0:00

This message comes from NPR sponsor,

0:02

Bank of America, with powerful digital

0:04

solutions for local and global businesses.

0:07

So you can make every move matter.

0:09

Visit bankofamerica.com slash

0:12

banking for business Bank of America,

0:14

N A copyright 2023.

0:17

This is Planet Money from

0:20

NPR. It's

0:24

been about six months since chat GPT

0:26

was released to the public. And basically

0:28

from the moment that happened, it felt

0:30

like this seismic shift. Because

0:33

all of a sudden people everywhere

0:35

realized just how powerful artificial

0:38

intelligence

0:39

already is. They began using this

0:41

AI chat bot to do all sorts of things,

0:43

to write raps, to take the bar exam, to

0:46

identify bugs in computer code and fix

0:48

them. I mean, all that stuff is pretty

0:50

cool, but at the same time, there's

0:53

been all this doom and gloom about

0:55

AI. Will it take our jobs?

0:57

Will it derail democracy? Will

0:59

it kill us all? And these aren't like

1:01

off the wall questions. Like serious people

1:04

are asking these questions right now. Yeah. It's

1:06

kind of easy to fall into this like doom

1:08

spiral these days. But then

1:10

a couple of weeks ago, I saw something

1:12

that gave me like this little glimmer of

1:14

hope. It was a study that looked at

1:17

this customer service department of a

1:19

big software company. And they started using

1:21

chat GPT to help workers get better

1:23

at their jobs. And interestingly enough,

1:25

it worked. Like it made the less

1:28

skilled workers at this company much

1:30

more productive. And at the same time,

1:33

it didn't do much for workers at the top.

1:35

So basically AI narrowed the productivity

1:38

gap between lower skilled workers and workers

1:40

with more skills. And Greg, I

1:42

think it's fair to say, you read a lot of economic

1:45

studies. Probably too many. And yet

1:48

you have been telling me, you've been telling

1:50

all of us that this finding felt really

1:52

big to you because it's different from

1:54

how we usually understand the way technology

1:57

affects workers. Yeah. There's a whole

1:59

generation.

1:59

of research looking at the effects computers

2:02

have had on the labor market. And over

2:04

and over again, what economists find

2:06

is that for decades now, computers have been

2:08

this major force for increasing

2:11

inequality. What this study shows

2:13

is that AI could be different.

2:16

And when I saw that, I was like, you know what? I

2:18

want to talk to David Otter. David

2:21

Otter, professor at MIT, widely

2:23

regarded as one of the greatest labor economists

2:25

in the world. Otter led a lot of that research

2:27

that found computers were this force

2:30

for a shrinking middle class. And I wanted

2:32

to find out if he thinks maybe this

2:34

new technological era we're in is

2:37

going to be different.

2:38

If maybe AI could be

2:40

a force for greater equality. Right.

2:42

So hello and welcome

2:45

to Planet Money. I'm Nick Fountain. And I'm Greg Griselsky.

2:48

And Greg, today's show is going to be a little different. We

2:50

found your conversation with David Otter so

2:52

interesting, so illuminating, so prescient.

2:56

That we're just going to run it. Today in the show,

2:58

the American middle class has been shrinking for

3:00

more than 40 years. Could AI

3:03

help reverse that trend?

3:12

When David Otter thinks about how

3:14

AI will affect the future of work, he

3:16

actually looks to the past. He sees

3:19

two major turning points when technology

3:22

fundamentally changed our economy. The

3:24

first turning point was a long time ago. We're

3:26

talking about the industrial revolution,

3:29

when machines began to replace work

3:31

that had previously been done by hand.

