How will automation change the nature of work?

We explore how new machines could impact the quality of work through 5 dimensions of automation

27 min readSep 17, 2017

By Benedict Dellot and Fabian Wallace-Stephens

Follow Benedict and Fabian on Twitter @BenedictDel @Fabian_ws

This article is an extract from the RSA report The Age of Automation: Artificial intelligence, robotics and the future of low-skilled work

Whether it is deep learning algorithms that can diagnose cancers in healthcare, machine learning software that can speed up recruitment practices in HR, or mobile droids that can deliver parcels to and from distribution centres, there is a sense among many that new machines are set to usurp humans from the workplace.

But of course, this is not the first time automation has drawn breaths of anticipation. Beginning with the industrial revolution in the late 18th and early 19th centuries, the world has seen a steady stream of new technologies come into play that have altered the nature of work. The spinning jenny, the flying shuttle, interchangeable parts, assembly lines, agricultural mechanisation and the personal computer are just some of the inventions that have replaced human labour over the years.

The American professors Eric Brynjolffson and Andrew McAfee draw a useful dividing line between two broad periods of automation: a first machine age where new technologies substituted for physical tasks, and a second machine age where cognitive activities became the focus of automation.

In this article, we explore how automation might play out in the future, drawing upon historical experience to give us clues. In doing so, our analysis extends beyond the traditional focus on job quantity to encompass matters of job quality. We believe the latter is a significant omission in recent studies on automation, with too many commentators fixated on estimating the number of jobs that could be displaced, at the expense of understanding how the remaining jobs will evolve. Not only are automation estimates an almost impossible exercise to get right, they distract observers from the way new technology will transform other aspects of the worker landscape, such as recruitment practices, wage growth, worker bargaining power, productivity levels, the degree of monitoring at work, and exposure to dull, dirty and dangerous tasks.

5 dimensions of automation

In 2004, the US academics Frank Levy and Richard Murnane published a book called The New Division of Labor, which speculated about the kind of tasks and jobs that might be off limits to machines. In the second chapter entitled “Why People Still Matter”, they suggested that self-driving trucks may one day be able to operate in structured environments, thanks to cameras and sensors that can capture sensory input. But they were doubtful autonomous vehicles would ever cope with busy streets or a task like turning in heavy traffic.

“Articulating this knowledge and embedding it in software for all but highly structured situations are at present enormously difficult tasks… Computers cannot easily substitute for humans in [jobs like truck driving]”
Frank Levy and Richard Murnane, The New Division of Labor (2004)

Levy and Murnane are two brilliant economists, but their inaccurate forecast reminds us to be careful when trying to ascertain what can and cannot be automated. Indeed, thanks to the developments such as machine learning and cloud robotics (outlined in What is the difference between AI and robotics?), new machines are beginning to encroach on non-routine activities that were once thought the sole preserve of humans. Amazon’s robots can scurry pallets around busy warehouses, soft gripping robots can pluck and bag delicate fruit harvests on farms, and machine learning algorithms can identify breast cancer with 89 percent accuracy — a better record than most pathologists. Faced with examples like these, it is hard not to agree with the likes of Martin Ford that humans are destined for the scrapheap.

Yet there are several reasons to be sceptical of such claims. The first is that there are still many things that machines cannot do, and which seem unlikely for them to master in the near future. Artificial intelligence and robotic systems struggle with open-ended conversations, they do not have hunches, they lack taste and cultural awareness, they have no strategic vision and cannot lead, opinions are beyond them, they cannot improvise, and by their very nature they are not authentic. While an IBM computer was able to beat Karry Kasporov at chess in 1997, there are still no machines that can ably tie a shoe lace, open doors, or turn the page of a book.

Researchers at Berkley University found it took their towel-folding robot ten hours to produce a stack of twenty-five towels (a new commercial robot called Laundroid takes three to ten minutes to fold a single item). This is what has become known as Moravec’s Paradox: high-level reasoning requires low computation, but low-level sensorimotor skills demand huge computational resources.

