Is AI making work safer?

Mary McCaughey (00:00:32:00 – 00:03:24:00): Hello and welcome to Eurofound Talks. Eurofound talks AI today. We've done it before. We will no doubt do it again, and AI is very much the subject of the moment, as we all know. But today we're taking a slightly different angle. We've talked about AI previously and when it comes to working life, we've talked about it in terms of whether it's going to actually destroy jobs or whether we will see a creation of new jobs, but this time we really want to look at what is AI doing to the worker. Is it actually making our jobs safer or are there dangers emerging that we need to be conscious of as we move into a future which embraces AI?
This is, I suppose, one of the most complex questions that we have to look at, because we know that AI can drastically reduce physical strain, it can take over hazardous tasks. It can take over repetitive tasks, and it does lead to fewer muscular skeletal disorders. That's a good thing. But of course it's introducing a whole new complex web of what we call OSH, which is Occupational Safety and Health challenges. These have not always been apparent at the start, but some are emerging and they are clearly going to be the issues that we will have to tackle as we go down the road. 

In this episode, we hope to delve into many of these issues and see how this digital revolution is actually impacting work across the critical dimensions: physical, psychosocial and organisational. These are all interconnected and as we know ourselves in Europe and in our most recent European Working Conditions Survey, which looks at nearly 40,000 people across the European Union and beyond, and their working life. We really can see how these physical, psychosocial and organisational dimensions feed into good quality jobs. 
We want to explore how the critical role of worker involvement and job autonomy to other areas that we've looked at before, whether AI systems empower them or constrain employees, when they're monitoring performance or dictating workflows. 
So these are complex issues and we want to look at this in the broader economic and political context. Clearly over the last little while, there have been nothing but developments in this area. We've seen a desire to drive forward on the competitiveness compass. 
And we've seen our President von der Leyen from the Commission outline massive investment in AI into the future with a view to boosting Europe's competitiveness against other global trading partners. So, this is going to be an interesting while that we have in front of us.

And today I'm particularly excited, because we're going to be joined by one of our sister agencies, EU-OSHA. EU-OSHA is a is a sister agency based in Bilbao, whose speciality is the area of expertise of Occupational Safety and Health. And today we're joined by two of our experts, one from EU-OSHA and one from Eurofound. Ioannis Anyfantis , who's the project manager at EU-OSHA, and Dragos Adascalitei, who's the research officer here in Eurofound and you will have heard him on other occasions talking about many other topics on our podcasts and webinars. So you're very welcome both. 

Dragoș Adăscăliței (00:03:24:00 – 00:03:29:00): Hello, Mary.

Ioannis Anyfantis (00:03:29:00 – 00:03:31:00): Hello, Mary. Very good to be here.

Mary McCaughey (00:03:31:00 – 00:04:07:00): So good to have you here today and I'm just going to start off to really just capture the zeitgeist. Dragos, I'll turn to you and we want to look at AI on the labour market and in the workplace. But like just in the broader context, I mentioned that Ursula von der Leyen had said that AI is essential for our future independence. She spoke about this massive investment that is going to go into AI gigafactories. Is this really a significant element of Europe's economic development? Is this playing out for the future competitiveness of Europe?

Dragoș Adăscăliței (00:04:07:00 – 00:06:30:00): Hello, Mary, and thanks for the invitation. I would say that if you look left and right and try to look at what is happening around AI in terms of both research and policy action, you can see that we're in the in the midst of a global race for artificial intelligence. And that is because emerging evidence suggests that, indeed, artificial intelligence holds the promise of massive competitiveness gains, of massive improvements in efficiency, but also of moving our societies and our economies to the next phase of development. So of course, look at it from this perspective: Artificial intelligence is a huge potential amplifier of both economic gains and broader societal gains. And you can see this kind of line of thought in the policy proposal that has come up from the Commission in the past six months or so. So in the beginning of the year, the Commission has put forward the AI action plan which outlines, let's say, the broad context and the broad set of policies that the EU should focus on in the next years or so, and more recently, so to speak, the Commission has put some meat on the bones through the apply AI strategy that comes and tries to actually, let's say, incentivize investment and usage. So not only not only investment, but also usage of AI in European enterprises. So put it all together, there's a lot of interest in the potential gains that the artificial intelligence can bring to our economies, and I would say legitimately so, right, because we've seen from previous waves of industrial and previous waves of investment or technological change and technological revolutions. That technology can actually, you know, move our economies forward, boost productivity, boost efficiency and actually create more jobs, and also contribute to an increase of better jobs. 

