Written evidence submitted by the RSA (AFW0039)

1. About the RSA Future Work Centre

1.1  The RSA launched the Future Work Centre in 2018 to explore the impact of new technologies on the world of work.[1] Against the backdrop of increasing concern about AI, robotics and digital platforms, we felt there was a need for further research to understand how these innovations could change both the quantity and quality of work, from how workers are recruited, to the degree to which they are monitored, to whether jobs are fragmented or cohesive. Using a combination of original research, scenario planning and hands-on Sector Labs, our ambition is to give employers, educators and policymakers the insights they need to prepare today’s workforce for tomorrow’s workplace.

1.2  The Future Work Centre is a partnership between the RSA, Taylor Wessing, Friends Provident Foundation, Google.org and a Fellow of the RSA.

1.3  A parallel initiative from the RSA is the Future Work Awards, which aims to identify and raise awareness of ground-breaking schemes from around the world that are improving the quality of work.[2] These range from modern lifelong learning programmes to innovative union models that serve gig workers to new HR practices that give workers maximum autonomy.

2. How can technology change the workplace?

2.1  Radical technologies including AI and robotics have become synonymous with automation. However, technology can affect workers in more ways than task and job displacement:

• Disintermediation – Where technology allows workers to connect directly with consumers, making it efficient to operate outside of traditional organisational models. Examples include gig platforms like Uber and Deliveroo (both powered by algorithms), which have allowed workers to find jobs 'on demand' and in small parcels. Another example is the e-commerce platform Etsy, which has enabled self-employed business owners to reach customers globally.

• Digitisation – Where technology turns physical goods and information into data that can be captured, shared and replicated at low to zero cost. Examples include news platforms that have digitised newspaper content, and search engines that have moved the process of searching for information online. One consequence is the loss of jobs in the production of physical goods. Another is the growing ease with which large companies can amass global customer bases, benefiting from the network effects of a more digital economy.

• Communication – Where technology enables workers to exchange information and collaborate more easily. Examples cover well known technologies such as smartphones, communication tools like Skype and Whatsapp, and infrastructure including cloud storage. Although on the surface rather mundane, these technologies have the potential to amplify the productive ability of workers while also threatening to erode their work-life balance.

• Management – Where technology shapes the coordination of work, including

recruitment practices, performance assessments, surveillance and monitoring in the workplace. Examples include AI-powered video software that can analyse job candidates during interviews, and scheduling software that can automate the planning of rotas based on expected consumer demand. These technologies could exacerbate bias, erode privacy and damage autonomy – or equally improve outcomes in these domains, depending on how the technology is deployed.

2.2  Mapping the full consequences of technology’s application in the workplace shows there are no simple conclusions to be drawn about the future of work. Yet even automation – a fifth force to add to those above – is itself multifaceted. At least four subtypes can be discerned:

• Substitution – The most conventional form of automation, substitution is where technology takes on a task that would usually be completed by a worker. Examples include self-driving cars, picking and packing machines used in warehouses, and software that can write news articles and company financial reports. 

• Augmentation – Augmentation expands the capability of workers, allowing them to achieve more and better-quality work. Examples include CAD software used by designers to produce higher quality images, robotic medical tools used by surgeons to make more precise incisions, and conventional search engine technology that allows researchers to find more relevant information. 

• Generation – As well as mimic what workers already do, technologies can generate tasks that are not currently done by humans (or only by a small number). ElliQ is a good example. This ‘elder care assistant’ can remind people to take their medicine, set up video chats with family and friends, and recommend physical exercises. Given very few people have carers on hand around the clock, this technology is not an immediate threat to workers.

• Transference – This is where technology shifts responsibility for undertaking a task from workers to consumers. Self-service checkouts, for instance, have not done away with the job of processing items through tills. Instead they’ve passed this responsibility on to shoppers. The same is true of most ticket or vending machines.

2.3  Automation is an outcome – the result of an action taken by machines – as opposed to a technology in its own right. Many commentators incorrectly conflate automation with technologies like artificial intelligence, when in fact the latter is an enabler of automation. Moreover, while AI and robotics continue to grab the media limelight, automation can be enacted through more basic technologies like traditional ‘expert systems’ software. The earliest factory machines that automated the production of cars, food products and white goods were not based on artificial intelligence. Nor are today’s self-service checkouts, which only use basic computing software.

2.4  Timelines are important. ‘The future of work’ is now a popular term used by policymakers, journalists and think-tanks alike, yet very few observers qualify their opinions with a definition of what future they are referring to. Whether it is 5 years from now, 20 years or 50 years will have a large bearing on what we might expect of technology, and whether indeed we can make any meaningful predictions at all. The RSA Future Work Centre is focused on the period leading up to 2030-35.

