by Ajay Agrawal
much replication, etc.). Some attention to the “direction” may bring much
larger returns.
6.4 Concluding
Remarks
The historical record suggests that dismal prophecies about the economic
and social impact of great technological advances rarely come to pass. Thus,
with AI poised to emerge as the new GPT, we should not necessarily envision
a future whereby humans will be rendered obsolete and mass unemployment
will be the “new normal.” At the same time, as many occupations will indeed
vanish, and many others will undergo signifi cant changes, it is important to
inquire into what sort of strategies may ameliorate the detrimental eff ects
of AI and enhance the positive ones. This is all the more important given
that in the twenty- fi rst century the public at large has much less tolerance
for bearing the costs of technical change and higher expectations for sharing
into its benefi ts here and now.
Therefore, we need to anticipate the required institutional changes, experi-
ment in the design of new policies (particularly in education and skills de-
velopment) in the professionalization of service occupations, and in aff ect-
ing the direction of technical advance. Furthermore, economists possess a
vast methodological arsenal that may prove very useful for that purpose—
we should not shy away from stepping into this area, since its importance
for the economy cannot be overstated.
186 Manuel Trajtenberg
References
Bresnahan, Timothy, and Manuel Trajtenberg. 1995. “General Purpose Technolo-
gies ‘Engines of Growth’?” Journal of Econometrics 65 (1): 83– 108.
Gordon, Robert J. 2016. The Rise and Fall of American Growth. Princeton, NJ:
Princeton University Press.
Heckman, James, Tim Kautz, Ron Diris, Bas ter Weel, and Lex Borghans. 2014.
“Fostering and Measuring Skills: Improving Cognitive and Non- Cognitive
Skills to Promote Lifetime Success.” Report prepared for the Organisation of
Economic Co- operation and Development, Paris. http:// www .oecd .org/ edu
/ ceri/ Fostering- and- Measuring- Skills- Improving- Cognitive- and- NonCognitive
- Skills- to-Promote- Lifetime- Success .pdf.
Hirschman, Albert. 1970. Exit, Voice and Loyalty. Cambridge, MA: Harvard Uni-
versity Press.
Mokyr, Joel. 2017. “The Past and the Future of Innovation: Some Lessons from Eco-
nomic History.” Paper presented at the conference on the Economics of Artifi cial
Intelligence, Sept. 2017, Toronto.
Summers, Lawrence H. 2016. “The Age of Secular Stagnation: What It Is and What
to Do About It.” Foreign Aff airs, Feb. 15. http:// larrysummers .com/ 2016/ 02/ 17
/ the- age- of-secular- stagnation/.
II
Growth, Jobs, and Inequality
7
Artifi cial Intelligence, Income,
Employment, and Meaning
Betsey Stevenson
The evolution of artifi cial intelligence (AI) evokes strong emotions in
people. Some imagine a dystopia in which people are replaced by machines.
Machines will develop the content we read, and the entertainment we enjoy.
Artifi cial intelligence will pick our friends and our politicians, and ultimately
take away any sense of human agency. And worst of all, those machines
will deprive us of work. Human beings will lose meaning and income, and
perhaps ultimately, be driven to extinction.
At the other end of the spectrum are those that envision the potential
for utopia. With machines doing all the work, people will have plenty of
income, yet very little unpleasant work to do. Instead, people will spend their
days enjoying art and music. They will pursue their passions unburdened
by the need to provide for their basic wants. They will feed their intellectual
curiosity and fulfi ll the human demand for personal interactions. In short,
people will be able to enjoy their lives with the freedom from time and money
constraints that artifi cial intelligence provides.
So who is right?
7.1 Income Is Not the Problem
Economists think that we know the answer, or at least part of it. Most
economists believe that automation promises a future of higher income that
Betsey Stevenson is associate professor of economics and public policy at the Gerald R. Ford School of Public Policy, University of Michigan; a visiting associate professor of economics at the University of Sydney; a research associate of the National Bureau of Economic Research; a research affi
liate of the Centre for Economic Policy Research; and a research fellow of CESifo.
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/ c14026.ack.
189
190 Betsey Stevenson
stems from the higher productivity that artifi cial intelligence will provide.
