We use cookies to personalise and enhance your experience on this site. By clicking Accept or continuing to use the site, you agree to our use of marketing and social media cookies as described in our Privacy Policy.
Accept and close
This is a lecture given by IFOW Co-Founder, Professor Sir Christopher Pissarides, at the British Ambassador's Residence in Brussels.
The main theme of my talk is that automation technologies (robotics and AI) are bringing many changes to labour markets. I then ask, how are workers in industrial countries affected? What is their impact on productivity and growth? Can the situation be improved, and if so, how?
A lot of what I will say is based on research funded by the Nuffield Foundation with title The Pissarides Review of the Future of Work and Wellbeing, based at the Institute for the Future of Work.
Although like many others engaged in research in the labour market, we believe that improving productivity is a paramount objective, we also believe that the wellbeing of workers should be a key objective as well. Higher productivity helps government achieve its social objectives, such as good health and education systems and social support to vulnerable groups. But to get the higher productivity we need to work, and if work makes us unhappy, gives us depressions and mental illness, is that a price worth paying? I hope you will agree that the answer is that we should pursue the two objectives in parallel and improve wellbeing alongside productivity.
Policymakers, and especially financial markets, focus too much on GDP growth. I think this is counterproductive – comparing differences in decimal points between expectations and realisations of GDP growth and making markets volatile in unpredictable ways in response to them, is bad for the economy. Measuring precisely GDP is so fraught with difficulties that small differences in growth rates should be ignored, not used to drive fluctuations in the cost of money, which then influences investment, and the income people get from their savings.
There are some encouraging developments. Judging by mentions in English language publications, wellbeing is on an upsurge. In the Labour Party conference in 2021, the then Opposition Leader Kier Starmer, expressed the view that the Treasury, when evaluating the benefits and costs of a proposed economic policy, it should also include in its calculus its impact on wellbeing. There are certainly a lot more references to wellbeing currently, both in policy and in the academic literature.
What is wellbeing? Wellbeing at work is simply the way that the worker feels about work. Are they happy doing whatever activity they are doing? Stressed? Relaxed? Sometimes objective measures, like medically diagnosed stress, mental illness, or absenteeism are used. But subjective measures, such as responses to questionnaires, are more encompassing and empirical studies are now using them more often. That’s the way that I will interpret wellbeing in this talk.
Changes in technology necessitate the transition of workers from one role (set of tasks) to another. They might also necessitate a job change, the closure of companies and the formation of new start-ups. “Role turnover” is more common than job turnover. But whatever the type, worker transitions take place continually. As a matter of routine, there is a lot of job creation and job destruction within sectors. The concerns about automation technologies is this: the disruption is bigger than in the past, in terms of the new things that workers need to learn to make the transition.
Companies can do certain things to achieve smooth transitions and adapt to the new technologies. Making sure that their workers enjoy a good working environment and providing good training possibilities can ensure that the transitions are smooth and benefit the company as much as their workers.
Since modern economic growth began, large scale innovations have brought one industrial revolution after another. Industrial revolutions are periods of intense technological change and social transformations, that brought the development of science, the reduction of poverty and the improvement of the quality of our lives. Each industrial revolution is characterised by a big new discovery, usually associated with energy - steam power, waterways, railways, the internal combustion engine and fossil fuels, electricity, and computers. The current one, many referring to it as the Fourth Industrial Revolution, is based on automation technologies, mainly robotics and artificial intelligence, driven by digital power.
Growth can benefit all in society if it is sustainable and inclusive, and modern digital technologies have the potential to achieve this objective, if used well. Unfortunately, the signs are not good so far. Inequality is high and not falling, many jobs created for the less skilled are not good, like those in the gig economy, and there is evidence of use of digital technologies for political purposes or in armed conflict.
But let me stay within the realm of economics. The Fourth Industrial Revolution is now driven by artificial intelligence. Robotics, the other main automation technology, have been applied widely in manufacturing and are now expanding in other sectors of the economy. The direction in which robotics are heading and their main impact on production have been well studied and are reasonably well known. Their main activity is “handling” goods in manufacturing, so they facilitate manual work. They take away boring tasks associated with manufacturing, so overall they are good for worker wellbeing.