3. How will technology shape the workplace?

Job numbers

3.1  A central concern of policymakers and the media is that new technologies will lead to the mass automation of jobs. PwC expects 7 million UK jobs to be displaced over the next 20 years (albeit with 7.2 million added).[3] The Bank of England puts the figure closer to 15 million by 2035.[4] Another thinktank, IPPR, warns that automation could take away 13.7 million jobs over the coming years.[5]

3.2  The RSA’s view, however, is that mass automation is unlikely in the short to medium term.[6] One reason is that new technologies, while becoming progressively more powerful, are still limited in capability. Ocado’s sophisticated automated warehouse system still requires humans to complete delicate and dexterous tasks.[7] Google Translate’s algorithms continue to struggle with full passages of text.[8] IBM Watson’s healthcare software was recently found to have made several ‘unsafe and incorrect treatment recommendations’.[9] More broadly, there is concern among Silicon Valley experts, including Gary Marcus, that deep learning approaches to AI – the lead method for programming – is coming up against technical hurdles.[10] This includes only being reliable with extremely large datasets, which are not available for every use case.

3.3  A second reason is that automation is more often of tasks than whole jobs. And because jobs usually encompass a range of functions, the automation of one task means workers can often pivot into new roles. No machine can wholly substitute for retail assistants, doctors, hotel receptionists, warehouse workers or financial advisers. A study by McKinsey identified at least 2,000 different types of work activity across all occupations, each of which in turn demand a bundle of separate capabilities (from mobility to sensory perception to natural language generation).[11]

3.4  Third, and as noted above, automation comes in different forms – only some of which involve removing tasks from workers. Augmentation automation means that technology can help workers to achieve more and better work in a shorter space of time. In our research last year, we came across a domiciliary care service in East London – Three Sisters Care – which is working with Bristol Robotics Laboratory to design a ceiling-based modular robotics system to help care workers lift and carry patients.[12] The purpose is to give staff more time to do fulfil their core responsibilities: to put older patients at ease, help them do exercises, give them company and manage their health conditions.

3.5  A fourth reason to be sceptical of mass automation claims is that they seldom account for the macroeconomic dynamics of shifting consumer demand. When a technology is deployed that boosts productivity, the extra output per worker often results in cheaper products for consumers. While some of this is saved, much of it is spent either on the same good or another good elsewhere in the economy, thus spurring demand for labour. This phenomenon of ‘Recycled Demand’ is thought to be one reason why the introduction of legal software in the US in the early part of the 21st century did not lead to a reduction in the number of legal clerks: the software reduced the price of the service, leading more people to make use of it.[13]

Job quality

3.6  Each of these factors points to a near-medium term future of abundant jobs. The question, then, is what these roles will look like. Will they be high skilled or low skilled? Will they be well paid or poorly paid? Will they be purposeful and bring out the best in people’s talents, or will they be dull, precarious and closely monitored?

3.7  Looking first at the sectoral and occupational make-up of our labour market, much of the evidence suggests we are likely to see a growth in what Adair Turner describes as hi-tech and hi-touch jobs.[14] By hi-tech we mean jobs that involve the creation, maintenance and explanation of new technologies. By hi-touch we mean jobs in service and experiential industries, including care, education, entertainment and hospitality. According to PwC, the number of healthcare jobs is expected to grow by 22 percent over the next 20 years, professional, scientific and technical service jobs by 16 percent and education roles by 6%.[15] In contrast, manufacturing roles are set to shrink by 25 percent and transport and storage by 22 percent. With hi-tech jobs well compensated and hi-touch jobs generally poorly compensated, there is a danger that inequality in earnings will widen.

3.8  Yet this prediction of a bifurcating labour market may be too simplistic. In many cases, high-skilled workers will be worse off as a result of automation. For example, if new cancer-detecting algorithms are deployed in healthcare, it may lower the barriers to entry to becoming a radiologist (who will have a shorter training regime) and thereby reduce their power to bargain over wages. Equally, technology could boost the earning power of low-skilled workers in hi-touch jobs. If new robotics or AI technology allows care workers, physiotherapists or teaching staff to do more or better quality of work, these improvements in productivity may translate into higher wages (assuming the surplus value is passed on to workers). Accenture believe that AI applications on their own could increase labour productivity in the UK by 25 percent by 2035.[16]

