Aristotle speculated that tools could end human labor if they could be made to work on their own. Centuries later, Keynes warned about the potential job losses from technological advancements. Neither of them, however, could have predicted the potentially massive impact of Artificial Intelligence (AI).
Natural language processing and generative models like Open AI’s ChatGPT have so dramatically improved machines’ ability to mimic people’s cognitive capacities that we are in a vastly different world from even that of a decade ago. These technologies’ ability to understand and generate human-like text, make predictions, and perform image and video recognition could change the labor market. The convergence of AI with robotics to create autonomous systems capable of highly complex physical tasks could be similarly dramatic.
Nonetheless, the potential effect of AI on employment across the range of known jobs remains controversial and uncertain. Estimations vary widely, subject to human bias and methodological limitations.
A New Index
In a recent study, we sought to overcome some of those limitations by using the extensive knowledge and advanced analytical capabilities of large language models to create a novel AI-Generated Index of Occupational Exposure (GENOE). The index estimates the likelihood of occupational vulnerability to AI for more than 750 professions over one-year, five-year, and ten-year time horizons. GENOE employs a novel methodology that utilizes large language models (LLMs) to conduct expert-like assessments, which we term “synthetic AI surveys.” This approach generates responses based on vast datasets, rapidly processes and synthesizes information, and provides consistent evaluations across occupations, offering a scalable alternative to traditional human expert surveys.
Our findings show that, on average, occupations have a 28% likelihood of being impacted by AI within the next year, with that rising to 38% and 44% over five- and ten-year horizons, respectively. Calibrating those findings to labor market data from the United States and Mexico, we estimate that 43 million and 16 million jobs will be exposed to AI, respectively, over the one-hear horizon, with that number increasing to 60 million and 22 million over five years, and 70 million and 26 million over ten years.
Findings on Latin America and the Caribbean
Extrapolating to the region as a whole, we see that about 84 million jobs will be exposed to AI within a year, rising to 114 million in five years and 132 million in ten years. And assuming that this average exposure is applicable to worldwide labor markets, 980 million, 1.33 billion, and 1.54 billion jobs would be exposed to AI over the one-, five- and ten-year horizons, respectively.
These estimates do not directly correspond to job losses. Occupational transformations and the emergence of new roles in some sectors will likely complicate the picture. Nor do they account for potential new job creation from AI advancements or unforeseen technological breakthroughs that could alter the work landscape in one direction or another. But they do indicate the large proportion of occupations that are vulnerable and may undergo significant changes as AI becomes integrated into the labor market.
Our analysis, moreover, includes some worrisome findings when it comes to gender and income equality. For example, in both the United States and Mexico, women have a significant presence in office and administrative jobs. This makes them more likely to be affected by AI. In both countries, AI is also expected to disproportionately affect lower- and middle-income workers.
Our new GENOE index provides significant advantages over other indexes aimed at predicting the impact of AI on occupations. Unlike many other indexes that disaggregate occupations into collections of separate tasks, skills, or abilities, our approach more realistically incorporates their interrelations. Journalists and technical writers, for example, share automatable routine tasks related to writing, editing and generating content. But journalism also includes non-routine roles like storytelling, evaluating the significance of news, and investigative reporting. These are all skills that require human judgment, critical thinking, and persuasion. They make journalism less likely to be transformed by AI than technical writing, a distinction our index makes.
Our index similarly takes into account the ethical, regulatory, and social context of jobs. Judges, for example, must weigh ethical and social considerations in their verdicts with judgment and empathy. For that reason, they have jobs that are less replaceable by automation than credit analysts working on loan approvals or credit ratings, though both make decisions that can affect people’s well-being and are based on evaluating extensive information. This too is a reality our index reflects.
Future research could expand on our work by incorporating elements, such as the potential for job creation and transformation through AI and the interaction between AI exposure and other economic and social factors. The GENOE index could be regularly updated to reflect a rapidly changing AI landscape.
The Policy Implications of Measuring AI’s Effect on Employment
In the meantime, the new index can help policymakers develop targeted interventions and support measures for workers in highly exposed occupations, including through education and training policies, unemployment insurance programs, and economic development strategies. It can assist businesses in making strategic decisions in areas related to workforce development and technology integration and provide individual workers with valuable insights for career planning and skills acquisition. AI is very likely to have an immense effect on employment over the next few years and in what remains of the 21st century, creating new challenges and new opportunities. As that happens, the holistic, forward-looking perspective of the GENOE index should provide crucial insights for policymakers intent on safeguarding the employment and welfare of people throughout society.
Leave a Reply