0:07

It's Monday, the 15th of May, 2023.

0:09

I'm Miranda Mitra,

0:11

The Economist's international editor. Welcome

0:14

to Editor's Picks, where you can hear three

0:16

highlights from The Economist this week, read

0:18

aloud. Here with The Economist,

0:21

we've put together a definitive assessment

0:23

of so-called peak China. The

0:25

idea that a slowing economy, a shrinking

0:28

population and other challenges mean

0:30

that the country's relative power is topping

0:32

out. One view is that a China

0:35

in decline would, paradoxically,

0:37

become more dangerous.

0:39

In The Economist this week, we look at several

0:42

aspects of this debate, from China's

0:44

growing military power to Xi Jinping's

0:46

fears of the Communist Party's collapse. And

0:49

we dive into the question of whether and

0:51

when the size of China's economy would surpass

0:54

America's at market exchange

0:55

rates. That was once

0:58

thought inevitable, but now there is some

1:00

doubt. If China's economy

1:02

does overtake America's, it will not

1:04

be by much. Mr. Xi is

1:06

unpredictable, but his country's long-run

1:08

economic prospect is neither triumph nor

1:11

disaster. As our cover story

1:13

explores, that could make the world safer.

1:17

Next, one of our economics correspondents

1:19

explains why your job is probably

1:21

safe from artificial intelligence. We

1:24

argue that the predictions of an imminent economic

1:26

revolution are overstated. And

1:30

finally, we turn to Mexico's

1:32

gangs and their operations beyond

1:34

drugs. We explore how they're becoming

1:36

criminal conglomerates.

1:40

The stories you're about to hear are just a sample of

1:42

what's on offer in the weekly edition of The Economist.

1:45

With a subscription, you can read or listen

1:48

to all of what we do. To get a

1:50

month's subscription for free, head to economist.com

1:53

slash podcast offer. The link

1:55

is in the show notes.

2:03

First up is Chinese power about

2:06

to peak?

2:08

The rise of China has been a defining

2:10

feature of the world for the past four

2:13

decades. Since the

2:15

country began to open up and reform its

2:17

economy in 1978, its GDP

2:21

has grown by a dizzying 9 per

2:23

cent a year on average. That

2:26

has allowed a staggering 800 million

2:28

Chinese citizens to escape

2:30

from poverty. Today

2:33

China accounts for almost a fifth

2:35

of global output. The

2:37

sheer size of its market and manufacturing

2:40

base has reshaped the global

2:42

economy.

2:43

Xi Jinping, who has ruled

2:45

China for the past decade, hopes

2:47

to use his country's increasing heft

2:50

to reshape the geopolitical order

2:52

too. There is

2:55

just one catch. China's

2:57

rapid rise is slowing down.

3:00

Mr Xi promises a great

3:03

rejuvenation of his country in

3:05

the coming decades, but the economy

3:07

is now undergoing something more prosaic,

3:10

a great maturation.

3:12

Whereas a decade ago, forecasters

3:15

predicted that China's GDP would

3:17

zoom past America's during

3:20

the mid-21st century at

3:22

market exchange rates and retain

3:24

a commanding lead, now

3:26

a much less dramatic shift is

3:29

in the offing, resulting in something

3:31

closer to economic parity.

3:34

This change in economic trajectory

3:37

is the subject of fierce debate among

3:39

China watchers.

3:41

They are thinking again about

3:43

China's clout and its rivalry

3:45

with America.

3:47

One view is that Chinese power

3:49

will fall relative to that of its

3:51

rivals, which could paradoxically

3:54

make it more dangerous. In

3:56

a book last year, Hal

3:58

Brands and Michael

3:59

Beckley, two scholars,

4:02

popularised the theory they called

4:04

peak China. The country

4:06

faces decay, they argue, and

4:08

has reached the point where it is strong

4:11

enough to aggressively disrupt the existing

4:14

order but is losing confidence

4:16

that time is on its side.

4:18

Their study opens with an imagined

4:21

war over Taiwan.

4:24

The peak China thesis rests

4:26

on the accurate observation that

4:28

certain tailwinds are turning to headwinds,

4:31

hindering Chinese progress.

4:33

The first big gust comes

4:35

from demography.

4:37

China's working age population

4:39

has been declining for about a decade.

4:43

Last year its population as a whole peaked

4:46

and India has now overtaken

4:48

it. The Communist Party's

4:50

attempts to convince Chinese couples

4:52

to have more children are not working.

4:56

As a result, the UN thinks

4:58

that by mid-century, China's

5:00

working age population could decline

5:02

by over a quarter.

5:04

Wave goodbye to the masses

5:06

of young workers who once filled the world's

5:09

factory.

5:11

Adding workers is one way for

5:13

an economy to grow.

5:15

Another is to make better use

5:17

of the existing population. But

5:20

China's second problem is that output

5:22

per worker is unlikely to rise

5:24

as fast as forecasters once

5:27

hoped.

5:27

More of its resources will go

5:30

to caring for the elderly.

5:32

After decades of building houses,

5:34

roads and railways, spending

5:36

on infrastructure faces diminishing

5:39

returns. Mr Xi's

5:41

autocratic tendencies have made local

5:44

entrepreneurs more nervous, which

5:46

may reduce China's capacity to innovate

5:49

in the long run. Geopolitical

5:51

tensions have made foreign firms eager

5:54

to diversify supply chains away

5:57

from China.

5:58

America wants to hobble.

