The impact of generative AI on jobs in Latin America | The Future of Work Podcast

This post was originally published on this site.

-Hello and welcome to the ILO’s Future of Work podcast.

I’m Sandra Kuchen.

From automating routine tasks to creating innovative solutions,

Generative Artificial Intelligence or AI is revolutionizing the way we work.

How is it affecting different parts of the world?

Who has access to these AI technologies and who doesn’t?

Does it risk creating a growing digital divide

which could fuel inequality?

A research paper recently published jointly

by the International Labour Organization and the World Bank

has been looking at these questions,

and in particular, it focuses on the situation in Latin America.

Joining me now are the co-authors of the paper.

Pawel Gmyrek is a Senior researcher at the ILO

and Hernan Winkler is a Senior economist at the World Bank.

Welcome Pawel and Hernan.

-Thanks for inviting us.

-Before we start our conversation,

I’d first like us to listen to two people from Buenos Aires in Argentina

about what they think of generative AI tools

and how they are using them.

-Hi, my name is Rosario.

I work in the human resources area of a company

that deals in health, more specifically a sanatorium.

In our case, to be honest, AI has had no impact.

Today, we don’t use any tools.

I think it would be a great idea to start using them

because there are many people in our sector

who do administrative tasks that I think could be 100% automated.

This wouldn’t mean replacing this part of the staff,

rather the contrary.

I believe it would give them the opportunity to do higher-quality tasks

that would have a greater impact for our sector.

In no way do I believe that AI can fully replace people,

rather I think it could increase the sector’s productivity.

-Hello, I’m Martina. I work at a software company,

but I work in an area

that has more to do with support as an account executive.

I use AI as a tool to complement my work

because it helps me speed up some of the things,

write some emails, but always under my supervision.

It’s one of the many tools I use nowadays.

I don’t think they’re going to replace us tomorrow,

but all the dynamics as we know them today

are going to change a lot.

New jobs will emerge, some will perhaps be eliminated.

I think we shouldn’t demonize it,

but we shouldn’t look at it as all roses either.

It’s a bit of a mix of the two and hopefully,

laws will begin to act faster to address these things

that are happening so quickly.

-Pawel, let’s start with you.

What are your thoughts on what you’ve just heard

and how does it link with your research findings?

-I think it resonates very well with the key findings of our paper, actually,

because what we show is that

there’s a lot of potential for these AI tools,

but also a great divergence among countries,

but also among occupations

in the extent to which such tools are accessible

and applied for the moment.

These differences are clearly visible

across sectors as illustrated by the two experiences

you just heard.

Rosario, as I understand, works in the personal care sector,

and that’s a sector that naturally has many tasks

where a human role is crucial,

not only because machines cannot do them,

but because we would want them to be performed by humans

in the first place, for example, taking care of patients.

Personal care is also among those sectors that could potentially have a lot to gain

from AI-driven augmentation of tasks.

It has a large amount of administrative tasks,

probably with the growing amount of admin procedures

that one has to comply with,

and such tasks could be made more efficient,

freeing up people to deliver better health services

or simply improving their working conditions.

Then the second speaker, Martina, is on the other side of that spectrum.

She’s in software development,

the ICT sector more broadly,

and such sectors are already highly digitized.

That means that innovations like AI

quickly spill over from the core functions like programming

to other functions like accounting or marketing,

where she works.

We have to remember that the IT sector is very specific in this case

because it’s used to constant tech-driven evolution.

When it comes to some tasks getting automated by technology,

many IT companies will come up with new tasks

and applications that drive competitiveness

rather than immediately think,

“How do we reduce the human factor?”

In other sectors, there might be much more of a temptation

to use these advancements just to look for labour-related savings,

which would be quite bad for individuals affected,

but also for this bigger macroeconomic picture.

These type of divergent effects across groups and sectors

is exactly what we’re looking at in our paper.

We try to go deeper

into the profiles of people in Latin America

who might be affected by generative AI,

and we try to provide a picture of who they are

in terms of their demographic profiles,

which jobs they are in, which sectors,

as well as their levels of education or types of contracts they have.

Essentially what we do,

we take all the existing occupations and look at their exposure to AI,

but then we look at the extent to which these jobs are digitized,

whether they use internet at work, do they have access to a computer,

and we leverage a vast number of country-level surveys

or about 920,000 individual responses

from the World Bank’s repository and ILO’s microdata repository,

and we get to that detailed overview,

which then lets us project more detailed potential effects of AI exposure

for those countries and for the region.

-It sounds like a very dense data set that you’ve been working with.

