in this conference,1 is
that AI has the potential of becoming a general purpose technology (GPT)
in the foreseeable future, 2 thus bringing about a wave of complementary
innovations in a wide and ever- expanding range of applications sectors.
Such sweeping transformative processes always result in widespread eco-
nomic disruption, with concomitant winners and losers.
The “winners” are primarily those associated with the emerging GPT
sector itself, and those that are at the forefront of the deployment of the
GTP in the main applications sectors. They tend to be young, entrepre-
neurial, and equipped both with the technical knowledge and the skills that
are made relevant by the new GPT. The labor force composition of Silicon
Valley off ers a grand view of who are the winners in the present informa-
tion and communication technologies (ICT)/ internet era. There are further
winners in those sectors that are ancillary to the core GTP circle, be it in
services
that directly benefi t from the growth of the GPT (e.g., the venture
capital (VC) industry, patent lawyers, designers, etc.), or in others that just
ride on the localized boom (e.g., upscale restaurants and entertainment,
gyms, tourism, etc.).
The “losers” are mostly those employed in sectors that structurally cannot
benefi t from the unfolding GPT (“laggards”), and those in industries where
the adoption of the new GPT renders many existing competencies and skills
obsolete, thus bringing about massive layoff s. They tend to be middle- aged,
have lower than average educational levels, and reside in areas that do not
have much diversifi ed sources of employment.
As economists, we tend to view the big sweep of economic growth since
1. See Cockburn, Henderson, and Stern (chapter 5, this volume).
2. See Bresnahan and Trajtenberg (1995).
AI as the Next GPT: A Political- Economy Perspective 177
the Industrial Revolution as the very embodiment of the “Idea of Progress”
(as conceived in the Enlightenment), and hence the rate of growth of gross
domestic product (GDP) as an unequivocal uptick in the welfare of society
as a whole. Sure, we do acknowledge that there are distributional conse-
quences, and sure, ever since Pareto we know that we are not allowed to
“sum-up utilities” (and thus the “minuses” of losers do not cancel out with
the “pluses” of winners). But those half- hearted qualifi cations become just
lip service—the truth is that we rarely dwell into the balance of winners and
losers, and in particular we do not pay much attention to the later. Para-
phrasing the well- known dictum of Isaac Newton, we may say that
We enjoy today higher standards of living because we are standing on the
broken backs of those that paved the way for technological progress, but
did not live long enough to benefi t from it.
Partly in response to these inequities, the post- World War II era saw the
creation of the welfare state, including unemployment insurance, transfers
to the disadvantaged, some form of health insurance, retraining programs,
and so forth. These “safety nets” were supposed to provide a reasonable
palliative to “losers,” but the truth is that we still do not have eff ective mecha-
nisms to prevent or ameliorate the costs of major technologically induced
transformations.3 Moreover, existing safety nets will quite likely fail to cope
with the juxtaposition of two new and powerful phenomena: (a) much larger
fl ows of GPT- displaced workers and (b) a new “great demographic transi-
tion.” Let us examine each in turn.
Regarding the extent of displacement: technological change always causes
disruption, as brilliantly articulated by Schumpeter’s notion of “creative
destruction.” Furthermore, there are infl ection points as a new GPT starts
working its way through the economy, when in relative short notice very
many sectors, competencies, and skills became laggards and obsolete.
However, as clearly envisioned in this conference, AI in its various incar-
nations seems to go much further, in that it has the potential to replace a
very wide swath of human occupations. Many argued forcefully that there
are no occupations that cannot be eventually replaced by AI, and that the vast majority of present occupations will indeed vanish within a generation.
The consensual view seems to be that a large proportion of employment
as we know it today will give way to smart machines, and therefore that x
percent of workers will be displaced, whereby x is thought to be signifi cantly larger than in previous GPTs. At the same time, the extent to which new, presently unforeseeable occupations may arise (denote them y percent) seems
to be constrained by the very nature of AI: presumably AI will be able to
3. Typically, these safety nets function reasonably well when dealing with the consequences of not- too- pronounced business cycles or with small, temporarily deprived groups of the population. Not so when there are major structural transformations or when the underlying conditions that led to welfare dependency become permanent.