5:59

China's capabilities in

6:02

some foundational technologies.

6:04

Its ban on exporting certain semiconductors

6:08

and machines to Chinese firms

6:10

is expected to cut into China's GDP.

6:15

All of this is dampening long-run

6:17

forecasts of China's economic

6:19

potential.

6:20

Twelve years ago, Goldman Sachs

6:22

thought China's GDP would

6:25

overtake America's in 2026 and

6:28

become over 50% larger

6:30

by mid-century.

6:32

Last year, it revised

6:34

that prediction, saying China would

6:36

surpass America only in 2035 and peak at less

6:38

than 15% bigger. Others are more gloomy.

6:45

Capital Economics, a research firm,

6:48

argues that the country's economy will never

6:51

become top dog, instead

6:53

peaking at 90% of America's size in 2035. These forecasts are, of course, uncertain.

7:01

But the most plausible ones seem

7:03

to agree that China and America will

7:06

approach economic parity in the next

7:08

decade or so, and remain

7:10

locked in this position for decades

7:12

to come.

7:14

How might China handle this

7:17

flatter trajectory? In

7:19

the most optimistic scenario,

7:21

Mr Xi would make changes to boost

7:23

productivity growth.

7:25

With income per person less

7:27

than half of America's, China's

7:30

population will be keen to improve

7:32

their living standards. He

7:34

could try to unleash growth by giving the

7:36

animal spirits of China's economy

7:39

freer rain and his people

7:41

more freedom of movement. The

7:43

Chinese government could stop relying

7:46

on wasteful state-owned banks and enterprises

7:49

to allocate capital. And

7:51

it could adopt a less prickly posture

7:53

abroad, easing geopolitical

7:56

tensions and reassuring firms

7:58

that it is safe to do business. in China.

8:02

Such reforms might ultimately

8:04

make China more powerful but

8:06

also one would hope less aggressive.

8:10

The trouble is that Mr. Xi, who is 69 and now probably

8:14

China's ruler for life, shows

8:16

no sign of embracing economic

8:19

or political liberalisation.

8:22

Pessimists fear that China will become

8:24

more combative as its economic

8:26

trajectory falters. There

8:29

are plenty of reasons to think this plausible.

8:32

Mr. Xi stokes a dangerous nationalism

8:35

to persuade ordinary Chinese that critics

8:38

of his rule are slighting China

8:40

itself. China's

8:42

military budget is forecast to rise

8:44

by over 7% this year in

8:47

line with nominal GDP.

8:49

Its military spending is lower

8:52

than America's but still catching

8:54

up. Its navy could be 50% bigger

8:57

than America's by 2030 and

9:00

its nuclear arsenal will almost

9:02

quadruple by 2035.

9:05

Beijing's economic power

9:07

may be peaking but no

9:10

other country is so capable of

9:12

challenging America globally, right

9:15

messers Brands and Beckley.

9:18

Yet the most likely scenario is

9:20

in the middle ground. The

9:22

speed of China's rise in the past two

9:24

decades has been destabilizing,

9:27

forcing adjustments in the global economic

9:29

and geopolitical order.

9:31

That phase of intense economic

9:34

disruption may now be over

9:36

and for all its troubles China's

9:38

economy is unlikely to shrink

9:41

triggering the kind of nihilistic

9:43

and destructive thinking that Messer's

9:45

Brand and Beckley fear.

9:48

Mr. Xi is unpredictable

9:50

but his country's long-run economic prospect

9:53

is neither triumph nor disaster.

9:56

Faced with decades of being

9:58

a near-peer of America's

9:59

America, China has good reason

10:02

to eschew hubris and resist

10:04

invading Taiwan. A

10:07

crucial question is whether the superpowers

10:10

can avoid misreading each other's intentions

10:13

and thus stumbling into a conflict.

10:16

Next week, we will examine

10:18

America's global leadership and

10:20

how it should respond to China in

10:23

the coming age of superpower parity.

10:27

Before we go on, we're always trying

10:29

to improve our podcasts, and that's why

10:31

we'd like you to help.

10:33

Tell us what you think of our shows by filling out

10:35

our new listener survey. To take

10:37

part, visit economist.com

10:40

slash EP survey. That's

10:42

all one word. It only takes a few minutes.

10:44

And thank

10:45

you.

10:59

Next, artificial intelligence and

11:01

jobs. Beyond the hype. The

11:05

age of generative artificial intelligence

11:08

has well and truly arrived. Open

11:11

AIs chatbots, which use

11:13

large language model or LLM

11:15

technology, got the ball rolling

11:18

in November. Now barely a

11:20

day goes by without some mind

11:22

blowing advance. An AI

11:25

powered song featuring a fake

11:27

Drake and The Weeknd recently

11:30

shook the music industry. Programs

11:32

which convert text to video

11:35

are making fairly convincing content.

11:38

Before long, consumer products such

11:40

as Expedia, Instacart and

11:42

OpenTable will plug into open

11:45

AIs bots, allowing people

11:47

to order food or book a holiday by

11:50

typing text into a box. A

11:53

recently leaked presentation reportedly

11:55

from a Google engineer suggests

11:58

the tech giant is worried about how easy

12:00

it is for rivals to make progress.

12:03

There is more to come. Probably

12:05

a lot more.