Mary McCaughey (00:06:30:00 – 00:07:25:00): OK, OK. So that's a good start. It puts it in that context. And I suppose we are seeing it as a key element to try and pivot us towards greater competitiveness in the current context. So that also gives it an added importance in in the world that we are operating in at the moment, but certainly also with respect to the conversation we're having today, and I would like to just draw a little bit on where do we stand on that because I know that our EWCS, the Working Conditions Survey that we just published recently, the first findings in September that showed for the first time, it took the temperature of workers and how they were consuming AI generative AI. 
For example, in the workplace, can you? Can you give me an outline there of what is the level of adoption across Europe? I mean, are we laggards in particular areas? Do we need to catch up in others? I mean, are Member States and sectors varying greatly? 

Dragoș Adăscăliței (00:07:25:00 – 00:10:09:00): Yeah. So indeed, as you say, the European Working Conditions Survey was fielded in 2024 and we're already in late 2025. Things might have changed since then because one thing that we know about the uptake of general TV AI is that, unlike with previous technologies. The increase in the number of users since its launch in November 2023 has been exponential, but what we know from 2024, based on the European Working Conditions Survey, is that on average 12% of workers are using generative AI technologies at work. Yeah. So, we're not talking about the use of generative AI technologies in private workspaces. Now when we talk about differences across countries, as with everything that that is in Europe, when we compare European countries, there are massive differences. So in general, the update of generative AI tends to be much lower in Eastern and Mediterranean countries and much higher in continental, in Nordic countries, and the same is valid for sectors such as ICT, banking, tend to have very high levels of uptake of generative AI. But also because this type of technology lends itself to the use in this kind of jobs and this type of occupations, and it's much lower in sectors like constructions and agriculture, where this, this specific type of AI technologies is not as useful. But in addition to the data from the European Working Conditions Survey, we also have data from Eurostat, which looks at the use of artificial intelligence in European enterprises and the data show that, on average, around 13.5% of European enterprises use one type of artificial intelligence system with larger enterprises being much more likely actually to use an AI technology. This is also because of the investments required to use and to buy artificial intelligence systems, but also because of the economies of scale. Larger companies tend to benefit more from artificial intelligence technologies, and in that respect, the recent apply AI strategy that the Commission published last week.

Mary McCaughey (00:10:09:00 – 00:10:42:00): Because otherwise we're dealing with potential of a growing kind of digital divide in AI, both across Member States, if I understand you correctly, but also across sectors and currently with our labour shortages in different sectors, that will also play into it, but the 12%. And to pick up or take up that you talked about at the start: There you are saying that because there's been about a year gap there in terms of taking a snapshot of how workers were using AI that could have grown.

Dragoș Adăscăliței (00:10:42:00 – 00:11:23:00): Definitely. And looking at surveys from national sources, some surveys report much higher uptake of the high in the range of 40 to 50%, and that makes a lot of sense because on the one hand, people use artificial intelligence in their day-to-day outside work, but also when they use artificial intelligence in the workplace, they don't necessarily use it because of an organisational policy, they might use this on their own initiative, right? So, there's a lot of, let's say, informal uptake in this technology in the end. And GPTs are general purpose technologies that people and I think are in in a phase of experimentation. If you want. We're at the beginning of the cycle exactly.

Mary McCaughey (00:11:23:00 – 00:12:13:00): Yeah. And perhaps a transitional phase. I mean, one that that leads me on to a question I wanted to ask. I referenced it at the start in the introduction about the fear factor that was at play when AI was introduced, about the loss of jobs. And I know there's a sort of a mantra which goes around at the moment that you're not going to lose your job to AI, but you will lose it to the person who knows how to operate AI, which would speak a little bit to what you were just saying. There was about the people using it themselves and developing themselves on a kind of a personal basis to a large degree also. But are those dire predictions of mass unemployment, are they? Are they rumours that we should abandon at this stage? 

Dragoș Adăscăliței (00:12:13:00 – 00:14:54:00): That's a very good question. And I would say that there are always anxieties in the world of work tech when there's a technological change. It was the same and it was, you know, the advent of the computer in the 1980s. It was the same in the 1990s with the advent of the Internet. And it's the same now. People are anxious about how a new technology is going to impact their work and whether that technology is going to take their jobs right. Ten years ago, it was about robots. Now it's about artificial intelligence.

