Technology has always changed employment, but the rise of robotics and artificial intelligence could transform it beyond recognition. Researchers at Oxford are investigating how technology will shape the future of work — and what we can do to ensure everyone benefits.
In a famous 1930 talk, John Maynard Keynes imagined a future 100 years hence in which technological progress automated much of human labour. By 2030, he estimated, we could all enjoy a 15-hour working week.
A lot will need to change in the next decade for that to become a reality, but it’s not impossible. Right now, advances in artificial intelligence and robotics promise machines that will take on all kinds of human tasks. Digital communication is creating an internet-dwelling labour force that can work remotely and on demand. And the self-employed are finding that new technological services like Uber and Airbnb can provide a flexible way to make a living. But phenomena like these give rise to a cascade of effects — not all necessarily desirable — that are fiendishly difficult to perceive and predict.
It’s perhaps not surprising, then, that the future of work is a topic of increasing fascination for University of Oxford academics. Both the Oxford Martin School and Green-Templeton College now run specific programmes that focus on the topic, with plenty of researchers — from the Departments of Engineering Science and Sociology to those of Politics and Economics — grappling with its complexity.
‘We see a need for bringing together different perspectives around the study of work,’ explains Dr Marc Thompson, a Senior Fellow at Saїd Business School and the Director of the Green-Templeton College Future of Work Programme. ‘Our role as academics is to contribute to the debate, both in terms of theory and to raise challenging questions and issues for those in government and industry. What will happen as a result of these advances? How will it affect people? Whose interests are being pursued? And what are the long-term implications?’
Rise of the robots
A series of recent studies from the University cut straight to the chase of technology’s impact on employment, focusing on how robotics and automation will affect the jobs that humans currently undertake. The authors, Dr Carl Benedikt Frey (@carlbfrey) and Prof Michael Osborne (@maosbot), come from quite different backgrounds: Frey is an economist interested in the transition of industrial nations to digital economies, Osborne an engineer focused on creating machine-learning algorithms. Together, they’re co-directors of the Programme on Technology and Employment at the Oxford Martin School.
‘It would be fair to ask why I’m doing work related to economics while we’re sitting here in the Department of Engineering,’ admits Osborne, gesturing to his surroundings. ‘But I’ve always had some interest in thinking about what machine-learning could mean for society beyond the industrial applications we usually consider, so when Carl approached me to speak about algorithms and technologies used in automation, and their effects on employment, it seemed like a natural fit.’ This is, of course, exactly the kind of multidisciplinary work the University excels at, and the reason the Oxford Martin School was established. Each of its programmes brings together researchers from different fields to tackle complex global issues that can’t be solved by academics from a single discipline.
Since meeting, the pair has set about developing ways to analyse which jobs that exist today could be at risk of being taken over by robots or artificial intelligence software in the next 20 years. First, they gathered together ‘as many smart people as [they] could’ to decide on 70 job roles that definitely could or could not be automated in the next 20 years. For example, they collectively decided that switchboard operators and dishwashers could definitely be replaced, while the clergy and magistrates certainly couldn’t.
The pair combined this list with data from the US Department of Labor’s O⋆NET system — a database which describes the different skills relevant to specific occupations. Osborne then built an algorithm that could learn from both pools of data to establish the kinds of skills that were common to automatable jobs. When shown other occupations and the skills they require, the software can classify them with a probability of being either automatable or non-automatable.
The pair found that the jobs least likely to be automated are those that require skills of creative intelligence, social intelligence or physical dexterity. These are what they refer to as engineering bottlenecks: current limits to technology that make humans irreplaceable. Osborne points out that it’s perfectly possible, for instance, to have an algorithm churn out an endless sequence of songs, but almost impossible to have it create a hit. Similarly, chat-bots may be able to communicate with you but they can’t negotiate a deal, and robots can assemble objects on a well-defined production line, but they can’t perform a fiddly task like making a cup of tea in your messy kitchen. In each case, it’s because humans draw on a huge wealth of tacit knowledge about culture, emotion, human behaviour and the physical environment that’s hard to encode in a way that a machine can act upon.