A recent study by McKinsey canvassed the views of technology experts to identify which work activities, out of a set of 2000, have the greatest scope for automation. High on the list were tasks that hang on processing or collecting data, and operating machinery in predictable environments. In contrast, tasks such as interfacing with stakeholders, being creative, planning, managing and developing people, and applying expertise to decision-making are all estimated to be significantly less susceptible to automation. These findings are roughly mirrored in Carl Benedikt Frey and Michael Osborne’s seminal University of Oxford study, which identified three groups of tasks — or ‘engineering bottlenecks’ — that remain difficult for machines to master:

Social intelligence — The ability to negotiate and persuade, to respond to the emotional cues of others, and to impart knowledge
Complex manipulation — The ability to deftly handle, move and control objects in a variety of settings, using fine muscle control
Creativity — The ability to conceive of novel ideas and to create art and design that pushes cultural boundaries

Some dispute the idea that competencies like creativity and social intelligence are beyond the remit of machines. Historian Yuval Noah Harari points to software that can write symphonies exquisite enough to dupe listeners into believing they are human made. Others point to the robo-therapy bot Ellie, which has been designed by researchers at the University of Southern California to study facial expressions and voice patterns, and elicit responses from patients. Yet many of these technologies offer only shallow forms of automation. ‘Deep creativity’ means pushing the boundaries of artistic endeavour rather than replicating symphonies of the past, while ‘deep communication’ means not only disseminating knowledge but helping others make sense of it, as well as engaging in open-ended and imaginative dialogue.

A second reason why the extent of automation may be less severe than some expect is because jobs are multifaceted and made up of a basket of tasks, only some of which are automatable. Hotel receptionists have to meet and greet customers, pick up and drop off keys, check bookings on a computer system, and move heavy bags through busy walkways. It is feasible that one of these tasks could be computerized, for example through automated check-in systems, but the result would most likely be an evolution of the job as new tasks fill the void. And even were all of these tasks automatable, it is difficult to see a single machine switching between them seamlessly as a human receptionist would. As one Nasa scientist put it, humans have an edge in being a “150-pound, non-linear, all-purpose computer system”.

The tendency of technology to automate tasks rather than whole jobs is often cited as a reason why the spread of ATMs in the 1990s did not lead to the significant loss of jobs in bank branches. Because cash dispensing was only one function of bank tellers, theory has it they were able to pivot into customer-facing roles and fulfil more unstructured tasks such as processing mortgage applications or answering queries about managing accounts and transactions. Looking at automation through the lens of tasks rather than jobs, the OECD estimate that only 10 percent of UK jobs are at risk of full displacement. However, they also calculate that a further 25 percent of jobs could see roles change considerably as AI and robotics chip away at routine functions.

The third reason to be sceptical about claims of mass job displacement is that technology results in more than just worker substitution (see below). AI and robotics will also complement what humans do, enabling people to achieve more and do better work. A good example is the use of ceiling-based modular robots in domiciliary care, where partially automated devices help overburdened social care workers lift patients out of bed and in sitting positions (see How AI & robotics are transforming social care, retail and the logistics industry a more detailed case study). Another example is the use of AI platforms in call centres. The airline KLM uses a chatbot to generate partial responses to customer queries on social media, which are then edited, refined and elaborated on by human agents in the team. 10 percent of messages are sent without alteration, most of them answers to basic questions.

Alongside substituting and complementing, AI and robotics will create work in their own right. This means machines fulfilling tasks that have never been done by a salaried human previously, or at least only by a minute fraction of the workforce. A case in point are the new concierge service bots coming onto the market for older people with low level care needs. ElliQ is an elder care assistant that can remind people to take their medicine, set up video chats with family and friends, and recommend physical exercises depending on how sedentary a person has been. Given that none but the wealthiest of individuals have carers on hand 24/7, this device cannot be seen as encroaching on human turf. Another example is the use of AI in agriculture to predict crop yields and determine where pesticide should be sprayed. These are not skills that farmers or their workers would ever profess to have, and thus machines should be seen in this context as creating value without displacing labour.

