Higher earners face greater AI exposure, study finds

Stay informed with free updates

Higher earners such as software engineers and data scientists are more exposed to the impact of artificial intelligence than lower paid workers, according to the latest research on the technology and the labour market.

Almost a fifth of employees would see at least half their tasks potentially affected by advances in machine learning, according to an analysis of more than 900 occupations published in Science on Thursday.

The paper highlights the uncertainty over AI’s impact on the labour market. The IMF expressed “profound concerns” this week that generative AI could stoke inequality and disrupt work, including in highly skilled industries.

“Exposure (to AI) can be good for workers, or it can be bad for workers,” said Daniel Rock, a co-author of the paper and assistant professor of operations, information and decisions at the University of Pennsylvania.

“At this stage, it’s very difficult for us to say what the long-term effects on labour demand will be,” he added. “But the exposure measure does tell you where to look for where things might change.”

Rock and his fellow authors from ChatGPT maker OpenAI and the Centre for the Governance of AI, a UK non-profit organisation, examined 923 occupations from a database of work and worker characteristics.

They used humans and a trained GPT-4 large language model to analyse if the fast-evolving technology could cut the time a person would take to complete a task by at least half, with no reduction in quality.

They concluded that 18.5 per cent of workers were in jobs that had 50 per cent or more of their tasks exposed in this way, skewing towards higher paid occupations.

Jobs most affected included blockchain engineers, clinical data managers, public relations specialists and financial quantitative analysts. Occupations with no exposed tasks included motorcycle mechanics, pile driver operators and stonemasons.

“Knowledge workers process information and you can think that what these large language models are doing is turbocharging our ability to process information in different ways,” Rock said.

The research echoes similar findings elsewhere. City of London finance professionals would be most affected by AI applications including image recognition, language modelling, translation and speech recognition, according to a UK government study published in November of AI’s job market impact.

The paper in Science was a significant addition because it estimated the scale of AI’s implications for various jobs, said Sarah Bana, an assistant professor of management science at Chapman university.

The research suggested the technology would have different impact from computerisation, which most affected less well-paid jobs based on routine cognitive tasks.

“There is a lot that needs to happen for these estimates to be realised, as the authors note, but it suggests very profound impacts on how we do what we do,” said Bana, who was not involved in this research but works with Rock on other projects.

While the study provided an “interesting overview of hypothetical future scenarios”, it showed the need for more research on employees’ “needs and concerns”, said Mhairi Aitken, ethics research fellow at the UK’s Alan Turing Institute.

“It’s crucial that studies such as this one are complemented by insights into actual experiences with AI,” Aitken said.