News - 10.04.2026

The impact of artificial intelligence on the labour market 

Artificial intelligence is moving quickly from a specialised technology to something that is beginning to affect every day working life. Employees are already encountering AI tools in administrative systems, recruitment processes, customer service, and professional tasks that were until recently considered uniquely human. For unions like VR, these developments raise important questions: What kinds of jobs and skills are most exposed? Who benefits from higher productivity? And how can workers shape the way these technologies are introduced into workplaces? 

To explore these issues, Victor Karl Magnússon, a specialist at VR and a doctoral student at the University of Oxford, spoke with Dr Eszter Czibor, an economist who examines science, innovation, and technology through a socio-economic lens. She combines an academic background in organisational and behavioural economics with extensive experience advising policy makers and practitioners on innovative support, technology adoption, and digitalisation programmes. Eszter has held senior research roles at UK-based policy organizations, working with partners like the European Commission, the British Council and UNDP. She holds a PhD in Economics from the University of Amsterdam and was a postdoctoral scholar at the University of Chicago. Now based in Reykjavík, Eszter has been researching the local startup and innovation ecosystem in Iceland and teaching a course on the AI economy at the University of Reykjavík, focusing on how artificial intelligence is reshaping work, innovation, and organisational structures. 

People often talk about AI as if it is one monolithic thing. In reality, there are many kinds of technologies which are grouped together under the heading of AI. Among these different kinds of AI, which are likely to matter the most for working life and the labor market? 

You are right, there is so much confusion around what we mean by AI. It doesn’t help that even though the term ‘artificial intelligence’ has been around for 70 years, there is no single definition that everyone agrees on. A common thread in many descriptions is the focus on “intelligent machines” that can simulate human learning or decision making and operate with some level of autonomy. Much of the recent popular interest in AI comes from the fact that we can now easily interact with advanced models through chatbots, using plain language rather than code, to generate new text, images, and more. These are examples of generative artificial intelligence or GenAI: models that can create complex original content based on patterns they learn from their training data. Predictive artificial intelligence systems, on the other hand, tend to focus on solving narrower, more specific classification, optimisation or decision-making problems such as detecting tumours in medical images or predicting flood risk. 

So where might the biggest future impact come from? While some leading researchers argue that purpose-built predictive AI solutions are more likely to be accurate (and are far more energy-efficient), many social scientists see generative AI as a potential new general-purpose technology: like electricity or the internet, it could reshape large parts of our economy and society. 

Much of the discussion about AI is about the future. But AI is already here. Where, concretely, is AI already being applied in the workplace? And how are the effects of narrow, predictive AI and generative AI different in this context? 

Given the excitement around generative AI, it is easy to forget that many of the artificial intelligence systems that are already widely used in industry and organisations today are predictive models, from smart logistics systems and automated bookkeeping to catch forecasting and species identification for fisheries. These applications rarely displace workers, but they often change the nature of their job and their required skills. Sometimes, these changes are welcome: when routine tasks are automated and workers can focus on key activities, or when workers are trained in new skills and achieve better results with technology, job satisfaction increases. But not all impacts have been positive: predictive AI technologies can reduce worker autonomy, increase workplace surveillance, or amplify bias in hiring decisions. Even though generative AI technologies are so young, they are spreading fast in workplaces: in a 2025 survey in the US by the St Louis Fed, nearly 40% of respondents said they have used AI in their workplace. Usage intensity is still low though: less than 6% of respondents’ work hours were spent using generative AI.  Overall statistics mask key differences: younger people and those with higher education are more likely to use AI professionally. Interestingly, there is also a clear gender divide: research after research finds that women use AI at work less than men, even for the same tasks and in similar occupations. 

Ironically, some of the first to experience tangible negative impacts from the arrival of generative AI technologies are the very people whose work has been used, often without consent or compensation, for training the underlying models: recent research has found that the introduction of image-generating AI technologies led to a 17% decrease in the number of freelance job posts related to image creation.  

You say that predictive AI tools can sometimes reduce worker autonomy. What are some examples of that? And if we take autonomy seriously as a normative ideal, what can unions do to make sure that AI is integrated into workflows such that we preserve and perhaps enhance worker autonomy, instead of eroding it? 

