Examining the impact of innovation on the economy is no simple task. After all, technology eliminates people’s jobs in the short term, but it also creates new, better jobs in the long term. Technology also boosts economic productivity, but many fear that such gains will not be fairly distributed in a coming age, as many pessimists envision a future economy composed of venture capitalists and their butlers.

Will the rise of robots and artificial intelligence deliver
a prosperous economy? And what policies should we adopt to ensure that all
Americans are included in this prosperity? On this episode, Carl Benedikt Frey
discusses the impacts of technological progress on the economy, past and
present.

Dr. Frey is the Oxford Martin Citi Fellow at Oxford University, where he teaches economics and economic history. In 2013, he co-authored a widely-shared paper with Michael Osborne titled “The Future of Employment: How Susceptible Are Jobs to Computerisation?” in which he estimated that 47 percent of jobs were susceptible to automation. This year, he returned to the subject with his new book, The Technology Trap: Capital, Labor and Power in the Age of Automation.

What follows is a lightly edited transcript of our conversation. You can download the episode here, and don’t forget to subscribe to my podcast on iTunes or Stitcher. Tell your friends, leave a review.

Pethokoukis: At the most recent Democratic debate, there was a mini–debate that broke out on the issue of technology, automation, and jobs. On one side, you had some of the Democratic candidates who seemed very concerned that robots, for lack of a better term, were going to take many jobs, and there was going to have to be a big policy response.

Leading the charge on that side has been dark-horse candidate Andrew Yang, who wants universal basic income because he’s really worried about technological unemployment. In fact, he describes the problem this way on his website:

“In the next 12 years, one out of three American workers are at risk of losing their job to new technologies. And unlike previous waves of automation, new jobs will not appear quickly enough, or in large enough numbers to make up for it. To avoid an unprecedented crisis, we’re going to have to find a new solution unlike anything we’ve done before.”

So that’s the Andrew Yang, “Robots are going to take the jobs” position. That was one group. And then you had Elizabeth Warren, who seemed less concerned. This is what she said:

“So, the data show that we have had a lot of problems with losing jobs. But the principle reason has been bad trade policy. The principle reason has been a bunch of giant, multinational corporations who have been calling the shots on trade. Corporations that have no loyalty to America.”

So you have what
appear to be two different views on the issue. And I’m not sure they’re
actually talking about the same thing. Warren seems to be talking about what
has happened up until now, while Yang is worried about the future. So let me
start by asking you — what has been the impact of automation on jobs, particularly
in the 2000s? We’ve seen a decline in manufacturing jobs at least in the last
few decades. So is this an automation story or a trade story?

Frey: I think it’s a bit of both. I didn’t see the debate,
admittedly, but I think I actually disagree with both of those assessments.
First of all, I think anyone who tries to put the timeline on how fast automation
might happen is bound to be wrong.

It will always involve the human factor — organizations,
legislation, and people will have to adjust and get used to working with the
technology. So I don’t think there’s a way of trying to estimate how fast this
will happen.

It’s a very specific
forecast by Andrew Yang. He says 12 years. There’s a saying about the stock
market: If you want to make a forecast, you either give a number or a year, but
never both. He’s giving both — 12 years, and one out of three Americans.

That’s right — we’ve been accused of doing that as well.
Lots of people think that we published a study back in 2013 which said that 47
percent of American jobs will be gone in 20 years, which is about 13 years by
now. So maybe that’s what Andrew Yang is referring to, I don’t know.

But what we actually did in the study was try to address the
potential scope of automation purely from the perspective of technological
capabilities. We didn’t try to make any assessment of its pace, which would be
shaped by lots of other factors that have little to do with the technology
itself. But I do think, at the same time, that the impact of technology on the
labor market has been very significant in the long run.

It’s obviously been very significant — the majority of
people used to work in agriculture. The reason they don’t now is primarily due
to changes in technology and mechanization of agriculture and the appearance of
better paying manufacturing and service jobs.

If we look to the 2000s in particular, China’s import competition has clearly played a very large role, but the manufacturing output share in the US has been very stable over time. What’s been falling is the manufacturing employment share. So lots of stuff is still being produced in the US, just not by that many people.

So the long-term
story of labor market disruption from technology — where we used to mostly be
farmers, and now very few of us are — is very well known.

In the 2000s you have this interesting trade aspect from the China “trade shock,” as written about by David Autor. But now, is focusing on trade becoming a backward-looking thing? Even that paper by David and his coauthors seems to indicate that the trade story was pretty much at an end. So is it now back to normal, where we’re resuming the regularly scheduled programming when it comes to labor markets, where the key thing is really automation and technological progress once again?