It’s really impressive.

I’m curious, too, why Latin America?

Hernan, you work in the World Bank’s Latin American section,

and you’re originally from Argentina.

Maybe you can help us shed some light.

Why did you choose Latin America as your region of focus for this research?

-Yes, that’s a very important question.

We chose Latin America mainly for two reasons.

First, it is one of the regions in the world

that has been quite stagnant, actually, for more than a decade,

and this lack of dynamism is reflected in the labour market.

You see that the quantity and the quality of jobs

and labour productivity levels are very low

when compared to other, more dynamic regions,

such as East Asia.

While there are many ideas,

many hypotheses that could explain this poor performance of the region,

the usual suspect is often the slow rate of adoption of new technologies.

That is, people and firms are not adopting new technologies as rapidly

as other regions in the world.

When we found out about the emerging experimental studies

that are showing that generative AI can have positive impacts

on labour productivity of certain occupations,

we immediately thought, “Well,

can Latin American workers take advantage of this new technology?”

The second reason for choosing Latin America

is that it is one of the most unequal regions in the world

according to most measures of income inequality.

That means that it is one of the regions

where the income gap between the poor and the rich is the largest.

Some of the experimental studies that I just mentioned

show that generative AI can help reduce inequalities

among selected occupations

because they tend to help inexperienced or low-skilled workers

more than the rest of workers.

Naturally, we thought, “Wow,

there is a new technology that has the potential to increase

productivity and reduce inequality.

We should definitely study this a bit more.”

Spoiler alert,

things are not as simple or rosy

as they seem in these initial experimental studies.

-OK, so let’s dig a little bit deeper then.

What else has your research revealed

on the exposure of Latin American labour markets to generative AI?

Who might use this technology and in what sectors?

Hernan, do you want to start?

-Yes. We find that about 26% to 38% of jobs

in Latin American countries are exposed to generative AI.

What do we mean by exposure?

We mean that some of the tasks that workers perform in these jobs

can be easily delegated to this technology.

We should be very clear about this.

It doesn’t mean that all these jobs will face the risk of disappearing.

In fact, we find that only about 2% to 5% of jobs

are at risk of full automation.

On the other hand, we find that about 8% to 14% of jobs can be augmented,

that is, to become more productive thanks to this technology.

Then there is a significant fraction of jobs,

between 14% to 22%,

that are exposed, but it’s still unclear

if the impact will be more towards automation or augmentation.

It will depend on many things,

such as where this technology is going,

the government policies, and so on.

Maybe Pawel wants to add a little bit more?

-Yes.

Thanks, Hernan. That’s a great intro.

I think maybe I can link to the title of our paper,

which was Buffer and Bottleneck.

Then when we were writing up conclusions,

I was thinking, we should actually call it

“No Buffer and Big Bottlenecks”,

because one of the things we tried to explore is

to what extent this lack of access or digital divide,

the fact that some people at work

do not have access to a computer or Internet or both

might protect some jobs from the immediate impact of AI,

especially the negative one.

What we find out in the paper is that among the jobs,

even though they are fewer,

those jobs that are exposed more to the risk of automation,

among those jobs, most people already use a computer at work.

There’s basically no buffer in a sense that

that technology is already able to penetrate into those jobs.

However, there are big bottlenecks because in the region,

among the occupations

that could benefit from this type of productive transformation

where you either automate or speed up some of the tasks,

up to half of these jobs in the region of Latin America

wouldn’t be able to benefit from this technology today

because they don’t have access to basic or crucial digital infrastructure.

-I imagine our listeners, especially when they hear you

talking about automation,

there’s that question which keeps coming up

about replacement of people’s jobs with AI.

Based on your research findings, to what extent do you see

this generative AI technology starting to replace people’s jobs?

Hernan, do you want to address this?

-Yes, that’s a very important question and reflects the concerns of most workers.

In fact, that’s how we motivate our study.

When we begin, many workers manifest these concerns in surveys.

When will we begin to experience them?

This is a new technology,

so it is very difficult to predict the timing for its impacts on workers.

Based on our research, we can say two things.

First, that the impact may arrive sooner

for workers in occupations exposed to automation

because they are already highly digitalized,

as Pawel mentioned earlier.

Probably most of our audience has felt that

every time we try to contact

a public service provider or an airline company,

it is more and more common to be talking to a robot

and not to a real human.

Our findings suggest that this trend would actually accelerate.

Second,

impacts on jobs in the developing world

are likely to take longer than in rich countries.

