AI services like OpenAI’s ChatGPT have been pitched for potential productivity gains, but their effect has been to make people work more while benefiting less from their labor.

In a research paper titled “AI and the Extended Workday: Productivity, Contracting Efficiency, and Distribution of Rents,” economists Wei Jiang (Emory), Junyoung Park (Auburn), Rachel (Jiqiu) Xiao (Fordham), and Shen Zhang (Seton Hall) set out to examine how occupational exposure to AI affects work time.

“When…ChatGPT came along, we were all very mesmerized by how powerful it is, how much work it does,” said Wei Jiang, professor of finance at Emory University, in a phone interview with The Register. “So we, like other people, anticipated if AI is doing our work, we can work less. And I just find myself actually working longer. So I checked with a few friends, and every one of them says, ‘Hey, we’re actually working longer.'”

Jiang said she wanted to see whether the experience that she shared with colleagues reflected a broader trend, so she and her co-authors decided to analyze data from the American Time Use Survey (ATUS, [PDF]), an annual survey conducted by the US Bureau of Labor Statistics that tracks how Americans spend their time, from 2004 to 2023.

“People fill in fifteen-minute interval [summaries] about what they did in the previous day,” explained Jiang. “So you can actually trace out how much people work, and then we also link each occupation to their exposure to AI. So what we find is that higher exposure, if your job is more highly exposed to AI, you actually work longer.”

As Jiang and her colleagues observe in their paper, the arrival of ChatGPT in 2022 offers an opportunity to examine the labor impact of the technology.

Workers in occupations with higher exposure to generative AI experienced a significant increase in work hours and a decrease in leisure time following the introduction of ChatGPT

“Workers in occupations with higher exposure to generative AI experienced a significant increase in work hours and a decrease in leisure time following the introduction of ChatGPT,” the paper says.

Based on data from 2022 and 2023 (when ChatGPT saw widespread use), an interquartile increase in generative AI exposure corresponded to an additional 3.15 hours of work and a reduction of 3.20 hours in leisure per week.

Jiang takes it as a given that AI increases productivity, although there have been several studies suggesting the opposite.

“I don’t think there’s much controversy to say that AI brings productivity gains,” she said. But the problem is, workers may not be reaping the benefits. “I think the disagreement is about who enjoys the gains.”

Three groups potentially could benefit from greater productivity, Jiang explained: Organizations and their shareholders, employees, and consumers. 

In competitive labor markets, workers have less bargaining power to capture technology-enabled productivity gains, she said. And where the product market is competitive, much of the economic benefit gets passed on to consumers.

In other words, to the extent that AI enhances productivity, that extra value tends to be captured by employers or consumers rather than workers.

“I think generally people tend to agree that the AI technology seems most gains are going to the consumers and the firms, and not much to the large population of workers,” said Jiang.

The longer hours, the authors say, also follow from AI’s role in surveilling workers to enforce greater productivity. “Remote workers with greater exposure to AI surveillance technologies work longer hours post-pandemic,” the paper says, noting this effect is not observed among the self-employed.

Despite higher wages being associated with AI occupational exposure, the authors found that AI-exposed workers were less satisfied, based on employee surveys.

“The combined results suggest that while AI-driven productivity gains promise greater efficiency, they have resulted in longer working hours and lower employee satisfaction, especially in competitive markets and for occupations with higher AI complementarity, challenging the conventional expectation that technology frees humans from prolonged workdays,” the paper concludes. ®