Yes. First of all, manufacturing employment in absolute
terms peaked in 1979, I think. So I think the China shock has been part of a
longstanding process of deindustrialization, and disappearing manufacturing
jobs which has been driven both by trade and technology.

The two are somewhat hard to disentangle, and it’s almost meaningless because both are driven by technology. Without advances in air communications technology, it wouldn’t be feasible for companies to take advantage of cheap labor in countries like China and coordinate production with added distance.

Secondly, looking forward, I think there are a lot of jobs
that can’t be sent abroad, but they can be automated. That’s a growing share of
people working in non-trade sectors of the economy. When autonomous vehicles
arrive, a lot of truck, bus, and taxi drivers are going to be increasingly
exposed to automation, but those jobs aren’t going to be sent to China.

Similarly, there are roughly 3.5 million people working as cashiers in the United States today. They’re not exposed to import competition, but they are exposed to Amazon Go. I do think that there is a tendency, looking forward, that automation will play a bigger role in shaping the labor market, and trade is probably going to play a diminished role.

You mentioned truck
driving, which has become a very common example about how technological
progress is going to cost jobs. We read a lot about autonomous vehicles, and
long-haul trucks seem to be a big target. They’re not navigating Boston city
streets in February which would be very hard to do, but rather they’re
navigating a coast-to-coast interstate highway. They wouldn’t have to deal with
a situation where, for instance, kids are playing and a ball goes into the
street. That seems like a clear-cut example where you will have a lot of
automation, and that will cost a big share of truck drivers as those jobs just
disappear.

So there you go — it
looks like technological progress eliminating jobs for people who may not be
able to find another job very easily. What does that example tell us about
automation going forward?

I think the key point is what you mentioned last — people who
won’t find new jobs very easily. I think the story has not been one of rising
technological unemployment with 47 percent of jobs disappearing and leaving
people unemployed.

The story has clearly been one of deindustrialization, and
creating a situation where the outset of opportunities for, primarily, men with
no more than a high school degree have steadily deteriorated. And I think the
truck driver example is a case-in-point continuation of that.

So it’s not high-skilled and high-income jobs that are the
most exposed to automation going forward. It continues to be a lot of manual,
middle-income jobs that will most likely continue to put pressure on the wages
of those with no more than a high school degree.

To focus on this
example, because it’s so frequently cited: If you read a lot of headlines, it
sounds like autonomous trucks are right around the corner, and there will be
nobody sitting in the cabins. So there will be no humans involved, and this is
going to happen in the next 18 months. How close is that scenario to reality?

Well, my background is as an economist, so I rely mostly on my engineering science colleagues to make assessments along those lines. Most of them think that it’s probably 10 to 15 years away for Level 4 or 5.

That’s where nobody
is sitting in the truck?

Exactly. But that’s from the point of view of technological
capabilities. It’s a separate question for how fast all of the old vehicles
will circulate out of the system. That is probably going to depend a lot on
government policy, and frankly, subsidies that incentivize people to shift to
autonomous driving.

I think it’s also likely that adoption will differ quite
significantly because of safety. It’s much easier to navigate the city of Dubai
than it is to navigate the city of London. So I think it’s almost impossible to
provide a timeline for how long.

But what about
highway truck drivers? Could you have trucks going coast-to-coast on highways,
where there might be someone in the cabin just for handling tricky situations,
but where most of the driving is done by the machine?

Well, most of flying a plane is done by machine as well, but
you still need a pilot for takeoff and landing, and indeed to reassure
passengers as well. That was also true with elevator operators, by the way. So
a lot has to do with attitudes towards technological change as well, which I
think is important to remember.

A lot of people were terrified by the prospect of entering
an elevator with no operator being responsible for their safety, and it
actually took some time for people to understand that the technology was often
safer than the operator. In a similar way, I think it’s going to take some time
for people to adjust to that thought.

And for those reasons, adoption timelines are hard to pin
down. But I think that if we look at it from a view merely of technological
capabilities, it’s probably 10 to 15 years down the road.

What do you think about that example?

On Twitter, I noticed a tweet from an aspiring Congressional candidate who was trying to comment on the emerging automation debate, at least in the Democratic Party. This is the example he decided to use for the impact of automation — I think it illustrates some of your points:

“In 2004, Blockbuster Video had roughly 85,000 employees. In 2019, Netflix only has 5,500 employees. That’s what automation does. It replaces large numbers of jobs to create just a few new jobs.”

