AFTER DECADES of blue-collar jobs being snatched up by machines, advanced chatbots are now breathing down white collars. “Generative” artificial-intelligence (AI) tools, such as ChatGPT, have made significant progress in crafting human-sounding language and grasping context. So much so that they have leapfrogged humans in some tasks. This could make as many as 300m jobs redundant globally, according to Goldman Sachs, a bank. Several new papers consider which sectors will face the biggest shake-up (see chart).
View Full Image
A recent study by OpenAI, the startup that created ChatGPT, looked at the potential for automation across 1,016 occupations. Humans and AI separately rated how well software powered by large language models (LLMs), which are trained on vast chunks of the internet and then fine-tuned to specific functions, could undertake 19,000 tasks involved in the jobs. If the software, such as OpenAI’s GPT-4, was deemed able to reduce the time it takes humans to complete the task by at least half, without a drop in quality, the task was considered ripe for AI replacement (a score of one meant that the whole occupation could be done in half the time).
For other tasks the annotators imagined additional software that could be added to the model, such as computer tools that can automatically pull fresh data from the internet. They found that 80% of Americans could have at least 10% of their work tasks done by advanced AI tools. The figure rises to 50% of tasks for an estimated 19% of workers.
Most exposed are industries which rely on programming and writing skills. That echoes another study, published on March 1st by academics in America, which found that the industries most at risk of a shake-up were legal services and some areas of the financial and insurance industries. They point to telemarketers as the occupation most likely to be made redundant.
Teachers, especially those of languages, literature and history, are next on the list. What is striking in both studies is that, unlike past breakthroughs in machine learning, it is skilled and high-paying jobs that are most exposed.
This automation should not be feared. It could unshackle workers from mundane tasks and unleash greater labour productivity, which would be a boon for drum-tight labour markets in advanced economies. A study by Goldman Sachs, published on April 5th, suggests that generative AI could grow global GDP by 7% in the next decade.
But studies like this may overstate the potential for automation. The annotators responsible for mapping overlaps between LLMs and human capabilities may omit some tacit skills in professions they know less about. Human qualities essential for some jobs, such as empathy or charisma, will be overlooked. And not all tasks capable of being carried out by AI should be: Vanderbilt University in Tennessee had to apologise for using ChatGPT to write a condolence email to students after a shooting at another American university.
Many businesses may also lack the IT architecture or inclination to accommodate AI innovations. And those who embrace it will face practical and legal quandaries. When chatbots do not know what to say, they often fib (though so can humans). The “creative” output they generate is based on a mashup of data sourced from the internet, raising thorny issues around accuracy, privacy and intellectual property. For all their conversational panache, in the real world, AI tools will still need handlers. That may even end up creating new jobs.
© 2025, The Economist Newspaper Limited. All rights reserved. From The Economist, published under licence. The original content can be found on www.economist.com