What will Artificial Intelligence bring for the economy: vast job displacement or staggering productivity gains? To understand the issue, Carlo Pizzinelli and Marina M. Tavares argue, we have to take account of the social, ethical and physical contexts within which human labour will be affected. This means looking not only at levels of AI exposure but also, crucially, at measures of AI complementarity.  

The fast roll-out of Artificial Intelligence (AI), and in particular Generative AI, in different aspects of our lives has sparked talk of a new industrial revolution. Its implications for labour markets worldwide could be profound and complex, promising to enhance workers’ productivity while simultaneously posing the risk of vast job displacement, with trade unions in some sectors already mobilising against its deployment.

Assessing the impact of AI on employment is difficult because the adoption of this technology is still in its early stages and there is great uncertainty over the exact nature of its integration across various production processes. Even so, attempting an answer to this question is essential if we are to step forward into a future in which AI brings about a positive and inclusive structural transformation of the economy.

Understanding AI’s role through the “task framework”

So, where to begin? Borrowing from the academic literature on automation, a promising approach to understanding the potential impact of AI on labour markets is the so-called “task framework”, whereby workers’ occupations are viewed as bundles of tasks that need to be carried out. Recent studies have identified which tasks can be performed by AI, and developed measures of “occupational AI exposure.” Such tasks include image recognition, summarising large amount of information, translation, and coding. This approach provides valuable insights by establishing a well-defined framework. However, it also offers a narrow view.

Whether AI can perform a task does not necessarily imply that, in practice, it ultimately will be used for this purpose. More importantly, will the use of AI require human oversight or will it be allowed to fully replace workers in the performance of a given work activity?

This latter question is what policymakers are ultimately asked to respond to. And to do so, in the context of a general-purpose technology, requires thinking about the interaction of the labour market with a broader set of considerations, including social norms, ethical concerns and legal aspects.

Accordingly, in a recent working paper, we move beyond the basic “task framework” to consider not only measures of AI exposure but also an index of potential “AI complementarity”, or shielding, to capture these additional dimensions. This index, which accounts for the social, ethical, and physical contexts of occupations as well as required skill levels, indicates how likely an occupation is to be shielded from AI-driven job displacement. When paired with high AI exposure, it also provides insights into the potential for AI to complement tasks within that occupation.

Workers in jobs with high exposure and high potential complementarity, we argue, are more likely to benefit from AI integration: productivity is enhanced through the automation of routine tasks and the augmentation of more complex ones. Meanwhile, jobs with high exposure and lower complementarity would be more likely to experience the risk of direct substitution of human work from AI systems, potentially leading to reduced labour demand, lower wages and, in extreme cases, job losses. In other words, the same task may interact differently with AI depending on various contextual factors, such as a worker’s level of responsibility and seniority, and more broadly by societal preferences.

Jobs with high exposure and low complementarity are more likely to experience the risk of direct substitution of human work from AI systems

How AI is reshaping work

The legal field serves as a prime example of AI’s role in professional settings, where advanced technologies assist but do not supplant human judgment. The advanced textual analysis capabilities of natural language processing technologies implies that judges are highly exposed to new technologies. AI systems can help to parse through large volumes of legal documents to identify relevant precedents and inconsistencies, thereby speeding up case processing. However, given the significant repercussions that judges’ decisions have on individual lives, it is very unlikely that societies would allow unsupervised AI algorithms to make final judicial decisions in the foreseeable future. This means that AI is more likely to complement judges by enhancing their productivity rather than replacing them.

Paralegals are also highly exposed to AI, as their job entails research, textual analysis and drafting. However, similar to other clerical professions involving lower levels of responsibility and high-stakes decision making, the high exposure to AI in this case bears higher risk of worker displacement. This example illustrates the potential for AI to exacerbate existing inequalities by affecting those at the top more favourably than those in lower-paid positions.

Analogously to the legal field, similar disparities can be observed in healthcare, where AI assists in diagnostic processes but cannot replace the empathetic duties of medical professionals, and in finance, where AI handles transaction processing and fraud detection but leaves judgement and strategic decisions to human expertise.

The impact of AI on labour markets across the globe

To assess the impact of AI across the world, we estimate the share of employment in occupations with low and high levels of exposure and complementarity for a large set of countries. We find that a striking 40% of global employment is exposed to AI, reflecting the potential reach of these technologies across different jobs (Figure 1). However, this share varies substantially across countries. In advanced economies, about 60% of jobs may be affected by AI, roughly equally split between those with high and low complementarity. Consequently, advanced economies, which have a higher concentration of white-collar jobs, face greater risks from AI but also more opportunities to leverage its benefits.

By contrast, in the emerging markets and low-income economies, exposure to AI is estimated at 40% and 26% of employment, respectively, roughly equally divided into low- and high-complementarity jobs. Most workers in developing countries thus face fewer immediate risks of disruptions. At the same time, the lack of infrastructure or skilled workforces to harness the benefits of AI in these economies raises the risk that over time the technology could exacerbate existing inequalities among nations.

Figure 1.  Employment Shares (Per Cent) by AI Exposure and Complementarity

AI and inequality within countries

AI could also affect inequalities within countries, with workers who can harness AI enjoying rising productivity and wages, while those who cannot falling further behind. For instance, research shows that AI can help less experienced workers enhance their productivity more quickly. Younger workers may find it easier to exploit opportunities, while older workers could struggle to adapt.

The effect on income inequality will crucially depend on the extent to which AI will complement or disrupt high-income workers relatively more than low-income ones. In our paper, we find that the incidence of low complementarity jobs is fairly constant across the income distribution, while high-complementarity employment is markedly higher in the top of the income brackets (Figure 2). In other words, while the risk of disruption – including job displacement – seems to be evenly distributed across the income distribution, the potential benefits are skewed towards the more affluent households.

Figure 2: Share of Employment (Per Cent) in High Exposure Occupations by Earnings Decile

Shaping an inclusive AI future: policy recommendations

The impact of AI on global labour markets will likely be uneven. The risks of job displacement and the opportunities for productivity gains will depend not only on what the technology can do but also, crucially, on the wider context surrounding any given job, which is shaped by societal preferences for the use of AI.

Importantly, the effects of AI can be shaped by policies. Comprehensive social safety nets and retraining programs will be essential to ensure that the workforce can transition to new opportunities. Educational curricula should also be adjusted to equip new generations with the right skills from the start. Regulatory frameworks, such as the European Union’s AI Act, will also play a critical role for labour markets: in shaping the safe and ethical deployment of AI, these frameworks will determine how the technologies are applied in different jobs. Trade unions are also key to give workers a voice in this process. Poorer economies, where there is a more limited coverage and quality of digital infrastructure, will have to expand access to a larger share of their population.

As we stand in the early stage of this transformation, the task for policymakers, employers, and unions, is thus to collaboratively foster an environment in which AI can ultimately contribute to inclusive growth.

The views expressed in this article are solely those of the authors, and should not be attributable to the International Monetary Fund, its Executive Board, or its management – nor do they represent the position of LSE Inequalities or the London School of Economics and Political Science.