Just all that I've seen up to this point points actually into the opposite direction. That is, that artificial intelligence is not actually going to destroy massive amounts of jobs, but it's going to reshape the way work is done. So, in the next five years or so, I don't think we're looking at the technological unemployment because of artificial intelligence, but we're looking at significant changes in the way people are carrying out their tasks, the way people are organising their work and the way people are doing in general work within an organisation. So I think this is where the massive impact of artificial intelligence are going to be. That being said, some of the evidence points to specific negative impacts on the number of jobs in particular parts of the labour market, and I'll give you an example. The best example is online labour. Supermarkets where you have lots of freelance workers, people doing job to online platforms, people trying to do fulfil tasks rather than do jobs as bundles of tasks and there in the aftermath of the launch of ChatGPT. We've seen a massive decline in the number of tasks posted by potential buyers because they are not employers, right? And that points to the fact that indeed at the fringes of the labour market we might see a negative effect. Of generative AI in terms of the number of jobs and one particular occupation seems to be very much negatively impacted and that is the job of translators, right? The number of translator jobs have has decreased massively in the in the aftermath of the launch of ChatGPT, and continues to do so because this technology tends to automate very well translation tasks. So, the type of jobs that translators are currently doing on the nature of the translation job has changed significantly since.

Mary McCaughey (00:14:54:00 – 00:15:53:00): Yeah. So perhaps it's fair to say that it's, you know, to paraphrase marqueen, the rumours are exaggerated of the demise of the of the worker. But at the same time, it's different types of work that we could see emerging that perhaps we can't even imagine at this stage, which will be born out of this new development in AI.

I mean, I suppose if we're to imagine, Ioannis, that the work does continue in many of these areas that we are talking about and that many of these industries do retain AI and generative AI in the workplace. And that that grows when we talk about the physical aspects and particularly when we look at industries that are associated with a greater level of physical hazards, what sort of impact are we seeing AI having on the worker? Or indeed on the workplace?

Ioannis Anyfantis (00:15:53:00 – 00:18:00:00): First of all, thank you, Mary, for the kind invitation to participate in this podcast. And if we talk about how AI is sensing the game when it comes to workplace safety, especially in industries where physical hazards are a big concern like construction, manufacturing or logistics and mining, I think you. But also, we see that artificial intelligence has huge opportunities to offer, but also we should approach them with caution. So, on the positive side, the AI power, the systems have huge potential over the so-called 3D tasks, those that are dirty, dull or dangerous.

For example, autonomous machines and robots and now handling heavy lifting can perform welding and even working at heights. Also, AI driven computer vision is being used to monitor work sites in real time. Spotting any unsafe behaviours and ringing an alarm and preventing accident from happening when workers are getting too close, for example, in hazardous zones further to that, wearables and sensor-based systems are also making a difference. Tracking things like posture, fatigue or exposure to harmful substances.

But it's not all smooth sailing, because artificial intelligence introduces new risks, too. So when it comes to the physical part, take for example, exoskeletons, which they are reduced. They are designed to reduce the strain. But if they are not fitted properly or they are used too much, they can actually cause problems like skin irritation. On top of that, when we're talking about virtual reality system or augmented reality system, which are great for training. They could sometimes lead to disorientation or eye strain, and moreover, automation offer leaves behind fragmented manual tasks that are repetitive or ergonomically poor, increasing the risk of musculoskeletal disorders. So, if we introduce this kind of machines without appropriate consideration of the tasks that they are automated, they could introduce new risks to the workplace.

Mary McCaughey (00:18:00:00 – 00:18:27:00): And I suppose to be honest, that also implies that a kind of an additional level of training is required for some of this. How to wear certain equipment, how to adjust in in terms of managing the impact in your workplace of AI tools?

Ioannis Anyfantis (00:18:27:00 – 00:19:23:00): Definitely, training is very important when these kinds of technologies are introduced and training is needed for many different purposes in order for workers to develop those new skills that are needed in order to work with these technologies. But also, training would require a different way of thinking from the workers, because when working with artificial intelligence, think about collaborative robot or something, it is not similar as working with humans because there are also these social aspects issues related to communication and apart from that, all the psychosocial aside, because task go to high automation could cause additionally cognitive overload. Also, when talking about training, training could increase stress to workers. So, all these issues that have to do with training, reskilling, upskilling of workers. It's a huge discussion, yeah.