Robots have a positive impact on productivity, and they generally replace workers, but they also bring productivity gains and help companies create complementary jobs. Think for example of a car company that employs robots in its assembly plants and creates more jobs elsewhere to design new models, expand the technology and sell the products. The main impact of robots on jobs seems to be to bring benefits to early adopters, who “steal” jobs from their competitors. Overall, the most likely impact of robots on manufacturing employment worldwide was to reduce it. We have some evidence for this from French companies and at the country level from the OECD, where robot adoption has benefited exporters such as Germany and Japan, whereas it hurt others, like southern Europe and the UK.
AI can have many more applications than robots, being capable of influencing work at all levels except for the very top managerial level, where complex decisions need to be made. We don’t know precisely how much it is being used because it’s embodied in software that is not directly observed, unlike robots, which are self-propelled independent units. There is certainly a lot of hype – workers seem to live in fear of losing their jobs, and about 80% of employers say they are using it. But so far, the evidence shows that workers are not justified in their fears, and employers are using it in simple applications, like cameras to watch remote parts of the office or remote working. At governmental level, the most developed application of AI is in surveillance of all kinds, like watching movement in public places, airports, and offices.
The general view is that the use of AI is still very limited, but we must prepare for it because it’s coming. In my view, more worrying than this, is that we don’t know in what form it’s coming, as we have a lot of choice how to develop it. So far, the development of AI has not been in the direction that most people would choose it to be, to improve human wellbeing. My remarks about surveillance indicate what I have in mind. The increasing use of AI in military conflict is another, and the insufficient transparency with which it seems to be applied in many companies is behind the concerns expressed by workers’ representatives. In research that we did at the Institute for the Future of Work we found that in Britain many applications of both AI and other automation technologies have reduced the quality of work, at least as perceived by workers.
There are certain things that we know about AI, other than its potential, and these relate mainly to the enablers that countries need to have to take full advantage of it and develop it in socially beneficial ways.
To be able to make extensive use of AI countries need to have a strong digital infrastructure – powerful broadband, which is reliable, and universal. Their innovation capabilities need to be strong, with well-trained human capital to support them. They need to be open to collaborations and interact effectively with other similar countries. And they need to have good quality labour force institutions, including good social support to help workers’ transitions across jobs.
On these criteria, the US and China come out as top performers, partly because of their size and partly because of the emphasis and resources they have put into developing a good digital research environment. In the United States the best examples are Silicon Valley and the area around Boston and the top universities located there. In China the government has been providing strong support since 2010, when the country switched to the development of modern technology to take advantage of the “talent dividend,” following its expected demographic shifts. In the next tier come Germany, Japan and South Korea, and other smaller northern European countries. The rest of Europe is not as well prepared, but it is of course better prepared than countries located elsewhere, such as Africa and South America. Compared with the big two, Europe is well prepared in one aspect of the transitions that will be necessary in the technological revolution that is coming, in the provision of social support and training facilities. In terms of wellbeing, or subjective happiness measures, Europe scores higher than other countries, most likely because of the social support and lower working hours that it offers.
AI works with very large data sets so if it is to perform a task, the company employing it needs to have access to such data. Large digital companies like Apple, Google, Amazon and others like them have a strong advantage in the world of AI, and especially over the public sector. Public services will improve enormously if the public sector could have access to their data, and although it is difficult to bring this about in a free society, it is worth trying.
Traditional AI, which carries out commands based on the data, is finding applications, and it’s predictable where it’s going. Generative AI, which makes decisions and has much more capability to analyse data and language, can do many more tasks. Recent developments, which have reduced the computing capacity needed to execute it, are major advances in the application of gen AI. Large language models, such as Chat GPT, are ones that are threatening even professional jobs, such as paralegals, civil servants, and programmers.
Given the impact of generative AI like ChatGPT on the professions, many will need to rethink their recruitment and training procedures. For example, senior civil servants, accountants, and lawyers are recruited by promotion from junior positions, whose tasks are mainly the collection and processing of data, the writing of reports or the preparation of briefing documents. These tasks involve a lot of on-the-job training about the specific requirements of a firm, which enables the best performers to move up in the company. If these tasks are taken over by big language programmes, how are senior executives going to be recruited? It seems to me that the answer is in extending and expanding the education and training that candidates receive before they enter the profession. Currently, joining a large legal firm or the civil service does not require much advanced training, because of the on-the-job training provided by the firm. The problem is that in the future there won’t be enough positions at the junior and intermediate levels to continue the current practice. Professionals ought to start thinking what skills their recruits should have prior to entering more senior positions in the firm and inform higher education institutions.
