Eve here. This post shows careful negative findings about AI implementation in Germany. A careful reading will suggest that a critical part of the research population lies in related settings such as factories and logistics management.
I thought that narrow non-LLM applications in AI could be very productive. However, the US is full of stories, and even analyses of white-collar workers experience a decline in productivity thanks to AI, as they have to oversee and correct its errors. A 2024 survey found that a whopping 77% reported that AI had lowered its output. Of course, many facilities are less cautious and willing to take risks as they have to eat AI responsibility, pay the costs to fix errors, and show unchecked AI results.
At Adionion, German survey respondents have not seen an increase in employment safety concerns as a result of the adoption of AI. The authors of the following report point out that this response may be due to differences in strong labour protections in Germany. This isn’t the case in the US where “hiring of will” and where the media regularly lurks about how it comes to your job. ADP survey earlier this year found that over 30% of US employees were worried about the loss of Liveliofo, created by AI. This anxiety, if not long on average, should have a mental health impact.
Bicci Juntera, Associate Professor at the University of Economics at Pittsburgh, Luca Stella, and Associate Professor at the University of Economics at Milan. Originally published on Voxeu
Artificial intelligence can reduce the need for physically demanding tasks, but it can also invade job satisfaction, the cognitive load of intelligence, and amplify anxiety. This column reports a study on German research data. This finds no evidence that AI exposure hurts workers’ mental health or subjective well-being. However, looking at the independent use of AI tools in the workplace, there are signs that reduce life and job satisfaction. This suggests the need to banish current debate beyond the impact of AI on employment, productivity and wages. If you are transforming your job in ways that affect stress, autonomy, purpose, or health, your Muston must be at the heart of technology policy and labor regulations.
As governments and businesses compete to integrate artificial intelligence (AI) into the workplace, key policy issues have emerged.
While much of the academic and policy debate focuses on the impact of employment and productivity (e.g. Acemoglu etal. 2022, Brynjolfsson etal. 2025), concerns have risen about the quality of the work itself. AI can reduce the need for physically demanding tasks, but it can also invade job satisfaction, cognitive loads of intelligence, and amplify anxiety. These risks have increased with public discourse centres and recent policy proposals, including those that include AI governance and workers’ protection.
Recent contributions from Voxeu have examined how AI is reshaping macroeconomic productivity (Cerutti etal. 2025), changes in employment structure (Ildetzki and Jain 2023), and ways to create tensions between innovation and regulation (Frey etal. 2025). What is less understood, however, is how AI influences workers’ daily experiences – their health, job satisfaction, and psychological well-being.
Recent research (Giuntella etal. 2025) attempts to address this gap using a wealth of German longitudinal research data. Our central findings are carefully optimistic. So far, there is no evidence that AI exposure has occurred that has deprived you of world mental health or subjective well-being. In fact, slight improvements in self-reported physical health and health satisfaction have been observed.
However, this picture changes when you look at the use of AI tools self-healing at work. Here we discover that life and job satisfaction is conservative, yet construed, and suggest that the way we experience the world is important when it is automated.
From unemployment to quality of work: the lens spreads
Existing research documents the economic impact of AI from labor market polarization to changes in sector productivity. Felten et al. (2021) and Bonfiglioli et al. (2025) shows that AI adoption varies widely by occupation and region. Acemoglu et al. (2022) and Brynjolfsson et al. (2025) shows that AI can replace certain roles while creating other roles. However, Gihleb et al. (2022) and Nazareno and Schiff (2021) argue that technological change also has deep implications for physical and mental health.
AI can be different from past automation waves. Industrial robots are replaced by manual routine tasks, but AI targets cognitive and community functions. It can increase productivity and reduce boring tasks, but it can also impair workplace autonomy or increase cognitive demands. How these changes will affect global welfare is a question that will attract more attention.
Measurement of AI exposure
We use data from the German Socioeconomic Panel (SOEP), a longitudinal representative dataset that tracks workers’ health, satisfaction and job characteristics over 20 years (2000-20). Germany offers individually passive cases. There is a gradual acceleration rate of vocational training systems, robust labor protection, and AI adoption.
