ChatGPT has made generative AI easy for many people to try, and businesses are racing to find ways to use it. Its capacity to draft text, create graphics, and analyze information could reshape work. Whether that change brings broader prosperity or widens economic divides is still an open question.
A technology with competing economic outcomes
Recent generative AI systems are built on large language models trained on extensive data. ChatGPT, which is based on GPT-3.5, made the technology accessible through a simple chat interface. Users can ask it for things such as recipes, workout plans, and computer code, without needing coding skills.
Companies and investors see commercial opportunity. Microsoft has invested a reported $10 billion in OpenAI, while Salesforce said it would bring a ChatGPT app to Slack and announced a $250 million fund for generative AI startups. Google said it planned to use generative AI tools in Gmail, Docs, and other products. Yet the article notes that obvious killer apps had not emerged.
The larger stakes go beyond which businesses profit. Productivity growth has been weak since around 2005 in the US and most advanced economies, and wage stagnation has affected many people. Earlier digital advances brought substantial wealth to some investors and entrepreneurs, but widespread gains have been limited, and some workers lost jobs to automation.
That record makes the new technology’s promise harder to assess. It could raise productivity and help workers do more. It could also enable companies to automate tasks once considered difficult to replace, while directing the gains toward a small group of firms and workers.
How AI could change white-collar work
One analysis by OpenAI’s Tyna Eloundou, Sam Manning, and Pamela Mishkin, with the University of Pennsylvania’s Daniel Rock, estimated that large language models could affect 80% of the US workforce. They estimated that 19% of jobs could be heavily affected, with at least 50% of tasks in those jobs “exposed.” The analysis suggested that higher-income jobs may face more impact than in earlier waves of automation.
Writers, web and digital designers, financial quantitative analysts, and blockchain engineers were among the workers identified as potentially vulnerable. The change matters because earlier automation often targeted routine, step-by-step work. Generative AI can also produce writing and graphics, tasks many people had viewed as requiring human creativity and reasoning.
MIT labor economist David Autor argues that this opens up a wider range of tasks to computerization. He has studied how digital technologies displaced manufacturing and routine clerical work. His concern now is that companies could replace well-paid white-collar roles, pushing some workers toward lower-paying service jobs while those best positioned to use AI capture the benefits.
That outcome is not inevitable. AI tools could also help people with less experience or formal expertise build capabilities and compete for work that requires specialized knowledge. In that version, generative AI becomes a way to broaden access to knowledge work and help workers develop skills for fields such as health care or teaching.
Early evidence points to a possible skills boost
An experiment by MIT economics graduate students Shakked Noy and Whitney Zhang offers a preliminary sign of how AI might affect differences in performance. They studied hundreds of college-educated professionals in areas such as marketing and HR. Half used ChatGPT for daily tasks, while the others did not.
ChatGPT increased overall productivity in the experiment. The least skilled and accomplished workers improved the most, narrowing the performance gap between employees. Workers who were less effective writers improved substantially, while strong writers became somewhat faster.
The findings suggest that AI could help some workers develop skills they lack. Autor describes experienced people displaced from office and manufacturing work as “lying fallow.” If generative AI helps them gain expertise needed in areas with available jobs, it could support a stronger workforce.
But the experiment does not settle how AI will affect employment or the economy as a whole. It points to a potential benefit, while employers’ choices will shape who can use the tools and who benefits from the productivity gains.
Design choices will shape who benefits
Stanford economist Erik Brynjolfsson has warned that AI development can focus too heavily on copying human abilities. In his view, that approach risks producing systems that replace workers, put downward pressure on wages, and concentrate wealth. He sees ChatGPT as an example that has intensified debate about whether AI should substitute for people or give them new capabilities.
Brynjolfsson is also optimistic about the technology’s potential. He argues that businesses can use generative AI to expand what they offer and help workers be more productive. The difference lies in whether companies treat AI as a tool for new work and creativity, or mainly as a way to remove the need for workers.
History offers a reason to be patient about measuring economic effects. MIT economist Robert Solow observed in 1987, “You can see the computer age everywhere except in the productivity statistics.” The impact of computing became clearer later, as businesses found ways to use cheaper computational power and improved software.
Economist Avi Goldfarb says AI’s lasting contribution will depend on whether companies redesign processes and create new value for customers. Using AI to perform existing tasks a little better may bring only modest benefits. Larger gains could come when businesses discover new ways to work with writing or graphic design at scale.
The technology is still developing, and its overall economic effect remains uncertain. The choices made by companies and developers will help determine whether generative AI expands workers’ abilities and shares prosperity more widely, or reinforces existing advantages.