A famous example is the automated scheduling system introduced by Starbucks that was meant to optimise employee shifts based on predicted store traffic. Once deployed, it inadvertently created unpredictable work schedules and dramatically varying weekly hours for employees, hurting their financial stability and their autonomy over their time. Similarly, van drivers in the UK experienced a reduction in their autonomy and ability to take the initiative at work after the introduction of a smart logistics system. 

To avoid outcomes like these, unions should advocate for employees’ right to be meaningfully involved in the design and testing of systems that affect their working conditions, including algorithmic management solutions. 

Another point you mentioned is the disparity of use of generative AI along the lines of gender, age and occupation. Are economists worried about new forms of inequality in the labour market, between those who manage to adapt to the new technological reality and those who don’t?  

The optimistic scenario is that AI will democratise access to many forms of knowledge and skills. In certain settings, AI tools have been shown to help less-experienced workers catch up with their more productive peers. But there are also signs of AI amplifying existing inequalities: higher-income people are much more likely to adopt generative AI tools than those in the lowest income brackets, and AI chatbots tend to perform better in response to prompts by higher-educated individuals (or: people with more formal education). And the gender gap we discussed among early users of GenAI chatbots matters for more than just who benefits from productivity gains: it may also shape the direction the technology itself takes. These models are trained on vast amounts of internet data, much of which reflects existing gender biases and stereotypes, and they continue to evolve through user interactions and feedback. When women are underrepresented among users (and among the developers building these systems) GenAI tools will be optimised around male perspectives and experiences. Over time, this risks producing technologies that are less useful for women and that reinforce, rather than reduce, existing gender inequalities at work and beyond. 

John Keynes predicted in 1930 that technological development and automation would lead to a 15-hour work week within a couple of generations. While this did not quite materialize, some people now view the potential effects of AI with similar optimism for shorter working hours. Will this happen? And is the challenge technical or political? 

First, a caveat: artificial intelligence’s impact on jobs will be uneven, with little to no direct impact on many crafts, trades, and roles where human interaction is the focus. That said, AI could well be transformative for information-based or knowledge-centric work, with indirect effects on other parts of the labour market. 

To make a 15-hour-workweek future happen we would essentially need three things. First, AI solutions that reliably automate routine tasks and improve workers’ productivity on non-routine, expert tasks, making human labour more, not less, valuable. Second, extensive upskilling and reskilling programmes so workers can leave jobs that mainly consist of routine tasks and move to higher-value roles (some of which will be new types of jobs emerging in response to the AI-led transformation). Third, collective bargaining agreements that allow workers to benefit from their higher productivity in the form of shorter work hours. And you are absolutely right: whether any of this happens is as much a social and political choice as it is a function of technological progress.  

The struggle for a shorter workweek is a complex and long-term project. What are simpler things unions like VR do to prepare their members for a future in which AI permeates the workplace? 

I’d like to double down here on my previous point about the importance of training and skill building. We have good evidence from the UK, for example, that job ads increasingly feature AI skill requirements, and that employers value practical AI skills listed in job applications. The UK government is responding to these signals: they made a commitment to upskill ten million workers by 2030 and have already delivered a million AI courses since last summer. The Icelandic AI Action Plan also includes reskilling plans related to AI and data literacy, and I believe there is plenty of room here for unions and particularly VR to play a proactive role.  

Young workers enter the labour market at a time of uncertainty. How should they think about education and career choices in the light of AI? 

Young people are definitely on the front lines of AI-driven job transformation: there are some signs from the US and UK that vacancies and hiring for junior positions in AI-exposed roles might be falling faster than for other roles. I think this leaves three options for young people. They could choose a profession with less exposure to AI, such as skilled trades and crafts. Alternatively, they could focus on building subject-matter expertise in a chosen field, coupled with practical AI skills, so they can scale their knowledge with AI tools. Finally, a great option is to lean more into our uniquely human advantages. Social skills, empathy, initiative, critical thinking and judgement, as well as relationships and networks, are becoming more important than ever. 

If we look ahead ten years, it is clear AI will be part of working life. To conclude, what worries you about that future? And what excites you? 

I am very excited about the new types of jobs, discoveries and innovations that could be created in a future with powerful artificial intelligence. I do worry about the path to get there: many roles are likely to be transformed and some will disappear, so we need to start thinking proactively how we support workers through these transitions. And to be honest, I am also somewhat worried that AI’s ultimate productivity impact will fall short of what we hope for: if we are indeed in an “AI bubble” now as some warn, then we must be prepared for some painful adjustments when it bursts.