12:07

The development of AI raises profound

12:10

questions. Perhaps most pressing,

12:12

though, is a straightforward one. What

12:15

does this mean for the economy? Many

12:18

have grand expectations. New

12:21

research by Goldman Sachs, a bank,

12:23

suggests that widespread AI adoption

12:26

could eventually drive a 7% or almost

12:30

$7trillion increase in

12:32

annual global GDP over

12:35

a 10-year period. Academic

12:37

studies point to a 3 percentage

12:40

point rise in annual labour

12:42

productivity growth in firms

12:44

that adopt the technology, which

12:46

would represent a huge uplift in incomes

12:49

compounded over many years.

12:52

A study published in 2021 by

12:54

Tom Davidson of Open

12:56

Philanthropy, a grant-making outfit,

12:59

puts a more than 10% chance on

13:02

explosive growth, defined

13:04

as increases in global output of

13:07

more than 30% a year sometime

13:09

this century. A few economists,

13:12

only half jokingly, hold out

13:15

the possibility of global incomes becoming

13:18

infinite. Financial

13:20

markets, however, point to rather more

13:22

modest outcomes. In the

13:24

past year, share prices of companies

13:27

involved in AI have done worse

13:29

than the global average, although they have

13:31

risen in recent months. Interest

13:34

rates are another clue. If people thought

13:36

that the technology was going to make everyone

13:38

richer tomorrow, rates would rise

13:41

because there would be less need to save. Inflation-adjusted

13:45

rates and subsequent GDP growth

13:48

are strongly correlated, notes research

13:50

by Basil Halperin of the Massachusetts

13:53

Institute of Technology, or MIT,

13:56

and colleagues. Yet since the

13:58

hype about AI began in November, the research has been published. November,

14:00

long-term rates have fallen. They

14:03

remain very low by historical standards.

14:06

Financial markets, the researchers conclude,

14:09

are not expecting a high probability

14:11

of AI-induced growth acceleration

14:14

on at least a 30-50 year time horizon. To

14:18

judge which group is right, it

14:21

is helpful to consider the history of previous

14:23

technological breakthroughs. This

14:26

provides sucker to investors, for

14:29

it is difficult to make the case that

14:31

a single new technology by itself

14:34

has ever radically changed the economy,

14:37

either for good or ill. Even

14:39

the industrial revolution of the late

14:42

1700s, which many people believe was the result

14:45

of the invention of the spinning genny, was

14:47

actually caused by all sorts of factors

14:49

coming together, increasing

14:51

use of coal, firmer property rights,

14:54

the emergence of a scientific ethos

14:57

and much more besides. Perhaps

15:00

most famously in the 1960s,

15:03

Robert Fogel published a work about America's

15:05

railways that would later win him

15:07

a Nobel Prize in economics. Many

15:10

thought that rail transformed America's

15:13

prospects, turning an agricultural

15:15

society into an industrial

15:17

powerhouse. In fact

15:20

it had a very modest impact, Fogel

15:22

found, because it replaced technology

15:25

such as canals that would have

15:27

done just about as good a job. The

15:29

level of per-person income that America

15:32

achieved by January 1st 1890 would

15:35

have been reached by March 31st 1890 if railways had

15:40

never been invented. Of

15:43

course no one can predict with any certainty

15:45

where a technology as fundamentally

15:48

unpredictable as AI will take

15:50

humans. Runaway growth

15:52

is not impossible, nor is

15:55

technological stagnation. But

15:57

you can still think through the possibilities. and

16:00

so far at least it seems as though Fockels

16:03

railways are likely to be a useful

16:05

blueprint. Consider three

16:08

broad areas, monopolies, labour

16:11

markets and productivity.

16:14

A new technology sometimes creates

16:16

a small group of people with vast

16:19

economic power. John D.

16:21

Rockefeller won out with oil refining

16:24

and Henry Ford with cars. Today,

16:27

Jeff Bezos and Mark Zuckerberg

16:29

are pretty dominant thanks to tech. Many

16:32

pundits expect that before long

16:35

the AI industry will generate

16:37

huge profits. In a recent

16:39

paper, Goldman's analysts estimate

16:42

in a best case scenario generative

16:45

AI could add about $430 billion

16:49

to annual global enterprise software

16:52

revenues. Their calculation

16:54

assumes that each of the world's 1.1 billion

16:57

office workers will adopt a few

17:00

AI gizmos, paying around $400

17:02

in total each. Any

17:06

business would be glad to capture some

17:08

of this cash, but in macroeconomic

17:11

terms, $430 billion simply does not move the dial. Assume

17:17

that all of the revenue turns into profits,

17:20

which is unrealistic, and that all

17:22

of these profits are earned in America, which

17:24

is a tad more realistic. Even

17:27

under these conditions, the ratio

17:29

of the country's pre-tax corporate

17:31

profits to its GDP would

17:33

rise from 12% today to 14%. That

17:39

is far above the long-run average, but

17:41

no higher than it was in the second quarter

17:43

of 2021. These

17:46

profits could go to one organisation,

17:49

maybe open AI. Monopolies

17:52

often arise when an industry has high

17:54

fixed costs, or when it is

17:56

hard to switch to competitors.

17:59

customers had no alternative to Rockefeller's

18:02

oil for instance and could not

18:04

produce their own. Generative

18:07

AI has some monopolistic

18:09

characteristics. GPT-4,

18:12

one of OpenAI's chatbots, reportedly

18:15

cost more than $100 million

18:17

to train, as some few

18:19

firms have lying around. There

18:22

is also a lot of proprietary knowledge about

18:24

data for training the models, not

18:27

to mention user feedback.