3.9  Critically, workers do not need to be employed within productivity-growing sectors to benefit from wage rises. An economic concept called Baumol’s Cost Disease describes a process whereby rising productivity and wages in one sector, say manufacturing, lifts wages in adjacent parts of the economy, say among social care workers and beauticians, even if the latter have experienced no equivalent rise in productivity. This is partly because wages must rise across the economy to prevent workers leaving their jobs for the lead sectors, and partly because workers in the lead sectors have more spending power to channel elsewhere in the economy. This may explain why the pay of teachers and hairdressers has grown consistently over time despite the former teaching roughly the same number of pupils and the latter seeing the same number of clients as they did 50 years ago. Subtle phenomena such as Baumol’s Cost Disease and Recycled Demand are difficult to articulate and understand, yet no analysis on the future of work would be complete without them.

3.10          So far we have discussed wages, but what about working conditions in the round? While this inquiry is focused on the effects of automation, we would like you to consider how new technologies will alter the management of work (see the typology above). In recent years, several software companies have emerged to help employers recruit, monitor and manage their staff. This includes: Percolata, which draws on a combination of smart sensors and algorithms to build schedules for retail workers[17]; Mya, a recruitment tool that uses natural language processing (an AI capability) to engage with job candidates during recruitment rounds[18]; and HireVue, an AI-powered video tool that analyses candidates during job interviews.[19]

3.11          These technologies can be beneficial or damaging to worker wellbeing depending on how they are developed (or ‘trained’ as machine learning algorithms are) and deployed. There is a risk, for example, that recruitment algorithms which automatically sift through CVs could entrench bias within hiring decisions. On the other hand, they may remove discrimination by basing hiring decisions solely on the merits of individual candidates. The company Infor Talent Science claim their recruitment algorithm led to an average 26 percent rise in African American and Hispanic hires across the US industries where it was used.[20] The use of automated scheduling software is equally contentious. Software like Percolata could be criticised for removing the human touch from management decisions, or conversely praised for ensuring that rotas are set with indifference to workplace politics. 

3.12          A recent RSA/Populus survey (undertaken in June 2018) revealed several insights on worker attitudes towards new technologies[21]:

• 33% of workers think our terms of exit from the EU will lead to the most job losses, versus 27% who fear new tech entering the workplace
• Workers are divided on whether technology will lead to large job losses: 34% say that few of the jobs lost to automation will be replaced by new ones, while 28% say most of them will 
• Workers are equally concerned about how their jobs will change in nature and substance: 50% are concerned about tech leading to excessive monitoring and 39% worry about having less freedom to work as they would like
• Tech companies are overwhelmingly seen as the winners from tech entering the workplace (42% think so). Just 6% say workers have the most to gain, while 13% say consumers 
• Most workers say they lack a safety net in the event of their job being automated. Only 18% say the government could support them with living costs: 64% agree they would be in a situation where they would struggle to make ends meet (39% strongly so)
• Most workers say they lack a safety net in the event of their job being automated. Only 18% say the government could support them with living costs: 64% agree they would be in a situation where they would struggle to make ends meet (39% strongly so)

4. Are businesses adopting technology?

4.1  In the rush to understand how automation may affect workers, we often fail to ask whether our businesses and public services are automating at all. Separate polling undertaken last year by the RSA and YouGov – this time of business leaders – found that very few firms are making use of artificial intelligence and robotics. Just 14 percent say they have already invested in AI and/or robotics, or soon plan to. A further 20 percent say they want to invest, but that it will take several years before they will ‘seriously’ be able to do so. 29 percent, meanwhile, are aware of these technologies but believe they are either too costly to deploy or have not been properly tested.

4.2  Our survey findings mirror other data on technological adoption and business investment. While there is no available data on the purchase of AI systems that we know of, information on robot sales are helpfully collected by the International Federation of Robotics. Their data reveals that the UK purchases significantly fewer robots than France, Germany, Spain and Italy.[22] In 2015 the UK had just 10 robot units for every million hours worked, compared with 131 in the US, 167 in Japan and 133 in Germany.[23] The UK has a bad record in the adoption of even the most basic innovations. A recent ONS survey found that just 19 percent of UK workers feel their job has changed as a result of new software being introduced to their workplace in the last 12 months.[24]

4.3  Several factors could account for low rates of technological adoption. One is cost: in many sectors, especially low-skilled and low-paid ones, it remains cheaper to rely on human workers than to invest in technology. In farming, a soft fruit harvesting machine can cost as much as $250,000.[25] In social care, several companies are developing machines that can lift and carry patients, yet one – from the company RIKEN – costs between $168,000 and $252,000.[26] Other barriers relate to internal organisational inertia and risk aversion (although this may be warranted). Regulation can also stymie the take-up of technology. New GDPR rules, for example, ask more from businesses that used automated decision-making algorithms, which may translate into extra costs.

