by Ajay Agrawal
in section 14.4 (although we cannot entirely rule out a situation like that
depicted in fi gure 14.2 in which a Pareto improvement is not feasible).
14.2.4 Imperfect
Markets
Let us also consider a fourth case, which does not necessarily refl ect the
specifi c situation with advances in artifi cial intelligence, but which is im-
portant to understand and keep in mind when we evaluate technological
innovations.
Observation 4) If the economy is not fi rst- best ex post, then the utility
pos sibilities frontier may move inward in response to an expansion of production
possibilities. Furthermore, this may even be true with costless redistribution.
When we speak about an economy that is not fi rst- best, we mean an
economy that deviates from the Arrow- Debreu benchmark, that is, that
exhibits market imperfections such as information problems, missing mar-
kets, and price and wage rigidities, which can result in aggregate demand
problems, monopolies and monopsonies, and so forth. Typically, these mean
that the market equilibrium is not Pareto effi
cient. The utility possibilities
frontier represents the maximum utility of workers, given that of entrepre-
neurs, taking the market failures as given.
This case is illustrated in fi gure 14.3. The initial equilibrium is E , but
0
the innovation, which would have led to greater effi
ciency in the absence of
these market imperfections, makes workers worse off —and even with cost-
less redistributions, there is no way that both workers and entrepreneurs
can be better off .
An example, elaborated on by Delli Gatti et al. (2012a, 2012b), were the
agricultural improvements at the end of the nineteenth century and begin-
ning of the twentieth. The result was that agricultural prices plummeted,
and so too did incomes on farms and in the rural sector. But mobility is
costly—moving to the urban sector required capital, and many farmers
saw their capital disappear as the value of their farms decreased. Those
with loans often went bankrupt. Capital market imperfections (based on
information asymmetries) meant that farmers could not borrow to move to
the city to where the new jobs would be created. But as incomes in the rural
sector plummeted, they could not buy the goods made by the manufacturing
362 Anton Korinek and Joseph E. Stiglitz
Fig. 14.3 Potential Pareto frontier with market imperfections
sector. Workers in both the rural and urban sector were worse off .9 This
provides one interpretation of the Great Depression—in the short run, the
innovations at the time proved Pareto inferior.
Another example is given by the now- standard result that free trade may
lead everyone to be worse off in the absence of good risk markets (Newbery
and Stiglitz 1984). That result can be interpreted as one involving technolog-
ical progress. Assume that there was no way of transporting goods between
two countries. A technological advance allows goods to be transported
freely. Then, under the quite plausible conditions postulated by Newbery-
Stiglitz, welfare (of everyone!) in both countries could decrease.
The theory of the second- best (Meade 1955; Lipsey and Lancaster 1956)
reminds us that in the presence of market imperfections, improving the func-
tioning of one market may deteriorate overall welfare. There are reasons to
believe that certain innovations in fi nancial markets, for example, structured
fi nancial products and certain derivatives like credit default swaps, especially
in the absence of appropriate regulations, contributed greatly to the Great
Recession (Financial Crisis Inquiry Commission 2011).10
It is important to appreciate the result described in Observation 4 to
understand how crucial our institutions and our market imperfections are
in determining whether and how large a benefi t society will derive from
innovation.
9. In the central Delli Gatti et al. model, the agricultural sector has constant returns to scale and wages in the urban sector are rigid (e.g, because of effi
ciency wage considerations), so that
the agricultural innovations are unambiguously welfare decreasing. In one variant of the model, where urban wages are fl exible, wage decreases lead to still higher unemployment. Though it is possible that entrepreneurs gain more from the wage reductions than they lose from the loss of sales, social welfare is decreased with suffi
ciently inequality averse social welfare functions.
10. At a theoretical level, Simsek (2013) and Guzman and Stiglitz (2016a, 2016b) have shown that opening up new markets—through fi nancial innovation—can lead to greater volatility in consumption.
AI and Its Implications for Income Distribution and Unemployment 363
14.2.5 Ascertaining Whether the Economy Is
Best Described by Observation 1, 2, 3, or 4
It is not always easy to ascertain which of the four observations in the
four subsections above best describes the economy. Typically, the only thing
that we can observe is that an innovation has made some individuals better
off and some worse off . (In our analysis of AI, we assume that it has made
workers worse off and entrepreneurs better off .) The presumption is that
risk markets for innovation are highly imperfect (so Observation 1 does not
apply), redistributions are costly (so Observation 2 does not strictly apply),
and markets are imperfect (so Observation 3 does not strictly apply.) But the
costs of redistributions may be suffi
ciently low and the market imperfections
suffi
ciently small that fi gure 14.1 still applies: everyone could be made better
off . Alternatively, redistributions may be so costly that fi gure 14.2 applies.