Mary McCaughey (00:19:23:00 – 00:19:45:00): Yeah, it's critical for this. Yeah. Yeah. And I suppose then you, OSHA, part of the job that you're doing at the moment is carrying out an updated list on an ongoing basis, monitoring the emerging risks that are occurring.

Ioannis Anyfantis (00:19:45:00 – 00:20:23:00): Exactly. This is the core part of our current campaign that is about Occupational Safety and Health with the digital aids, where we are examining, we're identifying the main areas on which Occupational Safety and health will have. When introducing the workplace is changing the way we work and digital technologies now also are changing the whole way that people perceive work because the workplace is changing the tasks performed by workers are changing. And yeah, it's not like the typical.

Mary McCaughey (00:20:23:00 – 00:21:02:00): And tell let's talk a little bit then about the psychosocial elements that that you referenced there. I mean, what are we seeing in terms of the main concerns around psychosocial risks? And I know when we talked previously we talked about isolation potential, social isolation, maybe breakdown of trust in the workplace work intensity are these the areas? Are there others that we should be concerned about?

Ioannis Anyfantis (00:21:02:00 – 00:21:59:00): Yes, thank you for giving me the opportunity to talk about all these issues because at the OSHA we are closely monitoring how these changes affect mental health well-being and all these psychosocial risks. So, one of the key concerns is the increased work intensity. Because AI systems can automate task allocation, monitor performance and track productivity, and this kind of systems are often called AI for worker management systems. But this can lead to tighter deadlines, higher output targets and reduced recovery time, all of which contribute to chronic stress. Also, we're talking about chronic cognitive workload of workers. That is also a significant risk and our Euro-barometer survey shows that workers often feel that AI reduces their autonomy, accelerates the pace of work and increases surveillance factors that are strongly linked to psychosocial strain.

Mary McCaughey (00:21:59:00 – 00:22:15:00): And Ioannis are we seeing evidence of that? I mean, that's what workers are saying or reporting. But are you seeing also that there is evidence to support that?

Ioannis Anyfantis (00:22:15:00 – 00:23:30:00): Yes, apart from the surveys, our research studies actually provide evidence on all these, apart from the cognitive workload we have seen that loss of autonomy is another major issue.

When AI systems dictate how tasks are assigned or evaluated, workers may feel that they have little control over their work and even well-intentioned tools like gamification or performance scoring could create unhealthy competition, and fear of penalties. Also, our SNR survey confirms that in workplaces where diesel technologies are present, there is a marked increase in time pressure, irregular work hours and the blur that is between work and personal life. And these conditions are worsened by automation, which can extend availability expectations and reduce opportunities for rest. Social isolation also is growing risks because as AI automates more interactions, especially remote platform based on digitally managed environments. Workers may find themselves engaging more with systems and less with colleagues and.

Mary McCaughey (00:23:30:00 – 00:23:44:00): And, with people, that whole human-centred approach that we are trying to advocate and that we see as important in in terms of moving into the next phase of digitalization, that's we're in danger of losing out there.

Ioannis Anyfantis (00:23:44:00 – 00:24:12:00): Exactly, and in order to address this kind of issue, I think you also we promote the what we call human centre approach to AI, advocating for workers participation in the implementation of AI systems, establishing a clear communication channel between the workers and the employers, transparent and explainable algorithms, because sometimes an algorithm could be like a black box, so there should be a clear understanding of how the algorithm works and training and upskilling of workers in order to support this adoption.

Mary McCaughey (00:24:12:00 – 00:24:48:00): Sure. Yeah. That's good, in terms of they are good measures to put in place, but isn't the danger that all of this means that the worker begins to increasingly feel slightly out of control. And so therefore they lose trust in the environment in which they're working. I mean, have you seen that, when we're talking about the sort of the roll out of AI in the workplace.

Ioannis Anyfantis (00:24:48:00 – 00:25:26:00): Given our research and the interviews that we have conducted in numerous areas and applications, because based on our research, we have developed several case studies in several workplaces in Europe. When talking about AI based systems to automate tasks or platform work, yes there is great evidence that workers may lose trust to the system if there is no kind of transparent exchange of information and the workers participation on this.