For example, in Latin America,

millions of workers are in informal jobs

without any access to digital technologies,

and where face-to-face interactions are key.

The adoption of generative AI in this sector will not happen overnight.

-In your findings and conclusions for your research,

what is applicable, do you think, beyond Latin America and other regions

who are also contending with this rise and rapid spread of generative AI?

-That’s one of the reasons why we actually wanted to do this study

focused on Latin America,

because Latin America is a good proxy for non-high-income countries.

If you look at the current literature,

it’s mostly focused on the jobs

either in the US or in high-income countries,

and Latin America region has countries in nearly the full spectrum of income.

From lower-middle income to upper-middle income countries,

and therefore it can serve as a proxy for other regions

since most countries globally are not high-income countries.

Let’s imagine this.

We come to this observation about AI and the digital divide,

among jobs that could be augmented in Latin America,

up to half don’t use a computer at work.

Then what we demonstrate in our paper is that there is a very strong correlation

between the country’s level of income

and individuals’ access to a computer and internet at work.

Of course, there are some important exceptions,

countries that made it a priority to digitize

and have rates of internet access

higher than what their levels of income would suggest.

Generally, this relationship holds across the LAC region

but also across the world.

One of our conclusions on that basis is that

if you were to redo the same study for other regions,

you would find a similar relationship

between internet access at work or computer access at work and income.

-Reducing the digital divide is critical.

What are other recommendations you make in your research paper?

-As you mentioned in the opening,

it’s a joint paper by the ILO and the World Bank.

Yet, it was surprisingly easy to agree on the recommendations.

That is because the numbers and calculations

make it clear what the general picture is.

Of course, our study just gives a broad overview,

but these country-level policies can be made more nuanced,

more specific through consultation and dialogue with social partners

at the country level

and at the sector level.

There is clearly an important role for policymakers

to develop interventions in two directions.

On the one hand, minimizing the disruptions

resulting from sudden job losses

caused by the onset of this technology,

and on the other hand, maximizing the productive benefit

of the transition that generative AI can induce.

We know from studies, studies show very often

that protecting jobs

in times of disruptions has better macroeconomic results

than having such jobs lost

and then trying to make them reappear.

That’s especially important in regions with high informality

because if formal jobs are lost,

they might not come back to the formal sector.

Protection of jobs, helping workers reskill,

and using social protection systems to stabilize these transitions

is one of the two packages in the hands of policymakers.

We said that such policies

should pay particular attention to the gender dimension

since we show that women are disproportionately exposed

to the risk of job automation.

In addition, there’s a need to invest in workers’ foundational skills

so that they are able to work with these new AI tools

within the context of their existing occupations

and that they can seek additional productivity

and creativity benefits.

That fits neatly into considerations and discussions

around lifelong learning policies and programmes.

Finally, maybe from my side,

something that’s also important to the ILO,

we should recognize that

this type of technological transformation and the same tools

can actually offer opportunities for what we call e-formalization

through innovative government services

and through the use of digital technologies

and that policies should explore such tools for formalization

using the moment of this transition.

I’ll leave the rest to Hernan.

I’m sure he will have a lot to complete here.

-I think you provided a very complete picture.

I would just emphasize the part of the access to digital technologies.

Within Latin America, in many countries, it’s very expensive for households,

especially for those at the bottom of the income distribution,

to pay for a good access to a good, reliable internet service.

This is something where, of course,

there is a lot of room for policies for interventions.

Of course, it’s not just about the telecommunication sector.

It’s also about the incentives for firms, for people to adopt this technology.

This is done through promoting a business environment

that is conducive to competition, to firm entry,

to productivity growth,

and also to promote skills

that allow the workers to learn by themselves.

It’s this lifelong learning idea that Pawel mentioned,

that as a new technology gets introduced,

workers have the capabilities to learn

how to use this to be more productive at work.

-It sounds like there really is still a lot to be done to ensure

we get the best out of generative AI,

including creating a just transition to these new ways of working.

I’m sure we’ll be returning to this subject soon.

For now, that’s all we have time for.

My thanks to you, Pawel Gmyrek, and to you, Hernan Winkler,

for sharing your insights with us.

Thanks also to you, our listeners.

If you’re interested in finding out

more about the impact of generative AI on the world of work,

the ILO has recently launched an observatory

on AI and work in the digital economy.

You can also get updates on the ILO’s work by following our social media channels.

Our handles are @ILO on Facebook, LinkedIn, TikTok, and X.

On Instagram, we are @ilo.org.

Once again, thanks for listening.

Please join us again next time.

[music]