One thing that we can say with some confidence is that we are unlikely to face a situation of job shortage across the economy. We should not be asking, where are the jobs going to come from in the world of AI, but what skills will I need to work effectively in this world?
It is easy to guess the general type of skills that will be in demand in the age of AI. They are of two kinds, at opposite ends of the training that they will need. They are new technology skills and “soft” skills. But in what proportions will they be demanded, and how are they changing?
At CEP/IFOW we got all online job ads from Britain, about 2.5 million, and cleaned the data down to about 1.5 million per year. Firms advertise jobs with about 3,700 separate skills (on the Lightcast – formerly Burning Glass – definitions). We analysed in detail the skills required in 2016 and again the skills required in 2022. We find that “Old fashioned” skills dominate what employers want, in both years.
Although the top three skills demanded in each year are unchanging, and much more dominant than the ones below, new skills related to digital technologies are emerging. Recall that Britain during this period, like most other advanced countries, suffered from labour shortages. For the first time since vacancy records began, advertised job vacancies far exceeded job seekers. But although firms were looking for workers qualified to analyse data and do some technical tasks, the vast majority of employers were looking for workers who were able to communicate clearly, be confident in dealing with clients and colleagues, and support a good customer base; in other words, workers who were effective in person-to-person interaction.
The biggest changes, both of new skills emerging and old ones disappearing, are in IT. There is a change in the IT tasks performed, the type of servers used, cybersecurity etc. Dealing with people ranks high even in the IT sector – as in onboarding and talent acquisition, which, although listed under human resources, uses a lot of IT information.
IT is followed by Health services and then Engineering. These are also sectors which are experiencing high demand for labour and changing skill requirements.
With changing skill needs, firms need to spend resources on training. Government has a role to play here, subsidising training, especially for smaller firms. Changes are also required in secondary education, in which Britain’s system of early specialisation in a small number of A levels is not one that is well suited in the new world of fast changing technologies.
Should one train in STEM? The answer to this question might be obvious but there are some subtleties involved. I would say Yes, up to some basic knowledge – but don’t be misled that that’s where the future of human work is! Many STEM related tasks can be done by AI and as the use of AI is expanding, fewer STEM trained professionals will be needed. AI is not good at offering person-to-person services in non-technical sectors of the economy, and that’s where human labour will always be wanted. Especially in our ageing societies. AI machines might be good at diagnosing health problems and even performing operations, but their capabilities do not even get close to humans in providing care and support for older people.
The challenge for these person-to-person service jobs is how to make them attractive for large numbers of workers, especially new entrants to the labour market. These sectors are predominantly labour-intensive and low productivity growth sectors, so they need to be supported by the rest of the economy. The problem is aggravated by the fact that in countries like Britain and other European countries, the public sector plays a big role as employer. The problem is especially acute in the British National Health Service, which is doomed if it continues with the practices of recent years. We should be careful, because the way that we are going, ending up with a broken health system will not be the only problem that we will be facing; we are also going to get more inequality, between the haves (technical skills) and the have nots. And the inequality will not be corrected by making more of us acquire the technical skills, because the jobs will not be there.
I have focused on the impact of AI on jobs and in the changes that it is bringing to labour markets. What about productivity growth? Are we on the threshold of a major productivity boom? I have already said that robotics in manufacturing have increased productivity. Not by very much, but sufficiently to make the companies that adopt them first more competitive and take jobs from others that are slower to adopt them. What about AI? Some believe that AI will have the capacity to increase productivity more substantially. Personally, I doubt whether it will make much difference to the productivity statistics, although it should improve the quality of services that we get and reduce mistakes. This is very much like the impact that computers and software, such as Microsoft Office in the early days of digitalisation, had. You may have heard Robert Solow’s famous quip at the time, that we see computers everywhere except in the productivity statistics. I think something similar will be experienced with AI, for as long as we have no way of measuring quality and speed of service provision in our national statistics. Or even more broadly, because we have not yet found ways of measuring wellbeing in our GDP statistics.