Mary McCaughey (00:25:26:00 – 00:26:21:00): So turning back to you Dragos a little bit, I mean with all the work that has been done on job quality and we're pushing very much for job quality and of course there's a road map for job quality underway at European level at the moment. But but talk to me. I mean, what? What are the main concerns that we're seeing when it comes to algorithmic management and AI? I mean, are we actually we've seen an improvement in job quality over the years for the most part. But are the elements that people value now, which is actually according to our recent survey, they are things like trust their safety, they're positive work environments. You know, these are the things that people actually prioritise over in fact pay to a large degree in in many places are we are we looking to see that being under, under kind of threat.

Dragoș Adăscăliței (00:26:21:00 – 00:28:34:00): That's a very good question, Mary and I would begin by making a distinction first, because I think that's very important and the distinction is between algorithmic management and artificial intelligence, because I think also sometimes people use these terms interchangeably and this shouldn't be the case. So by AI we mean all those technologies that we are just discussing until now, generative AI, collaborative robots, advanced artificial intelligence that are deployed in factories and so on and so forth by algorithmic management we mean technologies, digital technologies that are used in the workplace to automate management functions. They can be married or put together with artificial intelligence, but they can also function without actually out of artificial intelligence systems. And I think that's an important distinction to make because we don't need to actually necessarily link all the potential negative outcomes that algorithmic management systems can bring to artificial intelligence, right? Those tend to be tend to be linked more that are more likely to be linked with automation of tasks, the use of software to monitor the hours, the use of software to allocate tasks, the use of software to allocate performance targets. And when we look at the European Working Conditions Survey, this is basically where we see the most negative impacts, right? We call it in in house an algorithmic penalty for workers because we see that workers who are subject to algorithmic management in in this way tend to have poorer working conditions across the board. 

So they have, they tend to be more stressed at work. They tend to report higher levels of work intensity. They tend to report lower control over their working time, and they also tend to report lower, like less freedom to apply their own ideas, and I think that's very important because it means that algorithmic management can potentially have negative impacts, not only on the individual, but more broadly on the organisational side. And that's where the danger lies.

Mary McCaughey (00:28:34:00 – 00:29:31:00): Yeah. But it's funny Dragos because you're talking about the impact on the individual who is a human, clearly a human worker. But then we also know when there's evidence that has emerged over recent years that that the skills into the future, you know, will value the soft skills, the communication, the, the sort of the, you know, the emotional intelligence. That's required for coordination and delegation in the workplace. That sort of human oversight. I mean, what are the mechanisms, though, that can be introduced to prevent organising organisations moving in a direction which really looks towards greater efficiencies full? And that the human dimension is undermined.

Dragoș Adăscăliței (00:29:31:00 – 00:31:03:00): That's a very good question again. And I think from our case studies, we do not see, at least in Europe, organisations trying to actually move towards full automation of tasks. On the contrary, I think we're at the stage when organisations experiment with new technologies and try as much as possible to involve workers in the decisions that lead to the adoption of a new technology. So, the case studies that we've seen in Eurofound reveal that when organisations try to involve workers in the piloting phase, try to have work a worker, say when the technology is being deployed across the organisation, but also when the when decisions are linked to how the technology is embedded into current work processes. This results basically in a successful adoption On the contrary, when technological deployment is a unilateral decision of the employer, this tends to result in favour. However, we haven't seen this kind of behaviour because I think organisations themselves understand and recognise the value of the workers themselves, so I don't think we're in this kind of 0 sum game scenario in which technology technologies are deployed by organisation. Just because there is availability, but they are deployed because they can actually bring significant improvements for both of them.

Mary McCaughey (00:31:03:00 – 00:32:16:00): Yeah. But I suppose in this drive for greater productivity and increased competitiveness, you know that we're in at the moment for the whole of the European Union there, there is a danger that we make a move towards increased efficiencies, whatever that means, which could look towards applying more of these kind of tools at the workplace. And I suppose it's a good message to take away from this, that without the human intervention, the human dimension, the human-centred approach that is actually going to be counterproductive. And I do think that's a good message to come from the discussion that we're having today as we move forward. Because I suppose Ioannis from your perspective, also from the OSHA enforcement perspective, I mean there would also be systematic changes required for like authorities like the national labour inspectorates to ensure that they also address these kind of new dynamic, emerging, organic changes and risks which are being caused by digitalization, what would they be?