3:34

So prior to the industrial

3:37

revolution, there was a lot of

3:39

artisans, people who did

3:42

all the steps in making a product. Right. So

3:44

whether it's a piece of clothing or building

3:46

a house or a tool,

3:48

the era of mass production created

3:51

an alternative way of making things. And

3:54

it was basically breaking things down into a

3:56

series of small steps that would

3:58

be accomplished in the future. sequence,

4:01

often by machines, managers,

4:03

and pretty low skill workers. And so

4:06

a lot of artisanal skill was displaced. I mean,

4:08

the Luddites rose up for a reason.

4:10

Carter says that at first the factory

4:12

jobs that displaced the artisans

4:15

required less skill and also paid

4:17

less. So kind of a bummer. But

4:19

then machines got more complex

4:22

and so did the things they could make.

4:25

We're talking automobiles instead

4:27

of textiles. And so factory

4:29

owners started to need workers with more

4:31

skills. Over time, that

4:34

work became more skill demanding

4:36

because people had to follow formal rules. And

4:38

if you're using a lot of expensive equipment and making

4:40

precise products and using expensive

4:42

inputs, you need people who are kind of can

4:45

follow those rules well.

4:46

So this created what you might think of as

4:49

the kind of middle skill, what I would call mass

4:51

expertise, right? This is like the

4:53

golden era that we hear a

4:55

lot about in the United States this

4:58

time when people could graduate

5:00

from high school with basic reading and

5:02

math skills and then go out and find

5:04

gainful employment, you know, jobs on

5:07

factory floors or jobs in offices

5:10

where workers had to understand how to

5:12

compile paper records or deal

5:14

with basic financial transactions.

5:16

For people who didn't have four year college degrees, these

5:19

were the relatively better paid

5:22

jobs, right? They're better

5:24

paid than, for example, food service, cleaning, security,

5:26

and so on. And the reason is that food

5:29

service, clean, security, they're valuable

5:30

pursuits. They do use important things in

5:32

the world, but most people can do them.

5:34

And so they're not going to be well remunerated.

5:37

For work to be well paid, especially

5:40

in an industrial economy, it needs to be expert

5:42

work of some sort. By expert, I mean one,

5:44

you need a certain body of knowledge or competency

5:47

to accomplish that a thing.

5:49

That thing must be worth accomplishing, right? And

5:52

not everyone can do it. And so it is

5:54

the case that the kind of industrial era helped

5:56

really grow the middle class. It created

5:58

this tailwind.

5:59

where people with a reasonable amount of education all

6:02

of a sudden made them highly productive in offices, highly productive

6:04

in factories, highly productive in sales. And

6:06

so, yeah, it created this huge rising

6:09

tide that was relatively

6:11

equalizing. Now, I don't want to say it's only technology,

6:13

right? There are institutions that went with this. There's

6:15

democracy. There was obviously the system that educated

6:17

people, but the technology helped. So

6:20

okay, the Industrial Revolution, it

6:22

killed off jobs for skilled

6:25

artisans. But then it created a whole

6:27

bunch of new jobs for middle-skill

6:29

workers,

6:29

jobs that gave opportunities

6:32

to Americans without a college degree. That's

6:34

turning point number one.

6:36

The second big turning point is

6:39

computers.

6:40

This is what a lot of David Otter's

6:42

research has focused on. He finds

6:45

that in the computer era, starting

6:47

around 1980 or so, all

6:49

of those middle-skilled jobs that emerged

6:51

from the Industrial Revolution, they

6:54

started getting automated away.

6:56

Think

6:57

robots taking jobs on assembly lines,

6:59

or computer software taking jobs from

7:02

administrative office workers.

7:04

At the same time, computers made higher-skill

7:07

workers much better at their jobs. This

7:09

elite group benefited a bunch

7:12

from using email, building

7:14

spreadsheets, searching the internet. They're

7:17

like trading stocks and information instantaneously

7:20

all over the world. So if

7:22

you're a highly educated worker, if you're

7:25

a doctor or an attorney

7:27

or a marketer or researcher, those people

7:29

are highly strongly complemented by

7:31

this

7:33

automation of these information processing and

7:35

routine tasks. On

7:37

the other hand, if you are someone

7:39

who does dexterous manual work

7:41

like food service cleaning, security,

7:44

entertainment recreation, there's really not much complementarity

7:47

there at all. It doesn't make you much better,

7:49

doesn't make you worse. However, you have

7:51

lots of people in the middle who

7:53

are now being pushed out of those

7:55

middle-skilled occupations, and it's just not very

7:57

easy to move up.