To support the impact that AI has on global well-being and health, we employ two complementary measures of AI exposure. Our main measure developed by Webb (2019) quantifies how honest the acceptability of a profession to AI is based on overlapping job tasks and AI-related patents.
Our quadratic scale is a self-reported exposure measure from Sato Nami in 2020, where workers were asked how often they use AI-related systems at work (including natural language processing, image recognition, information assessment). By limiting the sample to individuals who surrounded the labour market before 2010 (before the adoption of wide prelides of AI in Germany), we limit the potential bias of sewing to occupations based on AI exposure.
Evidence regarding AI adoption and WorldRS happiness
Our findings diverge depending on how AI exposure is measured. Using a task-based exposure scale (Webb 2019), you can find out: (1) There is no significant change in life or job satisfaction among workers exposed to AI. (2) There is no significant increase in financial insecurity or reported job instability. (3) Small but signer improvisation in voluntary health and health satisfaction.
The ASE results are summarized in Figure 1. In particular, increased physical health is distinguished from a simultaneous reduction in the physical burden of work for workers exposed to A.
Figure 1. The impact of AI’s impact on the world and health: the AI’s Webb scale
Also, we can see that the slite is decreasing overall with weekly working hours (approximately 30 minutes). These results suggest that AI may reduce physical burden without total employment stability, at least in the early stages of adoption.
However, when considering self-healing exposure to AI systems, the photos change. Workers who reported using AI tools at least weekly at work were more likely to report reduced life and job satisfaction. You will have a negative effect. The results of this analysis are shown in Figure 2.
Figure 2 Welfare and health of the impact of AI on the world: a SOEP-based AI scale
This divergence shows important insights. It is possible that how workers interact with AI tools is more important to their well-being than how their thumb-ups were “objectively” exposed to AI. This inference is consistent with the arguments made in Vries and Erken (2023) that recognition and adaptation shape the productivity potential of AI, suggesting that communication and workplace design are essential for successful AI integration.
Early stage notes
Several caveats should be considered when interpreting the findings. First, data will only be expanded until 2020. This is because the development of the generated AI (such as large-scale language models) played regenerated the potential reach of AI into a creative domain in white colour.
Second, our sample focuses on middle-aged and elderly workers who surrounded the labour market with their environment before AI spread begins. Young cohorts can experience AI adoption very differently – especially when shaping initial job experiences and career trappings.
Third, Germany’s powerful labor market Instat may have eased AI’s disruptive efforts. Therefore, our Findins cannot generalize to countries with more flexible workforce discoveries.
Conclusion and the meaning of poly
The AI transition is ongoing, but its long-term impact remains uncertain. Early evidence from Germany suggests that AI can be integrated into the workplace with the well-being of harmful workers, even reducing physical work intelligence. However, subjective experience is important. If WorldRS is overwhelmed, desks and surveillance, the psychological costs of AI can emerge well before the economy.
As the policy agenda of global AI evolves, from privacy and competition to skills and taxation – labor and happiness should not follow. Like the previous Industrial Revolution, it is not the technology itself, but the way it was implemented, the government, the experienced way it shapes its legacy.
What should policymakers take away from their findings?
First, our findings underscore the need to commission conversations beyond employment and wages. When AI transforms work in ways that impress stress, autonomy, purpose, or health before it reaches the heart of technology policy and labor regulations. As Martin and Howert (2022) pointed out, quality of work includes not only the enemy, but also working hours, safety and welfare.
Second, evidence from Germany supports the idea of the Institute of Materials. Work Council, joint decisions and employment protection could help Germany integrate AI more smoothly with fewer psychological costs for workers. Countries without such institutions may need to explore alternative safeguards such as sloug regulations, collective burgers, or Echic design standards.
Third, while the fear of mass unemployment may be exaggerated, concerns about the quality of degraded work are realistic and already observable. In this sense, poly, which focuses on reskilling and job matching, may overlook the broader human impact of AI integration.
Finally, we need better data. Differences between objective and self-reported exposures indicate the need for a richer task-level investigation and real-time indicators of AI use and worker outcomes.
See original bibliographic submission