18:29

There is however little chance of a single

18:31

company bestriding the entire

18:34

industry. More likely is

18:36

that a modest number of big firms compete

18:39

with one another, as happens in aviation,

18:42

groceries and search engines. No

18:45

AI product is truly unique,

18:48

since all use similar models.

18:51

This makes it easier for a customer to

18:53

switch from one to another. The

18:55

computing power behind the models

18:58

is also fairly generic. Much

19:00

of the code, as well as tips and tricks,

19:03

is freely available online, meaning

19:06

that amateurs can produce their own models,

19:09

often with strikingly good results.

19:12

There don't appear today

19:13

to be any systemic moats

19:16

in generative AI. A

19:18

team at Andresen Horowitz, a

19:20

venture capital firm, has argued. The

19:23

recent leak, purportedly from Google,

19:26

reaches a similar conclusion. The

19:29

barrier to entry for training and

19:31

experimentation has dropped from

19:33

the total output of a major research

19:35

organisation to one person,

19:37

an evening and a beefy laptop.

19:41

Already there are a few generative AI

19:43

firms worth more than $1 billion.

19:47

The biggest corporate winner so far from

19:49

the new AI age is not

19:51

even an AI company. At

19:54

Nvidia, a computing firm which

19:56

powers AI models, revenue

19:58

from data centres

19:59

is soaring. Although

20:02

generative AI may not create

20:04

a new class of robber barons,

20:07

to many people that will be cold comfort.

20:10

They are more concerned with their own economic

20:12

prospects, in particular whether

20:15

their job will disappear. Terrifying

20:18

predictions abound. Taina

20:21

Elundu of OpenAI and

20:23

colleagues have estimated that around 80%

20:26

of the US workforce could have

20:28

at least 10% of their work tasks

20:31

affected by the introduction of LLMs.

20:33

Edward Felton

20:36

of Princeton University and colleagues conducted

20:38

a similar exercise. Legal

20:41

services, accountancy and

20:44

travel agencies come out at

20:46

or near the top of professions most

20:49

likely to face disruption.

20:52

Others have issued gloomy predictions before,

20:55

in the 2000s many feared

20:57

the impact of outsourcing on

20:59

rich world workers. In 2013,

21:03

two at Oxford University issued a widely

21:05

cited paper that suggested

21:08

automation could wipe out 47%

21:11

of American jobs over the subsequent

21:14

decade or so. Others

21:16

made the case that even without widespread

21:18

unemployment there would be hollowing

21:21

out where rewarding well

21:23

paid jobs disappeared and

21:25

mindless poorly paid roles took

21:27

their place. What actually

21:30

happened took people by surprise.

21:33

In the past decade the average

21:35

rich world unemployment rate has

21:37

roughly halved. The share

21:40

of working age people in employment is

21:42

at an all time high. Countries

21:45

with the highest rates of automation and

21:47

robotics such as Japan, Singapore

21:49

and South Korea have the least

21:51

unemployment. A recent

21:54

study by America's Bureau of Labour Statistics

21:57

found that in recent years jobs classified

21:59

as at risk from new technologies,

22:02

did not exhibit any general tendency

22:05

toward notably rapid job

22:07

loss. Evidence for

22:09

hollowing out is mixed. Measures

22:11

of job satisfaction rose during

22:14

the 2010s. For most of

22:16

the past decade, the poorest Americans

22:19

have seen faster wage growth than the

22:21

richest ones. This

22:23

time could be different. The

22:25

share price of Chegg, a firm which

22:28

provides homework help, recently

22:30

fell by half after it admitted

22:33

chat GPT was having

22:35

an impact on our new customer growth

22:37

rate. The chief executive

22:39

of IBM, a big tech firm, said

22:42

that the company expects to pause hiring

22:44

for roles that could be replaced by AI

22:47

in the coming years. But

22:49

are these early signs a tsunami

22:52

is about to hit?

22:53

Perhaps not.

22:55

Imagine a job disappears when

22:58

AI automates more than 50% of

23:01

the tasks it encompasses. Or

23:03

imagine that workers are eliminated in

23:06

proportion to the total share of economy-wide

23:09

tasks that are automated. In

23:12

either case, this would, following Ms Elundu's

23:15

estimates, result in a net

23:17

loss of around 15% of

23:20

American jobs.

23:21

Some folk could move to industries

23:24

experiencing worker shortages, such

23:26

as hospitality.

23:28

But a big rise in the unemployment

23:30

rate would surely follow, in line

23:32

maybe with the 15% briefly

23:35

reached in America during the worst

23:37

of the COVID-19 pandemic in 2020.

23:40

Yet this scenario

23:42

is unlikely to come to pass. History

23:46

suggests job destruction happens

23:48

far more slowly. The

23:51

Automated Telephone Switching System,

23:53

a replacement for human operators,

23:56

was invented in 1892. took

24:00

until 1921 for

24:02

the Bell system to install their first

24:04

fully automated office. Even

24:07

after this milestone, the number of American

24:10

telephone operators continued

24:12

to grow, peaking in the mid-20th

24:14

century at around 350,000. The occupation did

24:16

not mostly disappear until the 1980s,

24:23

nine decades after automation

24:26

was invented. AI

24:28

will take less than 90 years to sweep

24:30

the labour market, LLMs are

24:32

easy to use and many experts

24:35

are astonished by the speed at which the

24:37

general public has incorporated

24:39

chat GPT into their lives.