7:59

manufacturing job, it's very unlikely you're going to get a law

8:02

degree or medical degree. So you're going to more

8:04

likely end up driving a truck, working in

8:06

a restaurant, working as a security

8:08

guard. And so the computer era

8:10

actually devalued that mass expertise and

8:13

massively amplified demand for elite expertise,

8:16

which has been really not so great. Right. It's

8:19

not great if you're not a, that's right. Elite worker.

8:21

Because it's pretty great if you're an elite worker. It's

8:24

true. It's been a lot. It's been a

8:26

great four decades for elite workers, especially in the

8:28

United States. But to put it in crude words,

8:29

technological change in over the last few decades

8:32

has increased inequality. Sure. And

8:35

now it feels like maybe,

8:37

like just maybe we're in a new era.

8:39

Like I was already starting to think this and then this new

8:41

empirical study came out by Eric Bernielsen,

8:44

Danielle Lee and Lindsay Raymond. And that looked

8:46

at what happened to a software company and its workers

8:48

after the company adopted an old version

8:50

of chat GPT. And they basically find that

8:52

this AI system makes their workforce

8:54

much more productive. More interesting

8:57

to this conversation, they found that only

8:59

some workers benefited from it and it was actually

9:01

the less experienced, lower skilled

9:03

workers that benefited from use of the technology.

9:06

And the more experienced, higher skilled workers

9:08

saw little or no benefit. And

9:10

to me, that kind of, it seems to be like reversing

9:13

what we've been seeing where it's complimenting

9:15

the bottom and not really doing much

9:17

for the top. And I just want to

9:19

get your reaction to those findings. Sure.

9:22

And actually, my students, Shaqed

9:24

Noy and Whitney Zhang also have a paper where

9:26

they did a sort of a related experiment working

9:29

with people doing writing tasks. And

9:31

these are people who are college educated and do like advertising

9:34

copy and so on. And some use chat

9:36

GPT and some didn't. And basically they found

9:38

that using the large language model, it

9:40

made everyone more productive

9:43

by saving them a lot of time, but

9:45

it pulled up the bottom very considerably.

9:48

So the people who are only pretty poor writers on

9:50

this scale became

9:52

average and people who were excellent

9:54

became a little better. And so it reduced

9:56

productivity inequality. So

9:58

it's very consistent.

9:59

with the paper by Rignalsson and

10:02

Lee and Raymond. So that's interesting. I didn't

10:04

know about that study. So now we have two

10:06

empirical studies that are showing that

10:08

it's pulling the bottom up and

10:11

maybe doing a little for the top, but maybe

10:13

not doing much. Right. So

10:16

there's a big implication there, yes?

10:19

There's a big possibility there.

10:22

So the good scenario

10:24

is one where AI makes

10:26

elite expertise cheaper and

10:28

more accessible.

10:29

So right now, if you want

10:32

to do a lot of medical procedures, you

10:34

need a medical degree. That takes a decade,

10:36

right? And that makes those people scarce, expert

10:39

and expensive.

10:40

But you can imagine that with

10:42

the right tools, you could devolve some

10:44

of those tasks to people who have know something

10:47

about medicine and healthcare, but they don't have to have that

10:49

level of education. And then they could do much more.

10:51

And we already have an example of that. So the nurse practitioner,

10:54

nurse

10:54

practitioners are just a nurse who has an additional master's

10:56

degree. My sister's a nurse practitioner.

10:59

Okay, great. And so nurse practitioners are well paid, right?

