Despite High Adoption, AI Chatbots Don't Boost Worker Wages or Hours

A new study examining the real-world effects of generative artificial intelligence (AI) chatbots, such as ChatGPT, has found that these rapidly adopted technologies have so far had a minimal impact on workers' wages and hours. The research, conducted by economists Anders Humlum from the University of Chicago's Booth School of Business and Emilie Vestergaard, a PhD student at the University of Copenhagen, analyzed a significant dataset from Denmark and suggests that despite high adoption rates and considerable hype, the anticipated widespread labor market transformations have yet to materialize.[1][2][3] This challenges some narratives predicting imminent, sweeping changes to employment and earnings due to AI.[3]

The Danish study focused on 11 occupations considered highly exposed to AI, including roles like accountants, customer support specialists, software developers, and journalists, encompassing approximately 25,000 workers across 7,000 workplaces.[1][2] Data was collected in late 2023 and early 2024 through large-scale adoption surveys linked to matched employer-employee administrative labor market data.[1][3][4] The researchers employed difference-in-differences analysis and used employer policies as quasi-experimental variation to estimate the impact of AI chatbots.[3][4] Their findings were stark: "AI chatbots have had no significant impact on earnings or recorded hours in any occupation," the authors stated, with confidence intervals ruling out effects larger than 1%.[1][3][4] This held true even when examining overall employment and wage bills at the enterprise level.[5] A direct survey question to workers about whether AI chatbots had affected their labor earnings as of November 2024 further confirmed that workers perceived no earnings impact.[3]

While the study indicated a lack of significant change in wages or total working hours, it did explore the nuances of AI adoption within workplaces.[1][3] It found that firm-led investments in AI, such as providing training and encouraging the use of these tools, substantially boosted adoption rates.[5][3] For instance, such company initiatives nearly doubled the adoption rate from 47% to 83% in the studied cohort.[5] Despite this widespread use – with about half the workers in affected job categories using AI – the reported time savings were modest.[6] On average, users reported that AI chatbots saved them just 2.8% of their work time, which translates to a little over an hour in a standard 40-hour workweek.[5][6][3] Other sources mention a slightly higher figure of 3% time savings.[4] This contrasts with some controlled experiments or specific use-case studies, such as one by Brynjolfsson, Li, and Raymond in 2023 focusing on customer support agents, which found productivity increases nearing 14%, and even higher for less experienced workers.[6][3] The Danish study suggests this discrepancy might be due to effect heterogeneity, meaning that while some specific tasks or occupations might see large gains, the broader average across many exposed occupations is more modest.[3]

The research also sheds light on how any time saved is being utilized, or offset. It found that for 8.4% of workers, AI chatbots have created new job tasks.[5][1][2][3] These new tasks include activities like editing AI-generated output, teachers monitoring the integrity of AI use by students, or specialists focusing on improving queries for chatbots.[5][3] In some cases, these additional responsibilities can even increase the overall workload, diluting the productivity boost from time saved on other tasks.[5][6] The study noted that 17% of users reported new workloads due to AI chatbots, with some performing more of their existing tasks and others taking on entirely new types of work.[3] Furthermore, the researchers estimated that only a small fraction, between 3% and 7%, of any productivity gains achieved through AI use translated into higher earnings for workers.[5][3] This weak wage pass-through, combined with the modest average time savings, helps explain the limited impact on overall labor market earnings observed so far.[3]

The authors of the Danish study suggest several reasons for these muted economic impacts despite high adoption. One key factor is that many tasks are not fully automatable by current generative AI.[2] Humlum noted, "Most tasks do not fall into that category where ChatGPT can just automate everything."[2] The current phase appears to be one where AI is a tool that assists with parts of tasks rather than a complete replacement for human labor in most roles.[2][7] The study indicates that the labor market might be in a transitional period, where companies and workers are still learning how to best integrate and leverage these technologies effectively.[5][2] The authors emphasize that "labor market rigidity delays the economic impact of AI, and success depends on how companies integrate these technologies."[5] This aligns with broader research suggesting that the true impact of AI will depend on factors like redesigning workflows and upskilling the workforce.[6][8][9] While some predict significant long-term productivity gains from AI, potentially adding trillions to the global economy[10][11], the current evidence from this comprehensive study suggests that, for now, the transformative effects on wages and working hours remain largely in the realm of expectation rather than an observed reality.[1][3] The findings also raise questions for the tech industry regarding the substantial capital expenditures on AI infrastructure when widespread economic benefits are not yet clearly evident.[1][2]