24:42

But reasons for the slow adoption of

24:44

technology in workplaces will

24:47

also apply this time around. In

24:50

a recent essay, Mark Andresen

24:52

of Andresen Horowitz outlined

24:55

some of them. His argument focuses

24:57

on regulation. In bits

24:59

of the economy with heavy state involvement,

25:02

such as education and healthcare, technological

25:06

change tends to be pitifully

25:08

slow. The absence of competitive

25:11

pressure blunts incentives to

25:13

improve. Governments may

25:15

also have public policy goals, such

25:18

as maximising employment levels,

25:20

which are inconsistent with improved

25:22

efficiency. These industries

25:25

are also more likely to be unionised

25:27

and unions are good at preventing job

25:30

losses. Examples

25:33

abound. Train drivers on

25:35

London's publicly run underground

25:37

network are paid close to

25:39

twice the national median, even

25:42

though the technology to partially or

25:44

wholly replace them has existed

25:46

for decades. Government agencies

25:49

require you to fill in paper forms, providing

25:52

your personal information again and

25:54

again. In San Francisco,

25:56

the global centre of the AI surge,

25:59

Real Life.

25:59

Life cops are still employed to

26:02

direct traffic during rush hour.

26:05

Many of the jobs at risk from AI are

26:08

in heavily regulated sectors. Return

26:11

to the paper by Mr Felton of Princeton

26:14

University. 14 of the

26:16

top 20 occupations most exposed

26:19

to AI are teachers,

26:21

foreign language ones are near the top, geographers

26:24

are in a slightly stronger position. But

26:27

only the bravest government would replace

26:29

teachers with AI, imagine

26:31

the headlines. The same goes

26:33

for cops and crime fighting AI. The

26:37

fact that Italy has already temporarily

26:39

blocked chat GPT over

26:41

privacy concerns with France,

26:43

Germany and Ireland said to be considering

26:46

the option shows how worried governments

26:49

are about the job-destructive effects

26:51

of AI.

26:53

Perhaps in time, governments

26:56

will allow some jobs to be replaced.

26:59

But the delay will make space for the

27:01

economy to do what it always does,

27:03

create new types of jobs, as

27:06

others are eliminated. By

27:08

lowering costs of production, NewTek

27:11

can create more demand for goods

27:14

and services, boosting jobs

27:16

that are hard to automate. A

27:19

paper published in 2020 by David

27:21

Orter of MIT and colleagues offered

27:24

a striking conclusion. About 60%

27:27

of the jobs in America did not

27:30

exist in 1940. The

27:32

job of fingernail technician

27:35

was added to the census in 2000. Solar

27:38

photovoltaic electrician was

27:40

added just five years ago. The

27:43

AI economy is likely to create

27:46

new occupations which today cannot

27:49

even be imagined. Modest

27:52

labour market effects are likely to translate

27:54

into a modest impact on productivity,

27:57

the third factor. electricity

28:00

in factories and households began

28:03

in America towards the end of the 19th century. Yet

28:06

there was no productivity boom until

28:09

the end of the First World War. The

28:11

personal computer was invented in the

28:14

1970s. This time the productivity

28:16

boom followed more quickly, but

28:19

it still felt slow at the time. In 1987,

28:23

Robert Solow, an economist, famously

28:26

declared that the computer age

28:28

was everywhere except

28:29

for the productivity statistics.

28:33

The world is still waiting for a productivity

28:35

surge linked to recent innovations.

28:38

Smartphones have been in widespread use

28:41

for a decade, billions of people

28:43

have access to super-fast internet,

28:46

and many workers now shift between the

28:48

office and home as it suits them. Official

28:51

surveys show that well over a tenth of

28:54

American employees already

28:56

work at firms using AI

28:58

of some kind, while unofficial

29:00

surveys point to even higher

29:02

numbers. Still though, global

29:04

productivity growth remains weak.

29:07

AI could

29:09

eventually make some industries vastly

29:11

more productive. A paper by

29:14

Eric Brynjolfsson of Stanford University

29:16

and colleagues examines customer

29:19

support agents. Access

29:21

to an AI tool raises

29:24

the number of issues resolved each hour by 14%

29:26

on average. Researchers

29:29

themselves could also become more efficient.

29:32

GPTX may give them an

29:35

unlimited number of almost free

29:37

research assistants. Others

29:39

hope AI will eliminate administrative

29:41

inefficiencies in healthcare, reducing

29:44

costs. But

29:46

there are many things beyond the

29:48

reach of AI. Blue-collar

29:51

work such as construction and

29:53

farming, which accounts for about 20% of

29:55

rich world GDP, is one example. An

30:00

LLM is of little use to someone

30:03

picking asparagus. It

30:05

could be of some use to a plumber fixing

30:07

a leaky tap, a widget could

30:10

recognise the tap, diagnose

30:12

the fault and advise on fixes.

30:15

Ultimately though the plumber still has

30:17

to do the physical work. So

30:19

it is hard to imagine that in a few years time

30:22

blue collar work is going to be much more

30:24

productive than it is now. The

30:27

same goes for industries where human

30:29

to human contact is an inherent

30:31

part of the service such as hospitality

30:34

and medical care. AI

30:37

also cannot do anything about

30:39

the biggest thing holding back rich

30:42

world productivity growth, misfiring

30:45

planning systems. When

30:47

the size of cities is constrained

30:50

and housing costs are high, people

30:53

cannot live and work where they are most

30:55

efficient. No matter how

30:57

many brilliant new ideas your society

31:00

may have, they are functionally useless

31:02

if you cannot build them in a timely

31:05

manner. It is up to governments

31:07

to defang nimbies. Technology

31:10

is neither here nor there, the same

31:13

goes for energy where permitting and

31:15

infrastructure are what keeps costs

31:17

uncomfortably high. It

31:20

is even possible that the AI

31:22

economy could become less productive.