11:01

The median pay is about $150,000 a year. And

11:04

they do many of the things that only

11:07

medical doctors were allowed to do, right? They diagnose,

11:10

they prescribe, they treat, right?

11:12

And how is that

11:14

compatible? Partly it's a change in medical

11:16

norms and scope of practice boundaries. Partly

11:18

they're enabled by technology, right? There's a machine that

11:20

says, don't put those two prescriptions together. That

11:23

would be a problem. And this set of symptoms

11:25

is associated with this constellation of

11:28

diseases, check the following. And you can

11:30

imagine many ways in which people with

11:32

foundational skills in something could

11:35

use AI to make

11:37

that expertise go further. So

11:40

the good scenario is basically where AI

11:42

lowers the cost of

11:44

elite expertise, makes it more available

11:47

and increases the value of

11:50

basically the middle skilled workers of the future. That's

11:52

my good scenario. So to translate potentially

11:56

AI,

11:56

good. good

12:00

for the middle class. Good for rebuilding

12:02

the middle class. That's like the- Could be. That's

12:05

like the hope. That's the good scenario. That's like the headline. And

12:07

not just hope, we gotta make it happen. That's the headline

12:09

right there. Like David Autor hopes

12:12

that AI is good for the middle class.

12:16

No, no, no. Let's use AI

12:19

to reinstate the middle class.

12:21

What it will take to make that happen.

12:24

And also the other scenario David

12:26

Autor imagines, the one that doesn't

12:28

go so well for workers.

12:30

That's after the break.

12:37

Waylon Wong here with a plug for our latest

12:39

bonus episode, where we take you inside

12:42

a reporting on social media

12:44

influencers.

12:45

There's kind of a magic number

12:47

where it becomes, I can do this for

12:49

a living, and that's less than 1%.

12:51

And yet some Gen Zers say it's their

12:54

dream job. I mean, that way I can

12:56

earn money as I'm just like at

12:58

home, just playing video games. Follow us

13:00

down that rabbit hole and hear how our indicator

13:02

series on the dark and mysterious economy

13:05

of influencers came together. That's

13:07

in our latest bonus episode, available now

13:09

for our Planet Money Plus supporters. If

13:12

that's not you, it could be. Learn more

13:14

at plus.npr.org.

13:17

So for years, David Autor has

13:19

been looking at the effect of technology

13:22

on the labor market. And he finds that computers

13:24

made elite workers better at their jobs

13:27

and much richer. But at the same

13:29

time, computers also made

13:31

a bunch of good middle-class jobs disappear.

13:35

Autor thinks maybe AI could help reverse

13:37

that trend, lift a whole bunch of workers

13:40

back into the middle class by helping

13:42

them get better at writing or

13:44

research or I don't know, creating complicated

13:47

legal documents. Basically, AI

13:49

could allow them to do jobs currently

13:52

reserved for the upper echelon of

13:54

the labor market. That's what David

13:56

Autor calls the good scenario.