Or market failures may be suffi
ciently large and redistributions suffi
ciently
costly that fi gure 14.3 applies.
We emphasize that which situation we are in depends not just on the pos-
sibilities of ex post redistribution, but on the institutional fl exibility, which
determines the ex ante distribution.
As we noted, the second- best utility possibilities frontier is the outer enve-
lope of all conceivable constrained utility possibilities frontiers, which refl ect
all the conceivable institutional regimes in an economy and all the market
imperfections that the economy may suff er from. By institutional regimes
we mean all explicit tax and redistribution systems (from negative income
tax systems to universal basic income to the regressive tax system currently
in place), intellectual property regimes, job programs, education programs,
but even social norms such as those related to charitable contributions. Mar-
ket imperfections include all the market arrangements that diff er from the
Arrow- Debreu “optimal” benchmark, the conditions that ensure the Pareto
effi
ciency of the market. As we noted earlier, the term embraces imperfec-
tions in information, competition, and risk and capital markets (including
“missing” markets), but also rigidities in factor reallocation or in prices that
determine how easily factors and products reallocate and which may be
particularly important
in the context of technological progress.
Changing any of these institutions or market imperfections has an eff ect
on workers’ welfare. In general, it may be desirable to use a package of
changes to all these institutions to ensure Pareto improvements after tech-
nological change has occurred. For instance, in section 14.4.3, we show that
a combination of a change in the intellectual property regime and a change
in capital taxation can ensure that an innovation is a Pareto improvement.
Finally, we also note that the possibility of achieving a Pareto improve-
ment depends on how broadly we defi ne the classes of individuals that are
aff ected by an innovation. Our earlier example diff erentiated society, for
simplicity, into two categories, workers and entrepreneurs. More generally,
364 Anton Korinek and Joseph E. Stiglitz
diff erent categories of workers, for example, skilled and unskilled workers,
or workers in diff erent sectors or tasks, are diff erentially aff ected by innova-
tion. By the same token, diff erent categories of entrepreneurs or innovators
are diff erentially aff ected by innovation—for example, a given entrepreneur
will generally be worse off if she is out- competed by another’s innovation.
In the limit, if we consider the welfare of every single agent in the economy,
clear Pareto improvements in a strict sense will be very diffi
cult to fi nd. As
a result, our scope of analysis has to be targeted at the level that is relevant
for the question at hand.
From both a political and a macroeconomic perspective, it is desirable
that our welfare analysis focuses on groups that are suffi
ciently broad so
that they matter for the political or economic equilibrium. It may also be
useful to focus on groups that can be targeted with specifi c policy measures.
Having said that, there is also a useful role for social safety nets that insure
single individuals that lose out—for example, an innovator who goes broke
because he was outpaced by a competitor.
14.2.6 Endogenous Technological Progress
A fi fth and last point to emphasize is that there is no fi rst welfare theo-
rem for endogenous innovation. Generally speaking, the private returns to
innovation in an economy diff er from the social returns.11
Observation 5) The privately optimal choice of innovation may move the
utility possibilities frontier inward, even if redistribution is costless.
This implies that there may be benefi ts from intervening in the innovation
process to generate Pareto improvements, for example, by making it less
labor saving (see e.g. Stiglitz 2014b). Again, this does not specifi cally refer
to advances in artifi cial intelligence—it will probably not apply to most
examples of innovation in AI—but it is easy to think of examples where
privately optimal innovation may generate Pareto deteriorations, for ex-
ample, in the context of high- frequency trading in fi nancial markets (see,
e.g., Stiglitz 2014c).
14.2.7 Relationship between Technological Progress and Globalization
Many of the eff ects of technological change in general, and AI in par-
ticular, are similar to those of globalization. Indeed, globalization can be
viewed as a change in technology, that of trading with the rest of the world.
11. It is hard to know who fi rst had this insight. Certainly, Thomas Jeff erson, America’s third president, recognized it when he said that knowledge is like a candle: when it lights another, the light of the fi rst candle is not diminished. In the economics literature, it was clearly articulated by Arrow (1962) and Stiglitz (1987a). For a more recent statement of why social and private returns to innovation diff er, see Stiglitz and Greenwald (2015). These results hold regardless of the intellectual property regime. Poorly designed intellectual property regimes can (and do) impair innovation. For a simple theoretical model, see Stiglitz (2014a); for empirical evidence, see Williams (2010).