31:26

Look at some recent technologies. Smartphones

31:29

allow instant communication but

31:31

they can also be a distraction.

31:34

With email you are connected 24-7 which can

31:36

make it hard to

31:38

focus. A paper in 2016

31:42

by researchers at the University of California

31:45

at Irvine Microsoft Research

31:47

and MIT found that the

31:49

longer daily time spent on email,

31:52

the lower was perceived productivity.

31:55

Some bosses now believe that working from home,

31:58

once seen as a productivity

31:59

booster gives too many people

32:02

the excuse to slack off.

32:05

Generative AI itself could

32:07

act as a drain on productivity.

32:10

What happens, for instance, if AI

32:13

can create entertainment perfectly

32:15

tailored to your every desire? Moreover,

32:18

few people have thought through the implications

32:21

of a system that can generate vast amounts

32:24

of text instantly. GPT-4

32:27

is a godsend for a nimby facing

32:29

a planning application. In

32:32

five minutes he can produce a well-written 1,000

32:35

page objection. Someone

32:37

then has to respond to it. Spam

32:40

emails are going to be harder to detect.

32:43

Fraud cases could soar. Banks

32:46

will need to spend more on preventing attacks

32:48

and compensating people who lose out.

32:52

In an AI-heavy world, lawyers

32:55

will multiply.

32:56

In the 1970s

32:58

you could do a multi-million dollar

33:00

deal on 15 pages because

33:03

re-typing was a pain in the ass, says

33:06

Preston Byrne of Brown-Rudnick, a law

33:08

firm. AI will allow

33:10

us to cover the 1,000 most

33:12

likely edge cases in the first draft

33:15

and then the parties will argue over it for weeks.

33:19

A rule of thumb in America is that there

33:21

is no point suing for damages unless

33:23

you hope for $250,000 or more in

33:27

compensation since you need

33:29

to spend that much getting to court.

33:32

Now the costs of litigation could

33:35

fall to close to zero. Meanwhile

33:38

teachers and editors will need

33:40

to check that everything they read has

33:43

not been composed by an

33:45

AI. OpenAI has

33:47

released a programme that allows you to do this.

33:50

It is thus providing the world a solution

33:53

to a problem that its technology

33:55

has created. AI

33:58

may change the world in a way.

33:59

ways that today are impossible

34:02

to imagine. But this is not

34:04

quite the same thing as turning the

34:06

economy upside down. Fogle

34:09

wrote that his argument was aimed

34:11

not at refuting the view that

34:13

the railroad played a decisive role

34:15

in American development during the

34:18

19th century, but rather

34:20

at demonstrating that the empirical

34:22

base on which this view rests is

34:25

not nearly so substantial as is

34:27

usually presumed. Sometime

34:30

in the mid-21st century, a

34:32

future Nobel Prize winner examining

34:35

generative AI

34:37

may well reach the same conclusion. For

34:42

more on the future and economic consequences

34:44

of artificial intelligence, scroll

34:47

back to our Babbage podcast from the 26th

34:49

of April, which asked, How worrying

34:51

is generative AI?

34:53

Find that episode on your podcast app

34:55

or at economist.com slash AI

34:58

dash pods.

35:04

Finally, how Mexico's gangs are expanding

35:06

into every corner of society.