Well, it does tell you that there are fewer people in one
certain domain, but it doesn’t tell you anything about the economy as a whole.
And I think that lots of people miss a step — I mean, yes, the smokestack
industry of the late 20^th century did generate a lot of jobs. Where,
if you look at software engineering, there are not that many people employed in
those industries and occupations directly. But those people tend to earn quite
high wages, and their incomes support a lot of jobs in the local service
economy — people go out, take a taxi, go to a hairdresser, go to a shop, so on
and so forth.

And as research by Enrico Moretti suggests, that creates, on
balance, five new jobs in the local service economy. So often when we see such
examples, people haven’t thought too much about the secondary effects.

You address some of that in the book, and you differentiate between the sorts of technologies that replace jobs and the sorts that will also benefit workers and create new jobs or allow them to do their jobs better. Not all automation is alike — and I believe Daron Acemoglu has also written about the different sorts of technology, some of which indeed will put people out of work.

Maybe, though, that
company will be more productive, and maybe that will generate other kinds of
job growth. But there are also other forms of technology which will just help people
do their jobs better and also create new products and services.

It seems like, right
now, we need to know more about the enabling technologies rather than the
replacing technologies. Do we know how to create more of the enabling
technologies as well as the replacing ones?

As you mentioned, Daron Acemoglu and Pascual Restrepo have an excellent paper distinguishing between enabling and replacing technologies. And my book tries to look at their relative importance or prominence in the economy throughout history.

I think that difference episodes of technological change
have been more labor-replacing than others, and some are more augmenting and
enabling, in a happier way. And I think it’s clearly true different
technologies have different impacts on average people — so the telescope, for
example, didn’t replace any occupations doing any preexisting work. It allowed
us to do previously unconceivable things such as looking at the moons of
Jupiter.

If you think about the automatic elevator, it replaced
elevator operators even though there are more elevators in cities than ever
before. So it stands to reason that different technologies will have different
impacts on peoples’ jobs and wages.

Do you think we are
creating enough enabling technology now? That seems to be the concern of many
people — that these technologies are just going to replace workers, and they’re
not going to create new jobs and greater prosperity. Is that an issue, or is
that just because people are focusing on only the obvious cases where someone
had a job and now has a piece of software doing that job?

First of all, I think it’s much easier to look at the
horizon and say, “Okay, these jobs are potentially automatable.” It’s much
harder to try to envision the jobs that may appear in the future.

But with regard to the question of whether we can create
more enabling technologies to counterbalance the replacing ones on the horizon?
That sounds good in theory. But I don’t think we have the knowledge to do that
in practice. It’s very hard to steer technological progress from the top down.
I haven’t heard of any good example where that has been done in a convincing
way, so I just think that we don’t really have the knowledge of how to do that.

Before we get into
the history you have in the book, let me ask one thing about the 2000s
regarding technology and jobs. I was under the impression that the areas where
you were seeing jobs replaced were these middle-skill kinds of jobs.

So with a lower-skilled
job, or a “high-touch” job, you actually had to have someone there. Those are
fine, and the abstract thinking jobs are fine, but those middle-skilled jobs —
I’m not sure what the classic example is, maybe a receptionist — those were the
jobs really being impacted by technology. Is that true, and what sorts of jobs
or workers are most at risk going forward?

Yes, so if we look back to the 1980s, it’s clearly the
middle-skilled, middle-income jobs that have disappeared — the manufacturing
industry, and also a lot of clerical back-office work. Looking forward, our
estimates suggest that it’s mainly low-skill, low-income jobs – receptionists,
security guards, telemarketers – that are most at risk.

It’s clearly the case that some jobs on the higher end of
the skill spectrum are going to be effected as well. Like, document review is
no longer part of billable hours in law firms. Medical diagnostics are
something that computers are getting increasingly good at, but those
technologies are not going to replace entire jobs. They may replace certain
individual tasks in those jobs, and I think that’s an important distinction.

I think that the
concern is that the future economy will just be venture capitalists and their
butlers. Like, those are going to be the only two areas, and that’s not a real
attractive future. I realize that’s an exaggeration, but that does seem like
what people fear.

It is a concern — as I mentioned earlier, we do see jobs
emerging in the very skilled technology sectors where those people, as their
wages rise, tend to demand in-person services which are hard to automate.

Yeah, there is probably a concern that there’s going to be
an economy of software engineers and hot-yoga instructors. But I do think that
there is a lot more to that story.

So, clearly, a lot of jobs are going to appear that we
haven’t even conceived of yet. And also I think it’s important to remember that
many of these in-person service jobs that are emerging are actually more
pleasant. It’s nicer to work in a restaurant or a gym than in a factory, and I
think there’s a lot of non-monetary factors that we need to take into
consideration as well.