5. What should the government do about automation?

5.1  The RSA believes that low rates of technological adoption – be they of artificial intelligence, robotics, digital platforms or more prosaic innovations – will not lead to beneficial outcomes for UK workers or businesses in the long term. While our economy may be spared immediate disruption, if we do not act to accelerate automation our prosperity will weaken overtime and our country will become poorer relative to others. UK businesses that trade internationally risk becoming uncompetitive, with resulting job losses. Moreover, UK markets risk becoming more concentrated, as only the largest companies innovate and extend their lead in old market while entering new ones. This would not be a good outcome for workers, consumers or the diversity of our economy.

5.2  Equally, however, we cannot leap into the unknown and automate without due care and planning. The risk is that we exacerbate inequalities and entrench an already endemic problem of economic insecurity for millions of workers. We have a window of opportunity to bring about automation on our own terms – to adopt the right kind of technologies, to equip workers with the necessary skills to operate alongside them, and to make sure that the wealth created by these technologies – the profits – are shared fairly. With only 7 percent of workers believing they have the most to gain from the introduction of new technologies in the workplace, it seems our existing institutional set-up is not fit for purpose.[27] Our education, tax, welfare and regulatory systems all require reform.

5.3  Before jumping to recommendations about how the government should prepare workers and businesses for a more automated economy, first it is important to think about broader principles for managing risks. We call this set of principles a ‘social contract’, which we believe is in urgent need of renewal.[28] The RSA Future Work Centre will spend the coming months unpacking the different elements of this contract, but already we can suggest some of its features:

• Shared responsibility – It is not the task of the government alone to manage the effects of automation. Employers, investors and even consumers must be willing to intervene or have interventions made on their behalf. Investors in technology, for example, should be more mindful of the companies they invest in and use their power to influence how technologies are developed (e.g. encouraging creators of recruitment software to test their tools for bias). Consumers, who gain considerably from automation in the form of better and cheaper goods, may be asked to contribute more in the form of a transaction tax to support a stronger welfare system.

• Marrying flexibility with security – One reason the UK has the lowest unemployment rate in a generation is because of our employment regulation, which reduces the barriers to hiring workers. To the extent that automation leads to job losses, this flexibility will be essential in smoothing the transitions that workers need to move between one job, occupation or sector to another. However, this flexibility needs to be balanced with security, including adequate access to unemployment insurance, sick pay and pension coverage. Security can also be offered through a fairer tax system, one that taxes all forms of productive activity equally, be it from labour or capital – which is currently not the case (see, for example, differences in how the self-employed, employees and company directors are taxed).

• Learning to live with ‘low-skilled’ work – New technologies will spur the creation of new jobs within or serving hi-tech industries, but a large proportion of our labour market will always be made up of what is deemed to be low-skilled work: caring, cleaning, waitressing, delivery driving, and so on. Rather than always aiming to save people from this type of employment, a better response to automation would be to make this work better. Everyone should be able to describe themselves as a professional with a career pathway, developing within their roles not just moving out of them. One means of achieving this is to trial more forms of occupational licensing. Another is to establish personal training accounts (in place of the Apprenticeship Levy) that would give workers the funds and inspiration to pursue lifelong learning.

• Putting owning on a par with earning – Evidence suggests that automation, in its aggregate effects, may be shifting national income away from labour towards capital (i.e. the owners of the technology). Should this trend continue, it poses the question of whether having a job alone is enough to make ends meet. According to an RSA poll, already those who say they are ‘just about managing’ live in households with gross incomes above £31,000.[29] Government policy should put owning on a par with earning to boost people’s economic security, for example by giving people a literal stake in the companies that are becoming ever more profitable. The RSA recently laid out plans for a Universal Basic Opportunity Fund, which would be established in the form of a Sovereign Wealth Fund and pay out dividends to every citizen.[30]

5.4  These principles encourage us to think more expansively about how we can manage the effects of automation, however widespread it may be. For the full effects of automation to be managed, we will need to intervene upstream in the technological lifecycle, not just downstream. Some interventions can be enacted today, while others will need time to build political support for. Yet that doesn’t mean they are out of reach. Over the coming months the RSA will put forward further ideas and insights on this debate. We look forward to hearing the ideas of others who have submitted evidence to the BEIS inquiry, and of the inquiry research itself.

August 2018