Ioannis Anyfantis (00:32:16:00 – 00:34:20:00): Definitely based on our research findings, enforcement authorities must also undergo a series of technological and organisational transformations and actually the vast majority of them are already doing so. So you also said research on supporting compliance that includes case studies on different member states offer a clear mapping of these changes. So first we can see that inspectors need new skills in order to be able to assess those risks that are related to artificial intelligence, robotics and digital platforms. And this could include that the standing psychosocial and organisational risks as well and interpreting data from these smart systems also we can see that labour inspectrates today are taking advantage of technology and they are integrated digital tools in their workflows and for example they are using AI based systems in order to more effectively.

Target their inspections or they are using mobile inspection applications or they are even using chat bots that can automate the routine tasks and provide information to or relevant parties. Currently we are having several examples in Member States of implementation in these fields and also the inspector is should rethink the internal structure to encourage collaboration between host experts, data scientists and legal advisors, and finally stakeholder engagement is key because we have identified that enforcement should also focus on education and facilitation with enforcement powers. As the last resort for Labour inspectorate and workers organisation, social partners partners could play, could be part of the strategy. I think also we're also collaborating with the the senior Labour Inspector committee and we have seen and that actually they are all. We're also working on this area. They have developed a paper on digitalisation and the use of machinery and robotics using AI, presenting several case studies in order to inform labour inspectors, and this could be used as a training machine material for labour inspectors as well.

Mary McCaughey (00:34:20:00 – 00:34:54:00): Building on that, actually, when I was talking to Dragos, I was talking about the desire for greater productivity, increased performance, etcetera. I mean here, have you looked in the US or have you looked at what are the best ways of introducing AI systems that would allow for this increase in performance while at the same time ensuring that we offset the risks for workers that are emerging? 

Ioannis Anyfantis (00:34:54:00 – 00:37:37:00): Actually, about based on our findings, we have found that the most effective way to introduce AI systems into the work organisation is through a human-centred, participatory and prevention-through-design approach. This will ensure that prevent performance gains do not come at the expense of worker safety, health or autonomy, and through our research we have identified actually several important points relevant to that. First, work procedures should be redesigned so that artificial intelligence and human tasks are clearly defined, minimising psychosocial risks and.

Social effects and the AI should be assistive, not substitutive, supporting human decision rather than replacing it, particularly when humans are affected. For example, in our case study one of our case study, we have seen a manufacturer who used the AI as an assistive tool allowing workers to retain control over final decision worker participation from start is crucial. When workers are engaged in the design, deployment and evaluation of the AI systems, outcomes are far better for both performance and safety and this also promotes acceptance from the worker side and successful integration. And we have identified in these cases there were given names to the robots. So it looked that there was a good spirit of collaboration between them. Skills development was also found to be a key issue. Training was found to be very important and not only in terms of the hard skills and this risk healing procedure, but also in terms of soft skills, communication and development of critical thinking as well because there are several things that. Humans cannot afford losing further or more, as automation can leave behind fragmented manual tasks. Risk killing is both a challenge, but also an opportunity where the procedure can increase stress to workers. Also, AI based systems can. Should comply with legal and ethical standards included in transparent decision making. Nowadays, we are having apart from the Framework Direct Directive, that is a fundamental legislative provision for Occupational Safety and Health. This system should also comply with the machine regulation, the AI Act, the Data Act, while at the same time GDPR should be respected and human oversight is of high importance, as responsibility cannot be sifted or served with the AI system for this purpose.

Mary McCaughey (00:37:37:00 – 00:31:43:00): But, Ioannis are these just guidelines? Or can they actually be imposed upon work organisations themselves? I mean because it's all very well to have documents that sit in offices in Bilbao or in Brussels. But I mean, if these are just ideas that we think would be good if they're not rolled out and embedded in companies, how is this going to work?

Ioannis Anyfantis (00:37:37:00 – 00:39:08:00): For this purpose we have developed, when it comes to for example the automation of tasks in noir, a tool that the enterprises, especially small and medium enterprises could use in order to be able to identify all those risks that are reported in our studies. They can implement those measures in order to mitigate this kind of risks at the same time, since the labour inspectorates and the authorities who have adequate training, as well as workers, representatives and the unions have a more active role and workers participation on this. This could create a climate that could assist with all these good practises enterprises in order to successfully introduce this kind of technologies and automations. At this point I would just like to come back to the issue of trust mentioned earlier. And building trust in AI at work requires transparency, fairness and human oversight. From our point of view, especially when systems influence task allocation, performance evaluation or decision making and without clear accountability and working involvement, AI can feel unclear and disempowering, undermining confidence and increasing stress.