It is clear from the discussion so far that both workers and their employers are facing big challenges. There are a lot of uncertainties about the future and about the best approaches to work. How are workers coping?
Surveys of workers’ attitudes to work have proliferated recently, as part of life “happiness” surveys. They find that losing your job is one of the life events that gives most unhappiness (along with illness/death in the family and divorce). But being at work is one of the activities that gives most stress and dissatisfaction, especially some things associated with work, like commuting or seeing your boss. Only being sick in bed scores consistently below being at work.
One might ask - isn't that what the economist should expect? No one should feel happier working than pursuing leisure activities off work, otherwise there would be no reason to pay people to work. But there is room for improvement. When workers are asked about what would make them feel better at work, they mention better communication with managers, more transparency in the company’s activities and plans for the future, better social relations at work, more time flexibility and more autonomy in choosing their role in the organisation. More money comes low in the list. There are also variations across workers. For example, professional people feel better at work than non-professionals.
Recent automation technologies seem to be making things worse. Studies of connections between the risk of automation and subjective wellbeing find negative correlations. Workers in jobs that run a higher risk of automation express less job satisfaction than those in safer jobs. In the recent EU CEDEFOP (Dec 2022) survey of worker attitudes it was found that many more workers worry about losing their jobs to new technologies than it is justified with current technologies.
Clearly, worker wellbeing could be improved without violating what economists call the work/leisure trade off. But why should employers and government care? It turns out that even if employers cared only about their shareholders, there is evidence that worker wellbeing (provision of “good jobs”) improves productivity. More recently, employers care about wellbeing too, the S and G in ESG. This care might be partly driven by share values, but it is more likely driven by the fact that customers care about the social and environmental policies of companies, so in the long run, being kinder to ones employees and making sure that the jobs are good wins business.
Governments should care about wellbeing as an objective of policy: they should choose policies that improve citizens’ wellbeing, not just GDP growth.
Another bit of evidence that points to a connection between providing good jobs and productivity is provided by a recent McKinsey Global Institute study. They find that companies that care about workers’ subjective feelings at work and take measures, including training (at least 70 hours a year), find it easier and quicker to adopt the new technologies.
In other work at IFOW, based on interviews with workers, we found that some technologies improve wellbeing at work, others make it worse. For example, the use of laptops, tablets, smartphones and real-time messaging tools – are associated with an improved quality of life for workers. Workers also feel that they have more flexibility and autonomy at work. But technologies associated with AI, surveillance or automation increase anxieties, partly because of fear of job loss, but also because some good features of traditional types of work, such as enjoying uninterrupted free time away from work, might be lost.
Flexibility in the choice of hours by the worker is another frequently mentioned feature of good work. Several empirical studies of wellbeing at work concluded that having choice over hours is positively associated with worker wellbeing. This feature of good work has become more important since the pandemic. The new technologies that developed during the pandemic to facilitate remote working (zoom, teams, etc.) certainly help, but also, once home working was forced on companies, most discovered that flexible working time might even improve productivity, not only worker wellbeing.
We know from earlier work that as living standards rise, workers will want to work fewer hours. For example, the more productive Germans and Dutch work significantly fewer hours in a typical year than the Greeks or the Spaniards. How you spend leisure is a matter of choice. Some might want to use it for better work-life balance during the normal working year, others might prefer part-time work, and yet others might prefer longer annual leave. More recently the four-day week is becoming more popular as a way of reducing hours. Given the advantages of bunching hours of leisure, as in a two-day weekend, eventually the four-day week might become the norm.
Several recent experiments have been well received. The typical response of employers who participated is that the four-day week “improves health, finances and relationships: It simply makes you happy” (Tyler Grange). Of course, this is a voluntary reduction in hours, not forced by technology taking over jobs. Even in America there is evidence that there is increasing demand for fewer hours of work. The American Psychological Association in its annual survey last year asked workers what would make a future employer attractive to them. The things they emphasised are related to time flexibility, including working from home or working four days a week.
I conclude with the main message, kept brief on purpose. This is that the new technologies based on AI can improve the quality of work, productivity (although not very much) and worker wellbeing, but only if they are applied responsibly – and both government and companies have a role to play in this development.
Professor Sir Christopher Pissarides