But, even with those bottlenecks, the results suggest that as many as 47% of US jobs are at risk from automation over the course of the next two decades. It’s worth bearing in mind that the figures explain which jobs are theoretically automatable, rather than destined to be automated. ‘That may seem like a fiddly point,’ says Osborne, ‘but this analysis doesn’t take into account other factors that we absolutely do believe will have an impact on whether an occupation is taken over by a machine, such as human wage levels, social acceptance, and the creation of new jobs.’ But however you look at it, the numbers are difficult to ignore.
There’s an intuitive counter-argument to the claims that their analysis makes: for centuries, new technologies have been invented that have pushed humans out of work, but by and large most of us still continue to have jobs. In fact researchers elsewhere in the University have shown that the amount of work we all perform remains steadfastly consistent, irrespective of technological change.
Jonathan Gershuny, Professor of Sociology and Director of the Centre for Time Use Research, has spent a large part of his career tracing the way that we all use our time — to work, play, rest and everything else. ‘Fundamentally, there are three realms of activity,’ he explained from the bay window of his Woodstock Road office. ‘There’s paid work, unpaid work and consumption.’ Paid work is just that: the tasks we carry out in exchange for money, be it mining coal, writing a book or performing brain surgery. Unpaid work, meanwhile, is formed of tasks that you could pay someone else to do for you (but for whatever reason don’t), such as cooking, cleaning, gardening or childcare. And consumption is all the activity you absolutely couldn’t pay someone else to do for you — your night’s sleep, say, or eating lunch.
‘Why am I telling you all this?’ asks Gershuny, with a grin. ‘Well, when you define work quite widely like this, you arrive at a really quite extraordinary discovery, which is that work time — that is the sum of paid and unpaid work time — doesn’t change very much. Looking at all the data we have access to, the total is pretty constant, at about 60 hours per week.’ That’s just over a third of our 168-hour week, and a little more than the approximately 50-hour chunk we manage to spend sleeping.
He points to decades of evidence accumulated by his team — in countries including Australia, Canada, Israel, Slovenia, France, Sweden, the Netherlands and plenty more — that confirm the trend, as well as working time regulations from as far back as the Industrial Revolution. His latest dataset — a huge survey of British residents carried out in 2015 — was being downloaded in full the day we met, but a preliminary analysis already suggested that his observation holds true. ‘The truth is, we need work for various reasons: a time structure, a social context, a purpose in life,’ he explains. Indeed, what many people citing Keynes’ famous talk about the future fail to mention is that he went on to suggest that ‘there is no country and no people… who can look forward to the age of leisure and of abundance without a dread.’ In other words, he thought that most us couldn’t really begin to comprehend the reality of not working. Gershuny agrees, arguing that humans will simply endeavour to find new types of work to do in order to busy themselves, whether the robots take over the jobs we currently possess or not.
Dr Ruth Yeoman, a Research Fellow at the Saïd Business School who researches meaningful work in organisations and systems, points out that the human desire to find meaning in work is hard to ignore. She explains that the drive to work is so strong that people seek positive meaning in work that is considered by many people to be dirty, low status or poorly paid. ‘Hospital cleaners, for instance, interpret their work to be meaningful and worthwhile because they enlarge the scope of that work in their own minds,’ she explains.
‘It’s not just about cleaning, it’s about contributing to making patients well and the hospital welcoming.’
This phenomenon allows humans to justify all kinds of work to themselves as useful and relevant, it seems, regardless of what it actually is.
Frey and Osborne aren’t so confident that humans are resourceful enough to create new work for themselves, though. Frey has actually studied the rate at which new jobs are being generated as a result of technological change. His findings suggest that about 8.2% of the US workforce shifted into new types of jobs — that is, roles associated with technological advances — during the 1980s. In the 1990s the figure fell to 4.4% and in the 2000s it dropped to just 0.5%. The evidence suggests that the new industries we might assume to be the salvation of the labour force — such as web design or data science — aren’t creating as many new positions as we may hope.