What does all of this mean for job availability in the round? Recall the predictions of various studies that have explored the consequences of automation. Frey and Osborne forecast that 35 percent of jobs in the UK have the potential be fully substituted, while the OECD put the figure at 10 percent and McKinsey closer to 5 percent. Most recently, PwC bucked the increasingly conservative stream of estimates with a higher prediction of 30 percent. The reason why these estimates vary is because of methodological differences, for example with McKinsey using their own raft of experts to review the capabilities of machines, and the OECD viewing automation through the lens of tasks rather than jobs.

Rather than try to repeat another extensive estimation exercise, we chose instead to speak directly with employers about the prospects for job automation. They are the ones who ultimately buy and deploy new technology, and are therefore better placed than experts to determine its impact in the short-term, at least within their own organisation. Our RSA/YouGov survey found that business leaders on average believe 15 percent of jobs in their organisation have the potential to be fully automated within 10 years. This is a significant proportion and on a par with PwC and Frey and Osborne’s studies, if we take into account their timelines were around twice as long. However, as the graph show below, we found wide variation among respondents, with a fifth (22 percent) saying they see zero prospects for job automation in their business.

Regardless of the overall figure of automation, most studies reveal a technological bias against low-skilled and low-paid workers. The OECD calculates that 44 percent of workers with less than a high-school degree hold jobs made up of many highly automated tasks, compared with 1 percent for the college-educated. Similarly, Deloitte estimate that UK jobs paying less than £30,000 are five times more vulnerable to displacement than jobs paying £100,000 or more. These figures are given greater impetus by the numerous media stories of new technologies emerging in low-skilled occupations, whether it is robots that can flip burgers, delivery drones that can automate parcel delivery, or automated floor cleaners operating in hospitals and hotels.

Business leader estimates of the proportion of jobs in their organisation that could be fully automated within ten years:

Source: RSA/YouGov survey of 1,111 UK business leaders (Fieldwork conducted 10th-18th April 2017)

Share of employers who think a high level of jobs in their organisation will be automatable in the next 10 years, by sector:

While these are gloomy forecasts for low-skilled workers, it is important to note that some sectors will be more affected than others. In nearly every study of AI and robotics, including our own, two low-skilled sectors emerge as likely to bear the biggest brunt of automation: retail and logistics. The chart above shows that 15 percent of retail business leaders see considerable potential for job automation in their organisation (defined as more than 31 percent of jobs being automatable in the next 10 years), as do 21 percent of business leaders in transportation and distribution. In terms of middle and high-skilled jobs, both finance and manufacturing are seen as having high numbers of jobs that could be displaced by machines. Indeed, a recent US study by two MIT economists estimates that the deployment of one extra industrial robot (i.e. in manufacturing) per thousand workers reduces the employment to population ratio by 0.18–0.34 percentage points.

Yet the picture is markedly different for sectors that are bound up in the delivery of experiences and person-to-person interaction. Just 4 percent of business leaders in hospitality and leisure, 2 percent in medical and health services, and 3 percent in education see the scope for high automation among their workforce (although these last two figures should be interpreted with caution given low sample sizes). This echoes Frey and Osborne’s research, which finds many low-skilled but ‘human-centric’ jobs have little to fear from machines. Mental health and substance misuse social workers rank 4th of 702 on their scale of automation-proof jobs, while healthcare social workers rank 8th. The number of human-centric jobs is already growing at pace in the UK. Primary and nursery teaching professionals are up by 40 percent since 2011, educational support assistant are up 50 percent, and nursery nurses and assistants are up 25 percent.

Why are human-centric jobs less automatable? Because they require competencies that are currently very difficult for machines to replicate, such as empathising, forming authentic relationships and communicating in open ended dialogue. Each of these sits under the broad skillset of ‘social intelligence’, which as noted earlier has been identified as a bottleneck for AI and robotics engineers. The chart below shows the relative importance of this skillset to different low-skilled occupations, alongside creativity and manual dexterity. See the Appendix for a full explanation of our methodology.

Importance of “bottlenecks to automation” to selected low-skilled occupations:

New jobs and recycled demand

An analysis of the impact of AI and robotics on job numbers would not be complete without considering the macro ramifications, beyond changes at the firm level. One of these is the potential for new jobs to emerge as a result of the arrival of these new technologies. Many of these will be directly related to AI and robotics, such as roles in monitoring and repairing technology, engineering machine learning and deep learning algorithms, or reviewing and improving cybersecurity measures. Some of the fastest growing occupations in the UK are tech-centric, such as programmers whose number has risen by 40 percent since 2011, and IT directors, which have more than doubled in number over the same period (RSA analysis of Labour Force Survey).