In September 2017, the Chicago Booth IGM Forum’s Economic Experts
Panel asked forty- one economists from top universities in the United States
whether they strongly agreed, agreed, were uncertain, disagreed, or strongly
disagreed with the following statement: “Rising use of robots and artifi cial
intelligence in advanced countries is likely to create benefi ts large enough
that they could be used to compensate those workers who are substantially
negatively aff ected for their lost wages.”1
The answer was clear; no one disagreed with that statement. A few econo-
mists—10 percent—were uncertain, and the modal answer was agree, rather
than strongly agree. Yet, it is clear that economists believe that artifi cial intel-
ligence represents an opportunity for substantial economic gains. Indeed,
productivity gains have been at the heart of improvements in living stan-
dards from the beginning of time. And so, it is diffi
cult to imagine a world
in which productivity gains do not generate benefi ts suffi
ciently large that
we could compensate the losers.
Therefore, the relevant question is whether we would compensate the
losers. Here economists are more skeptical. Economics tells us that there
will be income gains, but our social and political structure help determine
how they will be distributed.
7.2 Who Gets the Gains from Automation?
Much of the skepticism about being able to successfully redistribute
income comes from a lack of trust that the political process will successfully
manage redistribution in a world in which income is primarily generated
by capital. The history of the last several decades has certainly not been
encouraging on that front. The share of income held by the top 1 percent
of the population has risen to nearly 20 percent, from around 10 percent
in 1980, while the share going to the bottom 50 percent of the population
has fallen to 12 percent from 20 percent in 1980.2 Currently we are failing
to redistribute the gains from technological advances, and so the concerns
that distribution will be a challenge are supported by our recent past.
7.3 What Will We Do with Ourselves?
Yet, the concern runs deeper than wondering whether as a society we
could manage to redistribute income. Most economists are concerned
about
how we will allocate jobs, and underneath that concern lies a belief that work
matters independent of the earnings that are generated by the work.
1. IGM Economic Experts Panel (2017).
2. World Wealth and Income Database. http:// wid.world/ country/ usa/.
Artifi cial Intelligence, Income, Employment, and Meaning 191
Essentially, many people are skeptical that people could successfully fi nd
engaging and emotionally rewarding ways to spend their time if they were
not working. One of the IGM Forum panelists, Robert Hall, expressed his
concern most concretely: “Those not in the labor force are unhappy and
inclined to opioids.”
So economists are fearful about what will happen if people lose employ-
ment opportunities, yet economic history provides economists with opti-
mism that employment will adapt. Which is why so many economists wonder
what, if anything, will be diff erent about artifi cial intelligence compared to
the industrial revolution or other important periods of rapid technological
change.
Economists’ intuition around the impact of technological change on
employment comes from considering how employment has adapted fol-
lowing previous periods of technological change. Here, once again, econo-
mists have a united view: technological change has not historically reduced
employment. This view of economists is seen in a February 2014 ques-
tion posed to the Chicago Booth IGM Forum’s Economic Experts Panel.
Forty- four economists from top universities in the United States were asked
whether they strongly agreed, agreed, were uncertain, disagreed, or strongly
disagreed with the following statement: “Advancing automation has not
historically reduced employment in the United States.”3
Economists are roughly united in agreeing with this statement, with only
4 percent disagreeing and 8 percent uncertain.4
Yet, when the IGM Economic Experts Panel was asked in September
2017 whether they strongly agreed, agreed, were uncertain, disagreed, or
strongly disagreed with the following statement: “Holding labor market
institutions and job training fi xed, rising use of robots and artifi cial intel-
ligence is likely to increase substantially the number of workers in advanced
countries who are unemployed for long periods.”
This is where economics lends a less clear answer and economists are
divided on this question: 44 percent agree, 26 percent disagree, and 31 per-
cent are uncertain. Is this a contradiction or a diff erent view about artifi -
cial intelligence compared to other technologies? I don’t think it is either.
Instead, I believe these answers refl ect the diff erence in what happens in the
long run versus the short run. In the long run, technological change leads
to prosperity and new jobs arise as we adjust to our new wealth, develop
new skills, and come up with new ways to use human skills. In the short run,
however, there is often a disruption.
3. IGM Economic Experts Panel (2014).
4. The fi gure of 88 percent is adjusted for respondents’ confi dence in their answer. Among all respondents, 76 percent agreed and 9 percent had no opinion.
192 Betsey Stevenson
7.4 The Long Run
One of the confusions around what will happen to employment and
unemployment stems from not separating short- run versus long- run eff ects.