13:59

But even in this good scenario,

13:59

scenario, no, there's going to be

14:02

a disruption of people who are

14:04

currently making, I don't know, 100 to $200,000 a year or something

14:06

like that. All

14:08

of a sudden, it

14:10

doesn't make as much sense to pay those people as much

14:12

anymore because you have a whole pipeline

14:14

of people who now do that job. That's

14:16

correct. It's possible that basically you will

14:18

see some expensive

14:20

expert work just less in demand that you will

14:22

need fewer managers for certain types of

14:25

decision making that, you know,

14:27

more like legal work will be done

14:29

by machines as opposed to by lawyers

14:32

and that they'll have lawyers, but they're supervisory

14:34

and there are fewer of them. So yeah, I

14:36

think it's possible. But you

14:38

know, in the long run, that means fewer people have to go to college,

14:40

which expensive. And

14:43

it also matters, right? This is not a zero sum

14:45

game, right? If it makes us all more productive,

14:48

we're wealthier as a result of that, right? So

14:51

even if it just places on what you do, but then the rest

14:53

of what you do, you do it 10 times as fast,

14:56

that's a gain in productivity. So

14:58

you're hopeful on the labor market thing

15:01

contingent on smart government

15:03

policy, essentially. Smart government, smart

15:05

private sector, smart philanthropy, smart

15:07

universities. And so maybe that will

15:09

evolve some disruption of people at the top. But

15:11

you know what, they've been doing so well for so long

15:14

that maybe, you know, you got to crack some eggs

15:16

to make an omelet.

15:17

That's right. And I don't think they're

15:19

just going to be thrown out of the top. You're not going to say,

15:21

that's right. I love that though. I

15:24

don't think they're all going to be just like thrown out of the top floor

15:26

of office buildings. You know, these things

15:28

happen gradually. Yeah. So

15:30

that's the optimistic scenario. I just want to bounce

15:33

off the dystopian kind

15:35

of, and maybe this is, I mean, the truly dystopian

15:37

is they become sentient and kill us all.

15:40

But the dystopian economic.

15:43

Or we actually, the more likely dystopian is

15:45

some, we use it to kill one another.

15:47

Yeah. Putting

15:50

that aside, putting

15:52

human existence aside, I'm focusing

15:54

on the economics. Other

15:58

than that, how was the play, Mrs. Lincoln?

15:59

Um, I can

16:02

still imagine like a narrative

16:05

or a potential future where it's actually AI

16:07

is inequality increasing. So one,

16:10

one scenario, obviously like companies

16:13

who own these systems will get insanely rich,

16:15

but then there's also like the downstream effects

16:17

where there's a whole bunch of industries where

16:19

a

16:20

bunch of people used to do the job,

16:22

but now only you need one or two people

16:25

to do it. So what do you think about

16:27

that sort of pessimistic

16:29

potential? I don't want to rule it out. I mean, you can,

16:31

so you can imagine a world where, you know, you just

16:33

need a few super experts overseeing everything

16:35

and every, everything else is done by machines. Right? So

16:38

that's one possibility.

16:40

Another possibility is one where like no

16:42

one's labor scarce, right? That's not a good

16:44

world because then

16:46

we have lots of productivity, but nobody who

16:49

owns it just the owners of capital, right? Then we have to have

16:51

a revolution and blah, blah, blah. It's not going to work out well,

16:53

right? Those things never work out well. So I

16:55

don't view those scenarios as highly likely. One

16:58

thing to recognize is that we are actually in a period

17:00

of sustained labor scarcity because

17:02

of demographics, right? We have

17:05

very low fertility rates. We have

17:07

large populations who are retiring

17:11

and we have radically restricted immigration.

17:14

And so the US population is growing at

17:16

its lowest rate since the founding of the nation.

17:19

And most industrialized countries and China

17:21

as well, by the way, are facing

17:24

this problem of they're getting smaller and older

17:26

or their populations are not growing. That's

17:29

a world where we need a lot more automation actually

17:31

to enable us to do things

17:33

we need to do, including care for the elderly.

17:36

So I'm not worried about us running

17:38

out of work and running out of

17:40

jobs. I am worried about the devaluation

17:42

of expertise. Just for

17:45

clarity though, because you just said I am concerned

17:47

about the devaluation of expertise,

17:49

but also though it sounded like you were excited

17:51

about the devaluation of expertise. I'm

17:55

worried about a world where no

17:57

one's labor is scarce. But let me give you an

17:59

example.

17:59

I mean by this. For example, you might

18:02

say, oh, Waze makes everyone

18:04

an expert driver. But

18:06

no, actually, it doesn't. It doesn't make anyone an expert

18:09

driver. It has the expertise. So there

18:11

was a time when London taxi cab drivers needed

18:13

to know all the

18:15

highways and byways of London, which took

18:17

years to master. It was an incredible feat of memorization.

