Artificial Intelligence, Education and Work: Why I'm still concerned about machines replacing us at?work

By Fernando Schapachnik, Associate professor of the Department of Computing, Faculty of Exact and Natural Sciences, University of Buenos Aires, Argentina.

Article developed for IADB educational blog (original version in Spanish here).

To be honest, I am not worried about machines replacing us, bluntly put, but I am worried about the growing advance of artificial intelligence (AI) in the world of work, and more specifically, that we are erring on what we believe are the public policy responses necessary. I’m going to explain myself.

Will there be a job?crisis?

There are all kinds of predictions about what is going to happen in the world of work with the impact of AI, and therefore it is difficult to bet on one single prediction, but I do believe that we can draw a picture of what is going to happen in Latin America.

The first thing to understand is that although the deployment of artificial intelligence solutions seems to be advancing much more slowly than was thought when ChatGPT burst onto the scene in November 2022, there will be an impact on the world of employment.

A study of the International Labor Organization in 2023 took standardised databases that break down each job into tasks, and analysed which of these could be carried out by generative artificial intelligence. Based on that, they identified 3 groups of jobs:

• On one hand, jobs where most of the tasks were automatable, which were therefore considered at high risk of total automation. This is “only” 2.3% of global jobs (75 million jobs).
• 13% of jobs (427 million jobs) were classified as complementable, meaning that an increase in the productivity of those who perform them is expected, given that generative AI will be able to automate some of their tasks, but not most.
• Finally, 9% (about 281 million jobs) remain in between, in what the report called the big unknown, because they could fall into any of the other two groups, depending on characteristics that are difficult to analyze in general, and on public policy decisions.

This gives us a base of 356 million jobs at risk. When analysing the particularity of some regions such as Latin America, it is important to also zoom in on the idea of complementation. In weak economies, with many difficulties, such as those of the countries in our region, it is hard to think that such an increase in productivity will not translate into some level of job redundancy. Simply put, if my colleagues and I become twice as productive because some tools help us do our jobs faster, how much do our company’s sales have to increase, so that none of us are considered redundant?

Classical economic analysis usually pairs this panorama of job destruction with the creation of new ones. Now, the question is what jobs will be created, and where?

If increased productivity and technological innovation led to the creation of new jobs, these would necessarily fall among those that cannot be automatized. Therefore, they would either require greater intellectual complexity or be manual jobs to fall outside the scope of automation. And this is where important challenges for Latin America arise.

On one hand, because manual work tends to be low-skilled and badly paid, and on the other, because in order for a Latin American citizen to be able to access a highly complex job, two conditions must be met:

1. The first is that his or her employer company locates the job in the same region, and not in their headquarters in central countries. It is still a pending issue for our continent that the design centres, the central offices and the management positions are located in the same territory as the factories themselves. It is still very common to find transnational companies with an innovation laboratory in Europe and offices in Miami from where the planning and commercial operations of all their Latin American branches are managed.
2. The second condition is that this citizen has the necessary qualities and qualifications for that highly complex job position. We cannot take the latter for granted: in Latin America, three out of every four students do not achieve minimum skills in mathematics (compared to 31% in the OECD), 55% do not have basic reading skills, 57% do not reach basic level of proficiency in sciences, and there is a high school dropout rate, as shown in the most recent IADB report on the state of education. In the population between 19 and 23 years old, only 32% attend higher education, according to a recent OEI study.

There are reason to be worried also in developed economies: the historical evidence related to automation, although it does not account for unemployment, does indicate an increase in inequality: a study by the Nobel Prize winner in economics Daron Acemoglu and Pascual Restrepo documents that between 50 and 70% of the change in the North American salary structure in the last four decades is explained by automation, and that this change has led, from the 1990s to the present, to a separation between the income level of the population with secondary schooling only vs. those who obtained a higher level of education:

“In particular, automation reduced the relative, and in some cases real, wages of workers specialising in routine tasks in industries undergoing rapid automation (such as those working in blue-collar jobs in manufacturing industries that introduced numerical-controlled machinery and industrial robots, or those in clerical tasks in industries that introduced software-based automation). In contrast, worker groups that were not displaced from their tasks, such as those with a post-graduate degree or women with a college degree, enjoyed wage gains.”