Part of the reason for that, argues Osborne, is that many of the new job roles being created are related to software, rather than hard, physical goods. ‘Software is pretty cheap with next to zero marginal cost of reproduction,’ he explains. That means that a small group of people can have a great idea and easily turn it into a product that’s used the world over, while barely growing the size of its team. The smartphone messaging service WhatsApp is a prime example: it was purchased by Facebook for $19 billion in 2014, when it served 700 million users. At the time, it had just 55 employees.
Trouble at the top
Counting specific jobs may, however, be overly simplistic when it comes to thinking about how the working lives of real people are set to change. ‘People often think about the work that people do as a monolithic indivisible lump of stuff,’ explains Daniel Susskind (@danielsusskind), a Lecturer in Economics at Balliol College and co-author of a new book called The Future of the Professions. ‘The problem is, that encourages the view that one day a lawyer will arrive at work to find an algorithm sitting in his chair, or a doctor turn up to a robot in her operating theatre, and their jobs will both be gone.’ Instead, he argues, we should be focusing on the separate tasks that make up job roles.
Susskind co-wrote his new book with his father, Richard Susskind (@richardsusskind), whose Oxford DPhil considered the impact of artificial intelligence on law. That was back in the 1980s, when AI systems were rudimentary and typically based on rules gleaned from human understanding. But five years ago father and son — the latter then working in the Policy Unit at 10 Downing Street — realised that a second wave of artificial intelligence was being developed that could have profound effects on professional careers. Since, they’ve been researching how technology might affect the working lives of lawyers, doctors, teachers, architects and the rest of the professions.
‘Not everything that a professional does is creative, strategic or complex,’ explains Susskind. ‘So while many professionals might think that all their work lies on one side of [Frey and Osborne’s] engineering bottlenecks, actually many of the tasks they perform are amenable to computerisation.’ For most, that means it’s unlikely that they’ll simply lose their job to technology, at least in the near future — but they can expect to see a significant change in the sorts of things they’re asked to do. In their book, the Susskinds describe twelve new roles that might appear within the professions — such as process analysers, knowledge engineers, data scientists and empathisers. ‘These are roles that sound unfamiliar to traditional professionals, that require skills and abilities that many of them are unlikely to have at this moment in time,’ they explain.
‘Many professionals have trained to be a professional in the 20th century, rather than the 21st.’
We’re already seeing professionals adapt so that they can work alongside more intelligent technological systems, though. Take, for instance, your bank manager. When you used to approach them for a loan, they’d carefully make a decision on whether or not you were a good risk, then either give you the money or send you home. Now, an algorithm determines whether or not you’re awarded the cash, and yet bank managers still exist. The role has simply changed, to become a customer service and sales job rather than an analytical or technical role.
Not everyone will be as lucky as the professionals whose jobs merely metamorphose, because if all of the tasks that make up a job are automatable, the job no longer needs to exist. Craig Holmes (@CraigPHolmes), a Fellow in Economics at Pembroke College and Senior Research Fellow at the Institute for New Economic Thinking, has been studying shifts in occupational structure of labour markets, and how they’ve moved away from middle-skilled work, with more people now doing high-skilled or low-skilled work. This phenomenon — referred to as the hollowing out of the labour market — isn’t in itself new: middle-skilled factory workers have been losing their jobs to robots for decades, for instance.
But the pace of technological development is now threatening other middle-skilled occupations that in the past we’ve assumed could only be done by humans. Job categories defined as associate professionals, for instance — the people that provide technical services that keep trade, finance and government running — appear increasingly likely to be taken over by machines. ‘In the case of, say, paralegals, there are now pieces of software that can sift through thousands of documents, pull out relevant precedents, and put them together using a very simple format, without requiring any human involvement,’ explains Holmes. ‘So a traditionally middle-tier research job can be perfectly performed by technology.’ The same story could play out in other sectors: large datasets of historical case notes and information from wearables could allow computers to make straightforward medical diagnoses, say, while smarter algorithms might remove the work of number-crunching accountants.
Like car factory workers replaced by robots in the past, Holmes imagines a number of possible futures for those discharged from mid-tier roles. Some, like the bank manager, will be able to assume different roles with similar titles. A small number may move upwards into roles that aren’t yet automatable. Others, sadly, may have to assume lower-skilled jobs or face unemployment.