35:10

On May 3rd, Mexico introduced a

35:12

law applying strict controls on the

35:14

import of chemicals used by

35:16

Mexico's gangs to make synthetic

35:18

drugs. The law is backed by

35:21

harsh criminal penalties. This

35:23

is a striking move by Andres Manuel

35:26

Lopez Obredor, Mexico's populist

35:28

president, who has shied away from tackling

35:30

the country's gangs, preferring to blame

35:32

drugs and disorder on family breakdown

35:35

over the border and poverty at

35:37

home. In truth, under his tenure,

35:40

gangs are increasingly powerful

35:42

and diversified. Mexico's

35:45

cartels have always been adaptable. In

35:48

the 1980s, they trafficked marijuana

35:50

and then cocaine from Colombia to the United

35:52

States. But in the past decade,

35:54

they have mutated into a much wider

35:57

array of groups with their tentacles

35:59

reaching beyond the border. the drugs trade into

36:01

extortion, people smuggling, arm

36:04

selling and illegal mining. We

36:07

look at organised crime trends through a drug

36:09

cartel lens, when today in Mexico

36:12

we have a mafia-like criminal landscape,

36:14

says Romain Lacour-Gromison, of

36:17

the Global Initiative Against Transnational

36:19

Organised Crime, an NGO based

36:21

in Switzerland. However, the

36:23

government response continues to be shaped

36:25

by a focus on narcotics,

36:28

not least because of the United States and

36:30

its war on drugs. The pressure

36:33

has only increased with the recent crisis surrounding

36:35

fentanyl. The head of the US Drug Enforcement

36:38

Administration, or DEA, last month

36:40

described the Sinaloa Cartel and

36:42

the Jalisco New Generation Cartel,

36:45

or CJNG, Mexico's

36:47

two biggest groups, as the

36:49

greatest drug threat our nation has ever

36:51

faced. Drugs continue

36:54

to make up a large part of the business,

36:56

especially for the two main gangs. Production

36:59

of cocaine cultivated in South America

37:02

and trafficked through Mexico has more

37:04

than doubled since 2014, says

37:06

the UN's Office, on drugs and crime,

37:09

while the price has not dropped. The

37:12

pivot to synthetic, such as methamphetamine,

37:14

which is sold locally, and fentanyl,

37:17

which is sent north, has only made business

37:19

more lucrative. The indictments

37:21

in the United States in April of the Chapitos,

37:24

the four sons of Joaquin Guzman,

37:27

known as El Chapo, the head of the

37:29

Sinaloa Cartel, who is now imprisoned in

37:31

Colorado, have shed light on the changing

37:33

shape of the trade. The DEA

37:36

estimates that a pill that costs 10 cents

37:38

to make in Mexico can be sold to

37:41

a wholesaler for 50 cents. Unlike

37:43

marijuana or cocaine, no land

37:46

is needed for synthetics. The chemicals

37:48

are imported through Mexican airports

37:50

or ports from China. Gangs

37:53

drop them off to labs, sometimes family

37:55

kitchens in northern states, before

37:57

pressing them into pills to take north. Perhaps 40%

38:01

of the Sinaloa Cartel's income

38:03

comes from drugs, estimates one official, half

38:06

of which come from synthetics. Gangs

38:10

have diversified for several reasons.

38:12

First, el narco, as Mexicans

38:15

call them, is simply a monster

38:17

that eats everything it can, says João Agrillo,

38:19

an author. But official policy

38:22

has exacerbated the problem. The

38:24

war on drugs begun in 2006 by

38:26

the then-president Felipe Calderón caused

38:29

groups to splinter and multiply.

38:32

By 2020, the number of gangs had

38:34

increased to more than 200, from 76 in 2010, according

38:38

to the International Crisis Group, or ICG,

38:41

a think tank based in Brussels. Not

38:44

all have the ability or connections to traffic

38:46

or make drugs. The DEA recognises

38:49

only nine major drug trafficking organisations

38:52

in Mexico. Second,

38:54

the security policies of Mr López Obrador,

38:57

who is often known as AMLO, have

38:59

given gangs room to expand. His

39:01

approach to security is known as abrasos

39:04

no balazos, that's hugs, not

39:07

bullets, a policy that seeks to tackle

39:09

root causes, but does little to confront

39:12

existing gangs. This

39:14

has let the groups deepen their expansion,

39:16

which started a decade ago, into

39:18

the legal economy, says Van de Felber-Brown,

39:21

of the Brookings Institution, an American

39:23

think tank. The biggest source of income

39:26

after drugs is, without a doubt, extortion,

39:28

says Eduardo Guerrero, a Volantia

39:31

Intelligencer Consultancy. Groups

39:34

extort money from everyone, whether

39:36

taco stand owners or pilgrims,

39:38

some of whom, over Easter, were forced

39:40

to take part in processions.

39:43

In 2022, almost double the

39:45

number of Mexicans reported having money

39:48

extorted than did five years before.

39:51

Only a tiny minority report.

39:54

The big money comes, however, from taxing

39:57

businesses in sectors such as agriculture

40:00

mining. Avocados, Mexico's

40:02

green gold, are a good example. The

40:05

country provides almost a third of global

40:07

supply, most of which is grown

40:10

in the western state of Michoacan.

40:13

The $3 billion worth of them exported

40:15

every year to the United States is a huge

40:18

source of income for the producers and

40:20

also for gangs. For the past

40:22

three years, Eric Rodriguez, a farmer,

40:25

has paid an annual fee of £10,000. That's $560

40:27

per hectare to Familia Michoacan, a

40:34

local criminal group. Mr

40:36

Rodriguez, not his real name, says

40:38

the gang comes with data about the size

40:40

of his farm and tells him to hold

40:43

back stock to push up prices. Ms

40:46

Felbab Brown's fieldwork in Mexico

40:48

shows how gangs also force fishermen

40:51

to sell their catch at a cut price, which

40:53

they then sell for a profit to restaurants.

40:56

They also dictate the terms of when and

40:58

what they can fish. Smuggling

41:01

people is another area of expansion.

41:04

Criminal groups used to leave this to small-time

41:06

smugglers known as coyotes. As

41:09

the border has tightened and the number trying

41:11

to cross has risen, so has the

41:13

cost, making it even more attractive

41:15

to the gangs. The US border

41:18

police apprehended migrants trying to cross

41:20

illegally 2.2 million times

41:22

last year, a six-fold increase on

41:24

a decade earlier. The cost

41:26

to transit Mexico and cross into

41:29

the United States can now be as high as $12,500

41:31

per person, according to a

41:35

study supported by BBVA, a

41:37

Spanish bank. Gangs

41:39

also kidnap migrants to extort

41:41

money from their richer relatives north of

41:43

the border. Mexico's immigration

41:46

agency freed over 2,000 kidnapped

41:48

migrants last year. Many more

41:51

are never found. Meanwhile,

41:53

a growing percentage of the 25 people

41:55

who disappear every day in Mexico are 12

41:58

to 15-year-old girls. many

42:00

of whom are likely to be trafficked into

42:02

prostitution. And though guns are

42:04

tightly regulated in Mexico, gangs

42:07

smuggle them in from the United States.