AI and Its Implications for Income Distribution and Unemployment 365
In particular, trade of advanced countries with developing countries is
“labor saving” (in the sense of Hicks): the demand for unskilled workers,
or workers in general, decreases, at any given wage, implying that while the
production possibilities curve moves out, and the utility possibilities curve
may move out, the new equilibrium entails workers being worse off , as in
fi gures 14.1 and 14.2. (In the absence of good risk markets, as we noted,
everyone can be worse off , as in fi gure 14.3). Thus, the issue of whether
globalization is welfare enhancing comes back to the question addressed in
this chapter: is it possible to ensure, either through redistributive taxes or
changes in institutions/ rules, that workers are not made worse off . Again,
there is a presumption that the gains to capital (or enterprises) could be
taxed to provide the requisite redistributions.12
As we discuss in greater detail below, one of the side eff ects of innova-
tion and intellectual property rights (IPR) is the creation of market power,
resulting in ex post ineffi
cient outcomes. Similarly, one of the consequences
of globalization is to weaken the market power of workers. This is important
because there is ample evidence that labor markets are far from perfectly
competitive. The requisite compensation and/or off setting changes in insti-
tutional rules to ensure that globalization represents a Pareto improvement
may thus have to be all the greater.
14.3 Technological Progress and Channels of Inequality
There are two main channels through which technological progress may
aff ect the distribution of resources and thus inequality: fi rst, through the
surplus earned by innovators and second, through eff ects on other agents
in the economy.
14.3.1 Surplus Earned by Innovators
Technology is an information good, which implies that it is nonrival, but
it may be excludable. Nonrival means that information can be used with-
out being used up—in principle, many economic actors could use the same
technology at the same time. If information about an innovation is widely
shared, it can be used by all of society and provide welfare benefi ts to any-
body who uses it. The excludable nature of information means, however,
that others can be prevented from either obtaining or using a technology,
for example, by withholding it from the public (e.g., as a business secret) or
12. Although a given country that opens up to trade is always made better off in a fi rst- best world, ensuring a global Pareto improvement after a country reduces its trade barriers may be even more diffi
cult than after technological progress has occurred, since changes in trade bar-
riers aff ect international terms of trade and lead to redistributions across all other countries that can only be undone via cross- country transfers. (See, e.g., Korinek 2016). Furthermore, within each country, gains from trade inherently require changes in relative prices, which means that large redistributions are even more likely than in the case of technological progress.
366 Anton Korinek and Joseph E. Stiglitz
by using social institutions such as intellectual property rights (e.g., copy-
rights or patents). This excludability may provide innovators with market
pow
er that enables them to charge a positive price for the innovation and
earn a surplus.
Society faces a diffi
cult trade- off in determining how to engineer the
optimal level of innovation. In a fi rst- best world, there are no agency prob-
lems in the process of innovation, and an optimal solution would be for the
public to fund innovations and make them freely available to all (see, e.g.,
Arrow 1962). In fact, this model of fi nancing innovation is common for basic
research and has given rise to some very signifi cant innovations in history,
including the invention of the internet. A closely related solution is the pro-
duction of innovations for nonpecuniary rewards, such as, for example, the
prevalence of open source technology, which is widespread in the context
of software and even artifi cial intelligence.13
However, in many circumstances, private agents are superior in producing
innovation, and when they fund innovation, they expect to earn a return.
The surplus earned by innovators then plays an important economic role
because it rewards innovators for what they accomplish—it represents the
economic return to innovation activity. However, as a result there is generally
some market power associated with innovations, especially when there is a
system of IPR in place, and this generally leads to ineffi
ciencies compared
to the fi rst- best allocation in which innovations are distributed as public
goods.14
We distinguish the following two cases, which determine whether inno-
vators earn rents, that is, payoff s in excess of the cost of their innovative
activity:
First, if entry into innovation activity is restricted, then the surplus or net
income earned by innovators is generally greater than the costs of innovation
activity. A natural example of such restrictions is when only a small number
of people are endowed with special skills that enable them to innovate. These
innovators then earn rents based on their exclusive abilities.
Restrictions to innovative activity may also arise from market structure: in
markets with Bertrand competition, the fi rst entrant who develops a costly
innovation may enjoy a monopoly position because any potential competi-
tor knows that if she enters, the incumbent will cut prices to marginal cost