The nature of those lower-skilled jobs will of course change too. The work of Frey and Osborne suggests that many low-skilled jobs — such as call centre workers, data entry clerks and dishwashers — will be readily automated in the future. ‘In some cases, the cost of technology will be so low that there’s no wage that people could happily accept that would make the job sustainable,’ admits Holmes. ‘In fast food restaurants, for instance, you can replace someone who takes an order with an iPad that will last for years. Nobody would accept a job that paid wages that low.’ But it’s not perhaps quite so gloomy as that, as personal service jobs will likely still require a human touch. ‘We’ll probably see an increase in the number of low-skill service jobs, because people value human interaction and many of those jobs currently seem not to be readily automatable,’ suggests Holmes. ‘That will provide more jobs, they just won’t be great jobs.’
A new digital workforce
While technology may be the mechanism through which many jobs are lost, though, it might very well also be the thing that enables people to take up new lower-skilled positions. ‘There’s been an explosion in connectivity around the world,’ explains Professor Mark Graham (@geoplace) from the Oxford Internet Institute. ‘Something like 3.5 billion people are now online. And that has some significant repercussions in terms of what work is, where it’s done and how it happens.’
Graham has been travelling the world to talk to people who find themselves in a new kind of labour market. In particular, he’s been interviewing individuals who perform work from home, provided to them by a slew of websites such as Amazon’s Mechanical Turk, UpWork, and ClickWorker. These sites all allow companies and individuals to outsource tasks: potential employers simply post a description of what they need doing to a website, then people interested in doing the work bid for it. The employer chooses someone to do the work, based on a combination of price, listed skills and ratings from previous employers; the worker carries out the task, gets paid, then moves on to another piece of work. The tasks being doled out vary — from transcription and translation to new kinds of work such as tagging images for artificial intelligence systems — but much of it is currently difficult or expensive to automate.
Technology has also created legions of new workforce members in more traditional sectors, such as transportation, hospitality, catering, cleaning and delivery. ‘There are increasingly more ways of commodifying bits of everyday life: using your car to be an Uber driver; your apartment to be an Airbnb host; your bicycle to be a Deliveroo rider; or your broom to be a Task Rabbit cleaner,’ explains Graham. This is what’s become known as the ‘sharing’ or ‘gig’ economy. Whether it’s Uber, Airbnb or Amazon’s Mechanical Turk, the business plan is much the same: create a digital platform which makes it easier to link a customer, who wants a service to be performed, with someone who’s willing to provide it, for a (very) competitive fee.
These new styles of working certainly bring some benefits: apparent flexibility for workers, more efficient use of existing resources and equipment, and reasonable prices for those seeking services. But, as Jeremias Prassl, an Associate Professor of Law and Fellow of Magdalen College, warns, this new workforce is potentially vulnerable. ‘Uber acts like an employer: it sets your wage, tells you the route to drive, hires you, and fires you if your rating falls too low,’ he explains. ‘Under any classical analysis, Uber performs all the usual employer functions. But in its contracts with “driver-partners”, the platform explicitly denies employer status, suggesting that the worker is very much a contractor. Legally, and through the language it uses, Uber tries to deny the fact that it offers employment.’ Through so doing, the company is able to avoid paying social security, pension contributions, redundancy pay and so on — all the usual rights an employee might benefit from.
Levelling the field
But Prassl, who’s written a book about the topic, points out that these kinds of contracts are nothing new. ‘From the perspective of an employment lawyer, zero hours contracts and the gig economy are old problems,’ he explains. ‘We’ve been grappling with the rise of so-called “non-standard work” for the last 30 or 40 years. It’s just that now they’re receiving more attention and sustained media coverage.’
The problem, as Prassl sees it, is that employment law is currently based on an old binary system. If you’re an employee you get rights — to, say, sick pay, notice of dismissal or paid holiday. But if you’re a contractor, you’re not afforded any of those rights. Employment law currently boils down to a simple question: How do you define whether or not someone counts as an employee? ‘What my research suggests is that maybe we should turn the problem on its head,’ he explains. ‘We could say instead: Who’s the employer?’
