Key Takeaways
- Despite fears, AI chatbots like ChatGPT have shown virtually no significant impact on jobs or wages so far, according to a recent study.
- Researchers analyzed data from 25,000 workers across 11 occupations often considered vulnerable to AI in Denmark.
- While many workers are adopting these tools, especially with employer encouragement, the expected economic benefits haven’t materialized yet.
- AI use is saving workers only a small amount of time (about 2.8% of work hours) and can even create new tasks, offsetting some gains.
- The study suggests transformative effects of AI on the economy haven’t arrived, questioning the massive investments being made in the technology.
Worries about generative AI chatbots like ChatGPT taking jobs or driving down wages appear unfounded for now. A new study analyzing their real-world impact found surprisingly little effect on the labor market.
Economists Anders Humlum and Emilie Vestergaard examined the effects of these AI tools on 11 different job types in Denmark during 2023 and 2024. Their working paper looked at roles often flagged as susceptible to AI disruption, including accountants, IT support, journalists, lawyers, software developers, and teachers.
Covering 25,000 workers and 7,000 workplaces, the researchers found minimal changes in employment levels or pay related to chatbot use. This contradicts much of the hype surrounding AI’s potential economic power.
The findings raise questions about the billions being poured into AI development and infrastructure by tech companies. Some firms, like Microsoft and Amazon, are reportedly rethinking their spending as widespread business adoption lags.
It’s not that people aren’t using these tools. “The adoption of these chatbots has been remarkably fast,” Humlum told The Register. He noted that most workers in exposed fields now use them, often encouraged by employers.
However, this adoption isn’t translating into major economic shifts. The study found that while company efforts to promote AI did increase usage, the impact on work quality and satisfaction was mixed.
Interestingly, the economists discovered that AI chatbots have actually created new tasks for about 8.4% of workers, sometimes even for those who don’t use the tools directly. This new work can cancel out potential time savings.
Humlum gave an example: teachers now spending time checking if students used ChatGPT for homework. Others spend time refining AI prompts or reviewing the quality of AI-generated content.
These new tasks could potentially lead to higher demand for workers if they involve high-value skills, but overall, the time saved by using AI was much lower than expected. The study reported average time savings of just 2.8% of work hours – a little over an hour in a standard work week.
This contrasts sharply with some earlier studies suggesting productivity boosts of around 15%. Humlum suggests those studies might have focused on specific tasks highly suited to AI, whereas his research used real-world data across broader job functions.
“Most tasks do not fall into that category where ChatGPT can just automate everything,” Humlum explained, adding that companies are still figuring out how best to use these tools effectively.
Furthermore, even when productivity gains occur, only a small fraction (estimated between 3% and 7%) seems to translate into higher earnings for workers. Much of the benefit might go to the companies, or the time saved might not easily convert into more valuable output.
While company training can boost AI adoption, especially among women who currently use the tools less than men, it doesn’t yet promise significant economic returns based on hard data like earnings and wages.
Humlum concluded that claims of AI being highly transformative need to reckon with the fact that, two years in, these tools haven’t significantly altered key economic outcomes for workers.