Summing up, although it would be an exaggeration to say that humans will be replaced by robots in their jobs, there are reasons to worry about the increase in inequality, particularly in Latin America.

The answer is training, right?

When discussing AI and work, two words are predominant as a policy response each time the disappearance of certain jobs is mentioned: upskilling and reskilling. Rephrasing them in less catchy terms helps clarify their meaning: job training for adults.

That phrasing allows us to wonder what is known about the effectiveness of such policies. And the answer is that a lot is known. There is extensive literature on the analysis of Active Labor Market Policies (ALMP for short), trying to analyze these types of interventions in a rigorous manner. The research question is how many people who took a specific training course later find a job, compared to others with similar characteristics, but who did not participate in such training. Such a framing allows us to analyse the effects of the training itself, and separate the influence of other factors such as the general employment demand of the economy in that specific region and at that specific moment. These types of studies are called randomised controlled trials (RCT for short) and they are the most reliable tool to understand the impact level of these interventions in the world of work. A percentage is obtained which represents how many people, out of every 100, will get a job as a consequence of having participated in the training but would not have gotten it without that same training.

The results are discouraging. A 2017 study by David McKenzie’s on developing countries found that out of 9 training policies analysed, only 3 were able to demonstrate improvements in the employability of their beneficiaries, averaging around 2.3%. That is, of every 100 trained people, less than 3 found employment.

These results appear consistently in the specialised literature. As an example, we can cite a 2016 study by the ILO focused on Latin America and the Caribbean, which analyzes 207 RCTs of AMLP interventions aimed at employability, of which 19 occurred in the region and except for one, all were about training, mostly on young people, and another study, this time from 2019, which analyzes 102 RCTs of interventions aimed at employability. The first finds that the positive impacts range between 5% and 20% (for a particular one, aimed at young people), and the second that the training had an average impact of 7.7%.

Put another way, the evidence indicates that, at best, for every 100 adults who receive job training, only 20 will find a job, and probably even less.

Job training for adults, upskilling and reskilling, cannot be our only public policy response to cope with the advance of artificial intelligence on jobs.

Then, what is the answer? What is it that we really should?do?

This article wants to reopen a debate that an excessive hope in training seems to have closed. Unfortunately, I do not have categorical answers for the short term. I offer instead some ideas that I deeply believe in, but require longer periods of time to blossom.

The first one is that it is more urgent than ever to ensure that everyone in Latin America has the possibility of accessing quality higher education, because that will be the requirement for many, perhaps for most, of the high-paying jobs that will not be automated. Some will require formal accreditations (degrees in engineering, medicine, etc.) and others will not, but they will require a cognitive demand compatible with higher education, precisely because they will be the positions that escaped automation by AI.

Today, more than ever, it is essential that school provides basic learning to the entire population, but also that it develops analytical capabilities and critical thinking (the higher levels of Bloom’s taxonomy). The objective is being able to address complexity. (Intellectual honesty requires equating demand with recognition: I cannot help but point out the contradiction between demanding more from school and not increasing its resources, between demanding more from teachers, and not prioritising their profession, especially by means of better salaries, as a central component.)

Critical thinking blossoms slowly, takes continuous effort and all the school time we can devote to it. It’s an expensive mistake to think that developing workplace readiness skills means using scarce school time to teach mundane, temporary, perishable contents that will mutate at full speed, based on fashion, temporal trends and fragile technological statuses. If it can be learnt through an online tutorial, it’s not school-level content.

Teaching workplace readiness skills, more than never, means developing strong cognitive foundations, so that adaptation can indeed happen, it means readiness to face complexity, the same complexity machines cannot cope with. And that requires critical thinking.

If I had to summarise what critical thinking is about, I would ask for a schooling that would educate us with a mixture of skepticism and discomfort. We have to be skeptical to be able to analyze what is proposed to us, to be able to avoid trends whose only virtue is being fashionable, and at the same time, to be non-conformists with the status quo, to push forward, to not accept things as they are. The advance of artificial intelligence in the world of work makes the need for a school that prioritizes thinking, which has always been important, to become urgent.

#ArtificialIntelligence #AI #LatinAmerica #Education #Upskilling