It seems like a subtle difference but, with the shoe on the other foot, he suggests crowd workers would be able to enjoy some kind of employment law protection. In this upended scenario, everyone could benefit from existing minimum standards like the minimum wage, working time regulations and discrimination protection, with their provision accounted for by whoever is legally deemed to be the employer. If companies failed to comply, workers could litigate employers in the knowledge that the damages were definitely owed to them.
It’s not just Prassl that’s worried about the vulnerability of employees. ‘One of the issues is that we confuse work with jobs,’ points out Ruth Yeoman. ‘There’s an awful lot of work in the world that has to be done, and one of the problems when we think about the future of work is how it all gets converted into jobs for which people will be paid. Sometimes people may contribute to society not through paid work, but through some other mechanism: voluntary work, say, or caring.’ And while those tasks may be hard work, or may not pay, they are necessary and many of them must be done by humans.
That’s why Stuart White (@StuartGWhite), Associate Professor from the Department of Politics and International Relations, is interested in how we could ensure everyone enjoys a basic standard of living — a concept he’s written about in the book Democratic Wealth.
‘The state could unconditionally pay every citizen an income, independently of their work status,’
he explains. White’s suggestion is that no tests of means or willingness to take a job would be imposed, so that everyone in the country received a basic payment every month. It’s worth noting that the idea is not intended to make everyone rich — far from it. Instead, it’s a means of giving individuals more flexibility, affording them power to decide when and how to be contributive and productive. ‘It’s a way of ensuring you don’t have people desperately scrambling into jobs to make ends meet,’ White explains. In turn, he argues, employers would make some of the least appealing jobs more pleasant — they’d be forced to, otherwise nobody would choose to do them.
Numerous mechanisms for putting such a policy into action have been proposed in the past. One option is to divert existing benefits and tax relief into a basic income that’s shared equally amongst the population. If those contributions didn’t stretch far enough, they could be topped up with revenue from further taxation — from land value tax, suggests White. Alternatively, the income could be provided by a state-owned investment fund from which the returns would be shared out equally. ‘There are lots of philosophical arguments about whether or not it’s all a good idea,’ he concedes. ‘But we’re moving into a world where there’s increased insecurity around work. Against that backdrop, a source of income that’s independent of work is a way of rebalancing power relations in the labour market.’
Whether or not you agree with the concept of a universal citizen’s income or the reform of employment law, these concepts are indicative of the kinds of discussions that Oxford researchers are increasingly leading. ‘I think the University needs to be asking these kinds of Aristotlean questions about whose interests are being met, who benefits from the changes… the moral questions,’ explains Marc Thompson. ‘It’s not something we should shy away from.’
Increasingly, then, just as Thompson hoped for when he set up the Green Templeton College Future of Work Programme, Oxford academics are working with business and governments to shape the debate about the future of employment. Frey and Osborne, for instance, have published reports with Citi and Deloitte about the impact of technology on employment; Mark Graham sits on the Department for International Development’s Digital Advisory Panel; and Richard Susskind acts as an IT Adviser to the Lord Chief Justice of England and Wales.
What remains, of course, is for policymakers, lawyers and industry officials to take the questions and suggestions raised by academics on board, then work out how best to use technological advance in all our favour. ‘These possibilities afforded by technology, automation and commodification of labour… they can all be shaped by policy, organisational change and simply choosing to do things differently,’ muses Thompson. ‘There are some important choices to be made about how we make use of them.’
Technology will make many jobs redundant, others easier, and create at least some new ones along the way. Keynes’ prediction of a fifteen-hour working week may even come true. But while humans are in charge, we can still choose for there to be some work that’s performed by non-robotic hands. ‘It would be very easy for there to be an automated pub where drinks are served from vending machines,’ concludes Mark Graham. ‘But nobody wants that. Because it would be depressing.’
Written by Jamie Condliffe, a science and technology writer based in London. He tweets @jme_c.
In keeping with one of the themes of the article we used 99designs to find an illustrator and worked with slouise.
Follow us on Medium, we’ll be publishing more articles soon that look at topics such as medical trials, developments in healthcare and more.
If you liked this article please click the green heart, it really helps to spread the word and let others find it.
Produced by Christopher Eddie, Digital Communications Office, University of Oxford.