Artificial intelligence isn’t hurting workers—It might be helping

But the authors caution: It is way too soon to draw definitive conclusions.

The paper, “Artificial Intelligence and the Wellbeing of Workers,” published June 23 in Nature: Scientific Reports, uses two decades of longitudinal data from the German Socio-Economic Panel. Using that rich data, the researchers — Osea Giuntella of the University of Pittsburgh and the National Bureau of Economic Research (NBER), Luca Stella of the University of Milan and the Berlin School of Economics, and Johannes King of the German Ministry of Finance — explored how workers in AI-exposed occupations have fared in contrast to workers in less-exposed roles.

“Public anxiety about AI is real, but the worst-case scenarios are not inevitable,” said Professor Stella, who is also affiliated with independent European bodies the Center for Economic Studies (CESifo) and the Institute for Labor Economics (IZA). “So far, we find little evidence that AI adoption has undermined workers’ well-being on average. If anything, physical health seems to have slightly improved, likely due to declining job physical intensity and overall job risk in some of the AI-exposed occupations.”

Yet the study also highlights reasons for caution.

The analysis relies primarily on a task-based measure of AI exposure — considered more objective — but alternative estimates based on self-reported exposure reveal small negative effects on job and life satisfaction. In addition, the sample excludes younger workers and only covers the early phases of AI diffusion in Germany.

“We may simply be too early in the AI adoption curve to observe its full effects,” Stella emphasized. “AI’s impact could evolve dramatically as technologies advance, penetrate more sectors, and alter work at a deeper level.”

Key findings from the study include:

• No significant average effects of AI exposure on job satisfaction, life satisfaction, or mental health.
• Small improvements in self-rated physical health and health satisfaction, especially among lower-educated workers.
• Evidence of reduced physical job intensity, suggesting that AI may alleviate physically demanding tasks.
• A modest decline in weekly working hours, without significant changes in income or employment rates.
• Self-reported AI exposure suggests small but negative effects on subjective well-being, reinforcing the need for more granular future research.

Due to the data supply, the study focuses on Germany — a country with strong labor protections and a gradual pace of AI adoption. The co-authors noted that outcomes may differ in more flexible labor markets or among younger cohorts entering increasingly AI-saturated workplaces.

“This research is an early snapshot, not the final word,” said Pitt’s Giuntella, who previously conducted significant research into the effect of robotics on households and labor, and on types of workers. “As AI adoption accelerates, continued monitoring of its broader impacts on work and health is essential. Technology alone doesn’t determine outcomes — institutions and policies will decide whether AI enhances or erodes the conditions of work.”