Ioannis Anyfantis (00:39:16:00 – 00:40:34:00): And also I would like to come back to the question referring to job control and autonomy. Regarding the evidence: There is growing evidence that AI is already impacting workers' autonomy and job control. According to our data, 30% of EU workers say that digital systems assign their shifts and over half say that this system determine their speed of work. While these technologies can boost efficiency, they can also risk reducing job control, increasing work intensity, and leading to this killing and surveillance and this could of course increase stress, disengagement and even burnout. Furthermore, according to our latest SNR survey that was conducted in 2024, only 4% of the establishments report using none of the digital technologies considered. Also the previous SNR. So if there even though there is a growing concern about the impact of technology, the lack of dialogue in almost 2/3 of the enterprises is worrying, especially as AI systems can intensify risks like stress, fatigue, isolation and loss of agency.

Mary McCaughey (00:40:34:00 – 00:42:21:00): And I think clearly, I mean, you're making the point here that this is a growing emerging issue that we have to monitor on an ongoing basis. And I think Dragos you made that point earlier also about the change even over a period of a year that we could expect to see in terms of take up and impact. And the trust issue I mean is fundamental in in all areas of our lives and work. And certainly, if this is going to impact on trust in the workplace, then it is something that needs to be addressed and you've raised many of the issues that of the contributory factors that could be brought to play in the workplace. But I think I would like to bring it back up again just to sort of saying, why is this important because you know, I suppose we have to move away from the individual workplace that wants to see greater productivity and competitiveness. But really to the global stage where we're seeing Europe placed against the other major trading blocs and we've, as we said at the start, Dragos, we've looked at you know other trading blocs and how they're moving forward. And I know we've put investment into AI or we're planning to put it in according to President von der Leyen and in a in a massive way or what we would consider to be a massive way, but I understand that's actually kind of like tuppence worth when it comes to the other trading blocks in terms of the investment that's been made in, in, in those areas.

Dragoș Adăscăliței (00:42:21:00 – 00:46:48:00): Yeah, I think that's particularly the issue for Europe because we're in a in a catch, we're in a position where we try to catch up with the big two other big blocks that are investing in. Yeah. And that's the United States and China. Just to give you an example of the scale of investments in artificial intelligence in the United States, last year, private investment in the technology was in the range of €500 billion, right. Moreover, that sort of investment which is much higher. When compared to previous cycles of investments in new technologies accounted for about 80% of economic growth in the United States. So that gives you a kind of an idea of what we're up against in the UK when it comes to artificial intelligence. So it's no small task, basically, A - to catch up with this big competitors including China and B - to kind of develop the type of artificial intelligence systems that we would like to see. Because it's important to mention that we already in Europe have a a framework of artificial intelligence that are that emphasising key principles such as transparency, unbiased data and so on and so forth, that lie at the core of the way we think.

AI is going to benefit our societies, which is very different basically to how both the United States and China are currently thinking about AI. So, we want AI for social good, and that's big tasks in itself, not to speak about the amount of investment that we need to put into that, so the policy initiatives that you've seen over the past year coming at the at the European level, try to fulfil exactly those goals, A - to kind of develop an European AI that is fundamental on the principles that we have, and B - incentivize more businesses. More companies, including the public sector, to use artificial intelligence even more because you know this the goals that we have at the moment that are driven by digitalization. Are interconnected with the goals that we're pursuing in other domains, be that greening, be that dealing with demographic change, be that trying to have better public services and so on and so forth. So you know AI is let's say, if you want a transversal theme that that cuts across various kind of domains that we care about. But as societies and the other thing that I wanted to say is to go a bit back at the discussion on productivity, because I think to me at least the issue of productivity on the issue of productivity lies everything, right? So, if AI ends up increasing productivity, improving working conditions, then we're in the best-case scenario, right? And there's a massive promise there, right? There's a lot of research currently being done on the issue of productivity that shows that there's a lot of, let's say, quite a few unfulfilled promises if you want. When it comes to artificial intelligence, there's, you know, randomised control trials that look at the impact of artificial intelligence and productivity, find massive improvements. So, in the range of 40 to 60%, improvements in productivity.