Again, the book is
also a history of technological progress. One of the classic questions is — why
did economic growth basically not exist for forever, and then suddenly we got a
lot of growth in the industrial revolution?

As you present it, it seemed like that was not a very interesting question for a long time, as if we’d had it figured out, but now it seems like we’re talking a lot about it again — especially since there have been concerns about productivity stagnating. What was the original catalyst for the industrial revolution as you understand it?

So I don’t believe in mono-causal explanations of economic
development. I think it was a blend of things that came together that made the
industrial revolution, but I think that one very underestimated factor that I
highlight in the book has to do with the structure of political power.

Before the industrial revolution, in most pre-industrial
societies, craft skills were a source of political clout. They didn’t have any
interest in technologies that threatened their jobs and incomes. And, fearing
social unrest, monarchs of governments typically sided with the guilds rather
than pioneers of industry, fearing that they might challenge the political
status quo.

And what happened in Britain was, first of all, with the
rise of Atlantic trade, the new merchant class emerged. They were the ones who
stood to benefit from mechanization because, with rising wages, mechanization
was what allowed them to remain competitive in trade.

Secondly, with the turnpike trusts that paved the way for
the construction of much better road networks in Britain, the integration of
markets meant that the political power of the craft guilds was gradually eroded
because it didn’t extend beyond their own city. And as markets integrated, they
were exposed to a lot more outside competition. Cities like Birmingham and Manchester
emerged from, essentially, previously rural areas which weren’t exposed to any
preexisting crafts guilds, and naturally they also became the masters of the
first industrial revolution.

And thirdly, as the political power of the crafts guilds
diminished, the threat from below diminished as well. But with growing
competition among nation states in Europe, the threat of foreign invasion
became much greater, and it therefore became increasingly hard to align
technological conservatives with the political status quo.

That is why I think that the first industrial revolution
happened in Britain. Science and other things clearly played a very big role as
well, but that was in the later stages of the first industrial revolutions and
even more so in the second industrial revolution.

If you look at a very simple chart of economic growth and rising incomes, it looks like there’s no rise, and then all of a sudden around 1800 it goes straight up. But that’s a very simplified version — there was a period where you had faster growth and productivity, but you weren’t seeing working class living standards rise. It’s called “Engels’ pause,” named after the German philosopher and coauthor of The Communist Manifesto.

Do you think we’re in
a kind of Engels’ pause right now? You read about all of these advances in
robotics and artificial intelligence, but we’re not seeing some explosion of
growth in peoples’ living standards. Is that where we are right now, and in five,
10 years, we’re suddenly going to see rapid productivity and income growth?

I think that’s broadly where we are, although I’m slightly wary of predicting the future because much depends on the political economy of technological change. I don’t think that we can take progression for granted.

Much of the positive attitudes that people have towards
technological progress in the 20^th century was a historical
exception rather than the norm. For most of history, people resisted anything
that they felt threatened their skills and incomes.

At the time when Engels was writing, most economists didn’t
even believe that technology could improve even a lot. What Malthus, Ricardo,
Marx, and Engels all had in common was that they didn’t think that
mechanization prompted wages to rise. It took the late 19^th century
to prove them wrong, and I think looking backwards from them, working people
saw that mechanization translated into higher wages. But if people don’t see
that happening today, I don’t think they are as likely to remain positive to
it.

If we go back to the 1980s, we see wages of prime-age men
with no more than a high school degree have been steadily falling for 40 years.
Their consumption possibilities have clearly not diminished, but I think if
people see diminishing prospects in the labor market, they could potentially
try to opt for policies that aim at slowing down the pace of automation.

And robot taxes are now being discussed on both sides of the
Atlantic. According to recent Pew Research, the majority of Americans think
there should be restrictions on the number of machines that businesses can
implement. So I really don’t think that the progression of technology can be
taken for granted.

Assuming that you
think that more technological progress is a good thing, is your concern that
all the good ideas have taken all of the low-hanging fruit, and now it’s very
hard to generate significant technological advances? Is it a concern about the
technology, or is it, again, this political and social backlash to technology,
where the ultimate barrier is our willingness to accept disruption?

So I’m not very concerned about all of the good ideas having
been arrived at, or that all of the great inventions have already been
invented. I think what economic history quite clearly shows with regard to the
first industrial revolution, the second computer revolution, and presumably
with artificial intelligence, is that it takes a long time for technology to
give a boost to productivity.

Plus, businesses need to figure out how to productively use
the technology and rearrange their production processes accordingly. If you
take the electric motor, for example: In the early days of factory
electrification, companies merely replaced the steam engine with an electric
motor as the essential power source in the factory. And all of the shafts and
counter-shafts that posed constant threat to peoples’ fingers, arms, and lives
remained intact.