178 Manuel Trajtenberg
perform most of the new tasks, and hence they will not constitute a good
enough counterbalance to the disappearing jobs, as has been the case in the
past. The prevailing view is then that the net displacement of employment
( x – y) will turn out to be signifi cantly larger for AI than in previous epi-
sodes of technological disruption, posing a serious challenge to traditional
economic policies.
The second part of the challenge entails a steep drop in birth rates together
with the extension of life expectancy (which has been steadily growing for
well over a century). These powerful demographic forces have resulted in
aging populations, with the concomitant increase in the dependency ratio
and the looming threat on the long- term viability of the pension system.
Notice that life expectancy is now increasing well past the retirement age, so
that a typical person in her fi fties contemplates a further stretch of twenty-
fi ve to thirty years of life. Thus, the prospect of being permanently laid off
at that stage in life has dire consequences for the displaced individual as
much as for society as a whole.
The joint eff ect of a large infl ux of displaced workers at the seemingly
unique infl ection point posed by AI, together with their longer life expec-
tancy, may thus create a formidable challenge that even the most advanced
welfare state will be hard pressed to cope with. Put diff erently, we cannot
aff ord to have many more, and longer- lived, unemployed or underemployed
people. This is what is at stake with the advent of AI.
There is yet another signifi cant development that magnifi es the challenge,
and that is the democratization of expectations. The growth in income per
capita involves not only a rise in material standards, but in other no less
important dimensions of well- being, including reduced uncertainty and a
concomitant heightened sense of control over our own lives, which entails
also the expectation of having a voice in processes that aff ect us (Hirschman
1970). Not by coincidence, economic growth and expanding democracy have
more often than not gone hand in hand within, as well as across, countries.
The Luddites of the early nineteenth century surely had their voice heard,
as did their like- minded emulators over the following decades. However, they
could hardly expect to make a dent on their fate: democracy was still highly
limited and living standards still very low for the vast majority, so that most
people were just consumed by the need to provide for their basic needs.
Much has changed since, and nowadays virtually every individual in
advanced western countries has come to expect to be entitled, at least in
principle, to full participation in every realm of society: the political, the
economic, and the cultural. The expectation is not just to vote in periodic
elections, but to have an infl uence via “participatory democracy”; not just
to hold a job, but to partake in the benefi ts of economic growth—this is
what constitutes “the democratization of expectations.”
We claim that in such co
ntext it has become much harder to have some
(many?) bear the costs of technological disruption (the losers), while others
AI as the Next GPT: A Political- Economy Perspective 179
reap the benefi ts (the winners). Moreover, the losers have become much
more skeptical of the vague promise that eventually the benefi ts will “trickle down” to them as well. With good reason: experience shows that the losers
typically remain on the downside, even if the welfare state somehow softens
their human costs. In advanced, democratic societies, people have become
more impatient, more demanding of government, more intolerant of false
promises, as well as of collective failures. Again, this should be surely con-
sidered a highly positive by-product of the rise in living standards.
The sharp split between winners and losers, if left to its own, may have
serious consequences far beyond the costs for the individuals involved:
when it coincides with the political divide, it may threaten the very fabric of
democracy, as we have seen recently both in America and in Europe. Thus,
if AI bursts into the scene and triggers mass displacement of workers, and
demography plays out its fateful hand, the economy will be faced with a
formidable dual challenge that may require a serious reassessment of policy
options:
• Governments may have to assume a wider responsibility for navigating
eff ective transitions from old to new GPTs, and not just for alleviating
some of the costs. As said above, the democratization of expectations
will not allow just for cosmetic adjustments—the political economy of
it will eventually force real change.
• In so doing, governments may have to consider courses of action
aimed inter alia at reducing signifi cantly the number of those that fall
in between the cracks during such transitions: actual and potential los-
ers are bound to become much less tolerant of their fate. This should be
done not by attempting to slow down the pace of technical change (that
would be silly and ineff ectual), but on the contrary, by making sure that
many more can be brought to partake in it.
6.3 From Threat to Promise: Strategies for the AI- GPT Era
In order to meet the above- mentioned challenges, governments will have
to design innovative strategies in the following key areas: