The Future of Work Podcast

-Hello, and welcome back to the ILO's Future of Work Podcast.

I'm Sophy Fisher.

There's a great deal of talk

and some concern about the effect that generative AI will have on jobs.

While it's clear that some new tasks and roles are being created,

there are many predictions that jobs and perhaps even entire skill sets

will become redundant or will even be destroyed.

I have with me today two people

who are extremely well qualified to throw light on this question.

Janine Berg and Pawel Gmyrek are two of the authors of a new ILO report,

Generative AI and Jobs:

A Global Analysis of Potential Effects on Job Quantity and Job Quality.

They've been looking at the effects, as the report says,

on both quantity and quality of employment.

Welcome to you both and thank you for joining us in the studio.

-Thank you for having us.

-Let me start by asking you a very basic question.

Let's define what we're talking about when we're talking about Generative AI.

Pawel.

-When we're discussing Generative AI a big simplification,

it's basically the latest family of the AI on the market,

which is AI that is pre-trained on large quantities of data in forms of text,

image, or video increasingly,

and which is able to reproduce this type of materials.

It's able to produce text or image,

also video.

These materials it produces often cannot be really distinguished anymore

from those produced by humans, on the surface.

Basically, the launch of ChatGPT

at the end of last year brought it to the public

and many more people have been able to interact with this type of AI

which has been around earlier,

but ChatGPT let us all experiment what it's like to use it,

at least in working with text on a daily basis.

-That's why there has been such a lot of discussion

about it in the last six months.

Let's move on to your report.

For me, the interesting thing about this is not only at job quantity,

but at job quality, too.

Janine, why don't you run us through the key findings?

-Basically, we did an analysis to see

what could be the potential impacts of Generative AI on the world of work.

We did the analysis of occupations, so broad occupational categories,

and within those occupations, tasks.

Every day you go to work

and you do lots of different functions in your day-to-day work, and these are tasks.

We looked at the tasks structure

and the potential effects of the technology on automating tasks

or how it could affect it.

The reason why we have this distinction between automation and augmentation is

because it really depends on

what tasks are in your job.

If you have a lot of tasks

that are highly exposed to this technology,

then there's a good chance, or there's a possibility

that your job could then be automated.

If you have some of your tasks that are exposed to the technology,

then they may be automated, but you still have a lot of other things

that you need to do every day

and so as a result,

we say your job could be augmented because those functions

might become automated through the technology,

but you're still needed to do other tasks.

-The ones that you're describing as susceptible to automation,

are you assuming that those jobs will be entirely redundant?

-We can't say,

and it's certainly not going to be from one day to the next.

I think that that's something that's really important

to take into consideration.

It's not like a doomsday thing where the technology is out

and the jobs are gone.

Usually, these processes take a long time to transition

and of course, a lot of companies might not adopt the technology

or they don't adopt it in full.

We also do a global analysis here.

There are lower-income countries

that might not be able to incorporate the technology.

There's a lot of caveats.

Our numbers are about the potential exposure,

which is really important to consider.

What we find is yes, there are some categories of jobs

that have the potential to be completely automated,

but the far majority of jobs are augmented.

-Which categories of jobs did you find were most likely to be affected?

-Overwhelmingly, it's clerical support workers.

There's 10 broad occupational categories.

People such as professionals, managers, service and sales workers, technicians,

associate professionals,

all of these people have about one quarter of the tasks

that they do that are exposed,

but that means that they will be augmented.

Whereas, with clerical support workers,

we find that they have a high-level exposure

in the sense of 25% of their tasks are highly exposed,

almost 60% of their tasks are medium exposed,

so there's a very high probability then

that there's this potential for automation of their occupations.

-When you're talking about clerical workers,

are you talking about people doing admin-type work?

Does that also trickle up to people doing accountancy and things like that,

people with a slightly higher skill set?

-It's mainly admin work.

It includes,

let's say, customer service, call centre work.

It could be hotel receptions,

and I think hotel receptions is a good example of the limitations as well.

People who have higher skills,

if you think of some certain basic level accountancy, certainly,

but if you think of other accountants' function,

let's say you're also a tax expert,

that expertise would still be needed, and so you would be using this technology,

but the potential for your occupation to continue is very high.

-One of the things you point out in this report,

which caught my eye,

was that a lot of these clerical jobs are primarily held by women.

Is what you're saying that the effect of Generative AI

is not equally distributed between men and women?

-Yes. That pretty much has to do very much

with the way that men and women are represented in those different jobs,

and those clerical jobs

that came up on top of highly exposed job categories

are largely overrepresented in many countries by women

when it comes to the employment share.

Basically,

when we separate the share of male and female employment in those jobs,

what we see is that the share of female employment

in this category with high potential exposure to automation

is more than two times higher

than the male share of such jobs.

That is particularly visible in high-income countries,

where also more said jobs in general are found

because as you go down the income bracket,

there are fewer clerical jobs of this type

in general within employment structures of countries.

We have to remember that these office jobs

have been a significant source of employment for women

in the typical process of development.

Also, when countries increase their income,

there is a certain segregation of jobs

into which men and women tend to go.

Typically, elementary jobs decrease,

and then as men and women go into employment,

many women choose to find jobs in these office jobs,

and they often found good jobs.

What we're also saying in the report is that there is this risk

that if this is the kind of jobs that could be affected,

the effect could be disproportionately harmful for women.

-What you're saying is that these jobs in some economies,

they're stepping stone jobs for women to get into the white-collar workplace.

If you remove that stepping stone,

it's going to make it much harder

for this particular grouping to get into formal white-collar work.

Is that essentially it?

-I'm not sure if they're a stepping stone as such.

I think there are jobs that simply exist in larger numbers,

and especially in high-income countries,

there are simply many people working in this type of jobs.

As a share of employment among men and women,

there tend to be more women working in this type of jobs,

but they are not necessarily a stepping stone.

They are just jobs in their own right,

and they have existed as, in many cases, good jobs over a long time.

-Now you've both alluded to the issue

that the effect on different income level countries

is different.

Do you want to unpack that a little bit more?

-Yes.

The reason why in our analysis we find that the greatest impact is

in high-income countries and then upper-middle-income countries,

and it really goes down depending on the level of economic development.

The primary reason for this

is really because of the economic structure of countries.

Countries at lower levels of economic development

have a much higher share of employment in agriculture.

They also have many more people in jobs

where you have to be physically present in transport,

food vending, all of these occupations

that would not be automated by this technology.

That's one reason.

The other reason, of course,

is that there's also important digital divides.

There's much less access to internet.

Even electricity can be a source of the problem.

Also, problems with digital skills.

As a result, you just don't have as many jobs in these countries

that have the potential to be affected by the technology.

Even if they do have the potential to be affected

just because of the occupation in itself,

the chances that that technology will be incorporated are much less.

Also, because the cost of labour is lower.

In terms of just looking at the numbers with the exposure,

for example,

with respect to automation

we find that in high-income countries it could be potentially 5% of employment,

whereas in low-income countries, it's less than 0.5% of employment,

so it's a really big difference.

For augmentation on the other hand,

the potential benefits of augmentation are almost similar.

It's a little bit lower in low-income countries,

but it's still very high, so 10% compared to 13% in high-income countries.

Again, the problem is

if you don't have the access to the technology,

if it can't be incorporated into the world of work,

then the potential productivity benefits of this technology

won't be realized by these countries.

-Yes. You are going to widen, as you've just referred to,

the digital divide between the haves and the have-nots.

-Exactly.

-I also want to drill down a little bit more into

what you were saying about augmentation because it sounds good.

I don't have to do the boring tedious bits of my job

and I can focus on the bits of my job where I can add value.

If your boss is an algorithm or your boss is AI,

doesn't this also not take away some of workers' agency,

some of workers' voice

and actually diminish the quality of the job?

Do you both want to chip in on that one?

-Sure. Of course, there is that risk.

The question of impact of algorithms on the quality of work,

it's not something that just appeared with the emergence

of these large language models or Generative AI.

That's something that we have been discussing at the ILO

for a long time.

What you refer to is part of it.

What we see looking at the tasks

that belong to different occupations is certain risks.

Let's imagine a job of a call centre worker

or someone who has to do with customer services.

If, now, this type of Generative AI is able to respond

to some of the easier questions to deal with queries

that are easy to automate,

that person might start receiving much more difficult questions.

If the performance framework on which they are evaluated, for example,

how much time they spend on the task

and how well they deal with it is not adjusted,

that's one part of a problem

because you basically change the landscape without adjusting the way

you look at a certain occupation.

The other part, of course,

is the question of stress that generates and pressure.

Imagine that as a technical support,

you suddenly start getting only angry customers

and difficult questions.

That's not the same as having a mix of that.

I think we all in our daily jobs have easier tasks and difficult tasks.

While this example of a call centre worker helps to illustrate,

let's imagine that in our daily jobs, it happens the same.

All the easier tasks are automated or are supported by a chatbot

that is pre-trained on a lot of knowledge that we had produced

and suddenly we are only faced with the more difficult tasks.

That's not an innocent change

and that's something that requires quite a lot of attention.

-Janine, do you think there has been much appreciation yet

of this potential effect on job quality?

-No. The overwhelming emphasis has really been

on the potential for job disruption and fears about the end of work.

I think what the paper really shows is that there are a lot of jobs

that are going to be affected by this technology.

That could be good and that could be bad.

It really depends on the design of the technology,

how it's integrated in the work environment,

do workers have a say on the design, on its integration,

is there a feedback process.

All of these issues really affect day-to-day experience at the workplace

and so they affect people's working conditions.

I think there hasn't been as much emphasis on this as there should be,

but I think that's starting to change.

-What do you think are the key takeaways

that you would highlight to policymakers from your paper?

What do you think they should be thinking about?

-There's quite a few policy messages that come out from the paper.

Related to this topic, specifically, I think the importance of workers' voice

is central both on technological design

and technological incorporation at the workplace,

but also processes.

There are some areas where you don't want

AI to be determining people's dismissal or people's pay.

You need to have a human in command.

There is definitely need some policy intention on those issues,

also on dispute resolution about having humans involved.

I think all of that is really, really important,

but of course, there's also policy issues

that relate to the transitions in the labour market,

because you know that some people might be losing their jobs,

and so you have to have income support mechanisms in place,

job training, skill systems in place, protecting workers during this transition.

There's quite a few issues that policymakers need to be aware of.

-And of course, the gendered issue that we talked about earlier on,

the fact that the impact on different genders

and indeed on different country

income groups is going to be significantly different.

Do you think people are starting to appreciate that?

Or are they still focusing too much on trying to regulate the AI itself

rather than regulate its impact on the world of work?

-Right. Most of the discussion about the regulation of AI,

and it's still incipient in most parts of the world,

really has been focused about AI in general.

There hasn't been enough attention on the world of work.

I would say there hasn't really been enough attention, too,

on just having systems of social protection in place

and systems of job training in place that can help workers

during all the different transitions that they face in the labour market.

This is something that we make a call for, for policymakers to be aware of.

-What is your next stage of your research?

You're going to keep going on this, I assume.

There's plenty more to do, yes?

-There's plenty to do, and of course, we have plans.

We have ongoing things happening already in terms of research.

One of the things we're looking at is basically the bias of these tools

and understanding a little bit more

how that can affect their integration to the world of work

because we are basically dealing with a situation

in which the AI has left the lab

and is now integrating and is being used by the public

and we are seeing that it's now increasingly possible

to design bespoke products on the basis of this technology,

and they are increasingly being integrated into the professional context.

We really have to understand much more the bias

that's in those systems

and how that could affect

if they are integrated into the world of work, but also--

-Because it reflects the biases with which it has been programmed?

-Yes, that's another thing about algorithms,

but we don't know exactly how

and that, of course, depends on when the model was trained,

what type of data it was trained on, which model, and so on.

We're trying to understand that a bit.

We also are very keen to work

more on country-level studies to unpack this global assessment

that we've done here into a country-level reality.

We also are looking at the new jobs created.

That is an area where we would like to develop some work,

which is understanding basically how much work is involved

behind the scenes in training of those models.

We know that this is an important issue, also, for policymakers to focus on

because these jobs are often not visible,

but training these models requires a huge amount of work.

Another area that I think we should increase the understanding

is the environmental impact of this technology.

As demand for it grows, as it's being integrated into work,

we know there's an environmental impact,

but we don't really know how to calculate and assess it.

We should probably look at that and of course,

all this work fits into the broader effort

of many other colleagues here at the ILO that look at regulations around AI

and how that can impact its application to the world of work.

-Of course, what makes your task even more difficult is

this is moving so fast.

We do a lot of research in the ILO, but in a lot of those other areas,

the rate of change is slower,

but here, it must be just very difficult to keep up.

-I mean, there's certainly literature coming out every day on the topics

that one needs to be on top of.

I think as far as the broad changes, we are following a line of history

and the technological adoption at the workplace I think continues.

It's new, but there is a lot that you can learn from the past as well.

-I'm sure we will be returning to this topic

because as we've said, it is moving so fast

and it is so interesting

and potentially will have such an enormous effect

on the world of work.

Unfortunately for this program, that's all we've got time for.

Thank you both very much to Janine and Pawel for joining us.

If you want to know more about their report

or its methodology and its background,

or the details of its findings, you will find it on the ILO's website,

and I highly recommend you download it and read it

because it's going to be important.

Thanks to you, our listeners, again for your time and attention,

and I hope you will join us again soon.

For now, goodbye.

[music]