When most of us think about artifi cial intelligence and increased automa-
tion, we are trying to think about what the long- run future holds, and our
intuition comes from considering how growth has changed how people live
across generations. It is not how it has changed our lives over the last fi ve
years, but instead contrasting how we live our lives—and if you are reading
this it involves large periods of intellectual contemplation—with how our
own family members ten generations back spent their lives. In the 1800s,
the vast majority of Americans worked in agriculture and very few of them
spent their time thinking about ideas. Today, 2 percent of Americans are
directly employed in agriculture. There are more people employed in the
public school system than in agriculture. In sum, few of us are in the jobs
or careers that our great- great- great- great grandparents were in and many
of us work in jobs today that did not exist a single generation ago.
One of the IGM panelists, Nancy Stokey, made it clear she was think-
ing about the long run: “If this had been true over the last two centuries,
almost no one would be working anymore.” When you take a really long-
run view, it has to be true that automation has not reduced employment, at
least not at as rapid a pace as the automation has itself occurred. In fact,
many economists regard it as a puzzle that paid work has been remarkably
stable even as nations have become increasingly prosperous, and its citizens
might have been expected to use more of their higher income to choose to
consume more leisure.
7.4.1 In the Long Run, Employment and Hours Worked Have Declined
Yet, despite our intuition, employment has tended to decline with techno-
logical progress. The diff erence between our beliefs about how technological
progress has impacted employment and what has actually happened refl ect
two things. The fi rst is that hours worked and employment has not declined
by as much as one might have predicted. The second is that economists tend
to think about employment in a model in which people who want to work
can fi nd jobs.
Hours of work have declined in most countries with productivity growth.
Figure 7.1 shows average annual hours worked in a handful of developed
countries since 1970. Annual hours worked declined fairly steadily in France,
Germany, and Japan. The United States and the United Kingdom had
smaller declines. Yet in each country, the annual hours worked fell.
To think more broadly about employment, childhood employment has
been almost eliminated in developed countries. And employment of young
adults, those age fi fteen to twenty- fi ve have declined as young people focus
on investing in further human capital. On the other end of the life cycle, life
Artifi cial Intelligence, Income, Employment, and Meaning 193
Fig. 7.1 Average annual hours worked
Source: OECD (2017).
expectancy has increased while retirement ages have fallen in most developed
countries.
Work has declined in terms of the number and share of our life in terms
of hours and days that we are going to spend working. The decline in work
has occurred through the interaction of economic growth with government
policies. For example, extended retirement has been facilitated by govern-
ment pension and retirement programs. The dramatic reduction in child
labor was facilitated by child labor laws. The demand for these programs
and regulations is itself facilitated by the higher income that productivity
growth creates.
Decreases in employment because of childhood education and retirement
are thought to be improvements in living standards and not something we
need or want to fi x. However, they do require income redistribution. Older
generations must support children, either through families or government
redistribution (such as child tax credits, child allowances, child health care
subsidies, etc.). Yet, most people agree that this is an improvement—few are
trying to get kids back into the workforce to fi nancially support themselves.
Something similar is true at the other end of the life cycle. While the elderly
can save for retirement, redistribution allows those who are retired to share
in continued economic growth.
7.5 Short-
Run
Disruption
The real uncertainty with artifi cial intelligence is what will the disruption
be like and how will we manage people through it. Most economists think
194 Betsey Stevenson
there will be people who are hurt through decreased demand for their skills.
There might be longer spells of unemployment and a larger need for worker
retraining. There might be jobs that workers do not want or are not qualifi ed
to do. While we can prepare a new generation for a world in which robots
do many of the jobs, preparing a generation midway through their lives is
harder. People are resistant to starting over, they mourn what they have
lost, and they resent a defi nition of progress that leaves them diminished in
status and income.
The loss of income should be easier to solve than the loss of status. So how
important is work and what do we know about it? Is work about the income
that it generates or about the meaning and order it gives to our days? Much
of the debate about the potential impact of automation on employment is
really a debate about how we will spend our time. So it is useful to separate
out the question of what will we do with our time if the robots take our jobs
from the question of whether we can fi nd a stable and fair distribution of
income in such a scenario. And it is useful to realize that the answers in the
long run may be very diff erent to what happens in the short run. Yet, how
we handle the short run will ultimately infl uence our long- run outcomes.
7.6 There Is Work outside of Employment
Work is a broader concept than paid labor. Paid labor is the result of
a trade- off between leisure, home- produced goods, and market- produced
goods. This matters from a measurement perspective because the 1970s was