42:10

Trafficking natural resources is lucrative

42:13

too, says the Global Initiative Against

42:15

Transnational Organised Crime. In 2014,

42:18

Mexico's government acknowledged for the first

42:20

time that gangs illegally mine

42:23

and export iron ore from Michoacan.

42:26

A report by Insight Crime, an investigative

42:29

outfit, last year found the activity

42:31

is thriving now under the CJNG. In

42:34

the state of Jalisco, gangs control

42:37

the timber market. Demand for Mexican

42:39

animals, though a niche business, is

42:41

growing too. Most are believed

42:44

to be sent to China, sometimes in

42:46

exchange for chemicals for drugs. Oil

42:49

is also a target industry. Gangs

42:52

siphon it off, sometimes cutting deals

42:54

with corrupt officials at Pemex, the state

42:56

oil company. They steal

42:58

and sell water in the states of Mexico

43:01

and Nuevo Leon. One

43:03

result of this expansion is a rise

43:06

in violence. Drug trafficking

43:08

requires little more than someone corruptible

43:10

at a border. But today's activities

43:12

are only possible where groups control

43:15

territory, causing violent clashes

43:17

across the country. They are particularly

43:20

fierce in Michoacan, where at

43:22

least 35 groups compete for resources,

43:25

now using military-grade weapons

43:27

bought with income from the drugs trade.

43:30

Homicide figures are alarming but

43:33

a poor measurement of the impact. Murders

43:35

fell by 7% from 2021 to 2022, to

43:39

a still shocking total of almost 31,000. But

43:43

disappearances, most of which are murders

43:46

with no body, have risen. And

43:48

like in the past, when gangs were only involved

43:50

in the illegal economy, today every

43:53

citizen and business is exposed

43:55

to being controlled by them, says Ms Falbab

43:57

Brown.

43:59

has only made things worse. After

44:02

the United States arrested Salvador

44:04

Cienfuegos, a former head of Mexico's

44:06

armed forces, the President in 2021 enacted

44:10

a national security law to curtail

44:12

the DEA's ability to operate

44:14

in Mexico. Relations

44:17

have subsequently improved and the government

44:19

has been carrying out more strikes against

44:21

gang operatives, as well as

44:24

cooperating on financial investigations.

44:27

Yet taking out kingpins, a policy

44:30

popular under Mr Calderon, has had

44:32

little effect. The President has

44:34

spread out the National Guard, a

44:36

federal police force he created in 2019,

44:39

to reassert the state's presence. But

44:42

in many places he has ordered them to stay

44:44

in the barracks. His tenure has

44:46

been marked by a general passivity towards

44:49

security and corruption, notes Mr

44:51

Guerrero. This has led

44:53

to a second consequence, deeper

44:56

corruption within the state which allows

44:58

gangs to diversify further. Officials

45:01

have always struck deals with organised crime,

45:04

but under the Institutional Revolutionary

45:06

Party, or PRI, which ruled Mexico

45:08

for 71 years until 2000, the

45:11

government called the shots. Democratisation

45:14

and the decentralisation of power have

45:17

tilted the balance of power in favour

45:19

of the criminals. The

45:22

recent conviction in the United States of

45:24

Jenero Garcia Luna, Mexico's

45:26

security minister under Mr Calderon, for

45:29

taking bribes from the Sinaloa Cartel

45:31

to facilitate drug trafficking was

45:33

a rare, proven case involving a federal

45:36

official. Criminal groups

45:38

are deeply entwined with local police,

45:41

mayors and politicians who can be threatened

45:43

or bought. This gives them

45:45

access to more sources of revenue through

45:48

public contracts to build roads and other

45:50

infrastructure. More worryingly,

45:53

rather than just threatening elected officials,

45:56

gangs are gaining unprecedented influence

45:58

over elections. Before

46:00

mid-term polls in 2021, nearly 40 candidates were

46:03

killed. Gangs

46:06

are having a greater impact on businesses

46:09

and society as well. Some

46:11

firms see no option but to employ

46:13

the groups to protect them, notes Teresa

46:15

Martinez of Tec de Monterrey,

46:18

a university. Extortion

46:20

by gangs is one reason for Mexico's

46:23

sluggish economic growth. Small

46:25

business owners say they struggle to pay up

46:27

or do not want to grow so big as to attract

46:30

attention. For the most part,

46:32

the gangs show little interest in building

46:34

ties with the communities they prey on.

46:37

But some do engage. On Children's

46:40

Day, a festival at the start of May,

46:42

the CJNG even handed out

46:44

toys. We are seeing the

46:46

transformation not only from drug trafficking

46:49

organisations to multi-commodity

46:51

ones, but now to actual local

46:53

strongmen exercising political

46:55

control, says Falco Ernst of

46:58

the ICG. Mexico's

47:01

gangs have sprawled to such an extent

47:04

that Mr Grillo and others have started

47:06

calling the situation an insurgency.

47:10

Certainly their mutation makes everything

47:12

harder to tackle. It requires a

47:14

broad national response, including

47:16

weeding out corruption, creating a functioning

47:19

justice system and re-establishing

47:21

the remit of the state, as well

47:23

as tackling drug consumption in the United

47:26

States.

47:26

But with the damage being wrought by

47:29

fentanyl, a shift from the narrow focus

47:31

on drugs north and south of the border

47:34

looks unlikely.

47:43

Thank you for listening to Editor's Picks. For

47:45

more from The Economist, subscribe at

47:47

economist.com slash podcast

47:50

offer. I'm Miranda Mitra, and

47:52

in London, this is The Economist.

48:00

you