It was only after four decades that the engineers figured
out, “Well, actually, we can equip every machine with its own electric motor.
Then you can sequence them according to a natural flow of production.” And
that’s essentially what gave us the age of mass production, it’s what allowed Henry
Ford to produce the Model T at a sufficiently low enough price for it to become
the people’s vehicle. And it gradually spread across industries.

I think we’re at a similar stage with artificial
intelligence today. Many of the AI technologies are just not sufficiently good
yet, but I think it’s important to remember that, to go back to the steam
engine, the early engines were mainly used to drain coal mines, and even then
they didn’t do particularly well. But eventually they became the prime motors
of the first industrial revolution.

So I’m not particularly concerned about the potential
future, or the potential of productivity growth. I’m more concerned about the
second point you raised — the resistance to the technology.

So what do you think
are the key policies to create a supportive political economy for technological
progress at this point?

Well, I think it’s hard to provide any sort of generic
policies.

Andrew Yang would say it’s UBI.

Right, so I don’t believe in that, because I don’t believe
that the challenge lies in people not having sufficient consumption
possibilities. I think that people are faring better as consumers today than
they did a decade, or two or three ago.

The challenge is more in the labor market, and economists
tend to think that the purpose of production is consumption but actually that’s
not true. People attach a lot of meaning to their jobs. I think that one of the
most consistent findings is that people who work are happier than those who
don’t. If you look at neighborhoods where jobs have disappeared, those are
associated with lots of social ills.

So I think the challenge is more about trying to help people
move into meaningful jobs, and making sure that their transition is not too
hard.

Do we know how to do that? You’ve likely been at a lot of conferences where this issue comes up, and someone will bring up, “Oh, well we need education reform, or job retraining.” Well, the job retraining part seems to be something we don’t know how to do particularly well.

So what is the
alternative for the 55 year old person who’s automated out of a job? Do we just
say, “Well, guess what? You have no right for the rest of your life to a
$90,000 a year job. Maybe you only have a right to a $55,000 a year job. That’s
just the way it is — maybe we’ll give you some wage insurance.” Is that really
the only option? Because it doesn’t sound like he’s going to become a coder.

No, I think that’s right. But I think there are certain
things that can be done, at least, to reduce the barriers between jobs and
geographies. Special occupational licenses are frequently mentioned — I think
it’s ridiculous that you need to take two exams to become a hair shampooer in
the state of Tennessee. If you then want to move across states, you might need another new license because they haven’t
harmonized across space.

Similarly, moving between places is made harder by the fact
that more people own their homes. I don’t think there is a particularly good
reason why we should encourage home ownership.

Secondly, if you look at the places where most new jobs are
emerging, they are primarily emerging in cities with skilled populations. You
see that house prices have risen very significantly in those places, but
housing supplies failed to keep pace. And we have a lot of zoning restrictions
that are responsible for that which could be abolished.

Thirdly, one example that is close to my heart because it’s
close to where I grew up in Sweden: Malmӧ was a city that was in trouble for a
long time after its shipyard closed down in the early 1990s. And the revival
essentially came with the construction of the bridge to Copenhagen, which
allowed people in Malmӧ – a city in decline – to tap into the booming labor market in
Copenhagen. Most of the people would commute back to Malmӧ and live
there because housing was cheap. Most of them would spend their money locally,
where they lived, which gave a boost to the local service economy.

I think a lot can be done in terms of actually connecting
declining regions and cities with expanding ones.

Looking
forward, which is the more likely scenario: Because of technology and advances
in living standards we’ll have unemployment and income growth that’s better
than what we’ve seen over the past few decades? Or, will we see slower income
growth and higher unemployment due to automation?

Do you see
a future of enabling technologies with good impacts, or replacing technologies
having the impacts that many anxious workers fear?

I think that one very important point that Daron Acemoglu
and Pascual Restrepo make in one of their recent papers is that even if the
technology is of the replacing sort, the worst periods for labor and
productivity growth are those with replacing technologies. The boost in
productivity growth tends to come decades later when businesses finally figure
out how to restructure their production processes to take advantage of the new
technology.

So I think that, even if this technological progress continues
to be of the replacing sort in the following decades, the long-run benefits for
labor are likely to be greater than these short-term pains they’re going
through now.

Excellent. Carl,
thanks for coming on the podcast.

My pleasure, thank you for having me.

The post The future of employment in an age of automation: A long-read Q&A with Carl Benedikt Frey appeared first on American Enterprise Institute – AEI.