On the other hand, if we look at other types of research that use surveys on artificial intelligence, the productivity improvements that they see there are actually very minor or zero actually. And there is the danger, right, because what we are now in the middle of a of a system in which we're investing. We're putting a lot of trust in, in the technology that holds the promise of productivity improvements and holds the problems, the promise of making our societies better. Of course, with the type of implications potential negative implications that we were discussing earlier, and Ioannis has highlighted some. But to be fair about this, we don't really know exactly how much AI is going to change in the in the next five years or 10 years in the world of work, right? So, there's there is also a lot of fiction when it comes to AI, because so much depends on this project succeeding.

Mary McCaughey (00:46:48:00 – 00:48:24:00): There's so many known knowns, unknown knowns, and unknown unknowns, Dragos. So, in this world, and I suppose the danger that we also have as European level is that we have, you know, so many strategies. We have the Apply AI Strategy, we have the AI innovation package, we have various other directions and initiatives that are being brought into play, but without them being embedded in a really active way. As you say, we're not really going to see the results bearing fruit and that's something that we will have to monitor on into the next phase if we're going to see the kind of benefits we want to reap from this and particularly if we're in competition with the other trading blocks, as you say you have invested and many billions more into this and then we have been able to do so far and so look as you know and as you can see, we could talk about this really probably for the rest of the week and we could come back next week and do the same again. But I would like to just if you like close this off as we do with each of our podcasts, and really, if I could ask you very briefly and really do be succinct on these, what are your three recommendations if you had policymakers in front of you to maximise the benefits of AI at the workplace while protecting workers from the emerging risks to their working conditions and two job quality. What would they be? Talk to me in 3 Ioannis.

Ioannis Anyfantis (00:48:24:00 – 00:49:25:00): Right. Quite challenging. In order to summarise everything in three bullet points as there is no silver bullet in order to achieve this, I would say that the 1st is that there should be a careful selection of tasks and processes in which AI should be used. So as a key message, don't use AI for the sake of AI worker consultation and participation is important. The second key message could be that the human centre approach should be followed with the humans in command. The AI should support decision making and not replace. Based in decision makers in particular areas where especially when it comes to to decisions that have to do with humans and workers should maintain job control and autonomy, and it's a very important issue and the last one is the equal access to information. And transparency, the system should only be used for the intended purpose and the data privacy should be respected at any cost.

Mary McCaughey (00:49:25:00 – 00:49:47:00): OK. OK. That's a good one, though. I like that one AI should only be used where AI should be used. I like that as a basic premise. Yes, Dragos.

Dragoș Adăscăliței (00:49:47:00 – 00:50:44:00): And for me, there are three kind of key issues that link to having a successful AI deployment and leveraging potential AI benefits. First, prioritising applications that augment and do not just automate worker tasks. I think this is important. There's a lot of research that shows that augmentation brings numerous positive outcomes for both organisations and job quality. The second would be involve workers at the point of deploying technologies and at the point and involve workers also in the monitoring of artificial intelligence applications in organisations and three redesign organisational processes. Because just deploying an AI in itself will not actually bring the promised productivity gains, we actually see that when organisations adjust other processes that involve all the deployment of artificial intelligence, this brings positive outcomes for both workers and and themselves. 

Mary McCaughey (00:50:44:00 – 00:50:44:00): And what's very interesting about what you have both brought to the table at the end there is that human dimension involving the worker, but also ensuring that that human centred approach to AI is maintained throughout every decision that we make in this area going into the future. I would like to thank you both that that really has been a fascinating period of discussion on an issue which is equally fascinating. Ioannis, thank you and thank you EU-OSHA for your contribution today and Dragos, thank you as always for your contribution. I would remind you that you can listen to Eurofound Talks on this subject and on many other subjects which are related to this. For example, on job quality, on digitalisation and various other topics. You can access Eurofound Talks on Spotify, Apple Podcasts for wherever. You get your podcasts EU-OSHA, and you're a fan both for wealth of data information online, which is relevant at both Member State and sector level that will be allowed, will allow you to sort of delve deeper into the information that you have been hearing today and you can access all of that on their websites. You can also watch our webinar, which took place with setup up earlier this year, which talked about how AI is changing the world of work in Europe and setup up is another sister agency who is specialised in training and skills development. Of course you should follow us on our social media channels. You should subscribe to our newsletter and if you have any queries from today's discussion. Please don't hesitate to contact us in Eurofound or in EU-OSHA, we're always delighted to respond. That is what we're here for. So, until next time, until Eurofound talks to you.