18:19

And then that made them really expert. They could get

18:21

you around London better than any other

18:24

driver. Well,

18:24

now you don't need to know that. You just need a phone.

18:28

And that's good for passengers. It's good

18:30

for consumers,

18:31

but it devalues the expertise that those drivers

18:34

have. Would you call that de-skilling, essentially

18:36

de-skilling? I would say it devalues

18:38

the expertise. So I realized that's not, didn't

18:40

meet you halfway there. But what I mean is. It

18:43

sounds like de-skilling to me because they used

18:45

to have a skill that was, I guess it was a

18:47

valuable skill and now they still have that

18:50

skill. They still have the skill. It's just not needed.

18:52

It's not scarce, right? So the expertise

18:55

of being a London cabby has

18:57

been substantially devalued.

18:59

Okay. So yeah, that's one of Otter's big

19:01

worries that what happened to London

19:03

cabbies kind of happens to the entire

19:06

labor force that AI makes human

19:08

expertise kind of irrelevant.

19:11

It devalues it, but David Otter

19:13

doesn't actually think that will happen, at

19:16

least not for all workers and not

19:18

anytime soon. He says people

19:20

still have all of these advantages of our AI.

19:23

Like we're more adaptable. We have

19:25

more common sense. We're better at

19:27

relating to other people. Not

19:29

to mention we have bodies. We have arms

19:32

and legs and we move around in the world. Like

19:34

there's a bunch of things about being a human that

19:36

still have advantages in the marketplace.

19:39

So AI

19:40

raises all of these different possibilities.

19:43

Some more promising, some kind of scary,

19:45

some very scary. And I just

19:47

want to end by getting sort of the

19:49

big picture gut check from David Otter.

19:52

I'm curious where your head is at. Like,

19:54

cause where are you?

19:56

Because you seem hopeful that it could rebuild

19:58

the middle class if we...

19:59

we channel it, but then it's also like there's

20:02

all these other who knows where this is

20:04

going. Yeah, I mean, I think there's

20:06

an optimistic, there's a positive scenario

20:08

for labor market,

20:09

but that's the labor market side. I

20:12

think there's many reasons for concern

20:14

about how AI could

20:16

be misused in all kinds of other ways,

20:19

right? From misinformation to

20:21

control of critical systems, to

20:24

surveillance and monitoring and coercion,

20:28

to very, very dangerous

20:30

weapons, smart weapons that can autonomously

20:33

do all kinds of terrible things. So

20:35

I think there's lots of reasons

20:37

to be scared about how it can be used.

20:40

The irony is the labor market is the least scary part

20:42

of this at the moment in my mind.

20:47

Well, thank you very much. I really appreciate it.

20:49

Sure thing, Greg. It was a pleasure speaking with you. Thanks. It

20:51

was really a lot of fun. If

20:54

you enjoyed this episode, we've got more

20:57

AI content on the way. Next

20:59

week, we will be launching a three part

21:01

series where we try to figure out whether

21:03

we can replace all planet money

21:06

with AI. Yeesh. If

21:09

you just want more insightful planet

21:11

money content about the economy from a

21:13

real life human being, subscribe

21:15

to the newsletter that I write. You can find

21:17

it at npr.org slash planet

21:19

money newsletter. This episode was

21:22

produced by Dave Blanchard and edited

21:24

by Molly Messick. It was fact checked

21:26

by Sierra Juarez and engineered

21:28

by Catherine Silva. Jess Jang is

21:30

Planet Money's acting executive

21:32

producer. I'm Greg Ryszalski. This

21:34

is NPR. Thanks for listening.

21:39

Support for NPR and the following message

21:41

come from Front Door. Home to-do lists

21:44

can seem endless. Repair the leaky

21:46

dishwasher. Fix the fridge. Get the

21:48

faucet to stop dripping. If only

21:50

there was a way