Many commentators are understandably dubious about the claim that new occupations will emerge to replace those that dwindle in size and importance. An investigation in 2013 by PwC found that just 6 percent of all UK jobs that year were of a kind that did not exist in 1990, while an OECD study found that only 0.5 percent of the US workforce is employed in digital industries that emerged during the 2000s. However, some new occupational types have gained a significant foothold in the labour market, such as IT business analysts and systems designers, the number of which has shot up by 31 percent between 2011–16. Furthermore, AI and robotics are likely to create new tasks that will subtly reinforce jobs, such as personal trainers using sophisticated monitoring software to create tailored fitness regimes.

We should also recognise the multiplier effects of job creation. While it is true that new tech-centric occupations and sectors are unlikely to replace all the jobs lost to AI and robotics, their creation will spur the formation of additional jobs in ancillary sectors to serve their needs. The Berkeley economist Enrico Moretti estimates that one additional job in the ‘tradeable’ industries of a given US city results in 1.4 jobs in the local non-tradeable industries (where ‘tradeable’ means products and services that can be exchanged over distances). Because they command higher earnings, every new job in the tech sector is estimated to generate 5 complementary jobs elsewhere. Note that this multiplier dynamic plays out strongest in city areas.

Alongside new jobs, a second macro ramification of AI and robotics will be shifting or ‘recycled’ demand. This is a well-documented phenomenon whereby rising productivity caused by new machines leads to a lowering of consumer prices, thereby freeing consumers to buy more of the product in question or to spend money in another part of the economy. One of the best examples of recycled demand can be found in the transformation of the 19th century garment industry. It is estimated that 98 percent of the labour required to weave a yard of cloth was automated as a result of new technologies, yet the number of textile weavers actually grew for a period because prices fell and demand was elastic. The same effect played out after the introduction of ATMs in the US, which reduced the cost of running branches and bank services, leading banks to open more branches and take on new staff.

There is no reason why the same phenomenon will not occur in the wake of AI and robotics automating jobs and tasks. A case in point comes from the legal industry. Despite many fearing that AI will shrink the number of entry-level legal jobs, an investigation by The Economist found that the number of legal clerks in America grew by 1.1 percent on average per year between 2000 and 2013. The authors speculate that the introduction of software, which was capable of analysing large volumes of legal paperwork, led to falls in the cost of legal services, which in turn raised demand for legal clerks. In the same vein, robo-advisory services may heighten demand for financial advisors, while machine learning-powered health diagnostic systems may counterintuitively lead to greater demand for health practitioners.

Recruitment

Just as new machines will affect the number of jobs available in the future, so too will they alter the way people access that work. Several start-ups have developed software aimed at transforming who and how organisations recruit. Arya uses algorithms to source potential hires partly based on their social media history, Entelo applies machine learning techniques to spot individuals who may be on the cusp of switching jobs, and Ideal.com has developed shortlisting software that can screen candidates’ CVs based on the role and requirements of the employer. Last year, the consultancy firm Deloitte began using a new algorithm to tap into a more diverse talent pool. Alongside academic results, the system will take into account entrenched obstacles candidates have faced, such as growing up in a deprived area.

Other tools focus less on identifying and screening candidates, and more on streamlining the recruitment process. Mya, for example, is a chatbot designed to engage with job applicants before and after interviews. Using natural language processing and generation, it can answer a host of different questions applicants may have, raise final but important queries from the recruiter (“Can you remind us whether you have line management experience?”), and schedule interviews with minimal human oversight. Mya’s founders say it can automate up to 75 percent of the recruitment process. Another innovative tool is qdroid, which is used by Google to automatically draft interview questions based on the attributes it calculates are pertinent to the job. Again, this is based on historic data about the characteristics of successful previous hires.

The attraction to employers of using these machines is clear: they promise faster hiring times, lower costs and better job matching. But what do they mean for workers? One concern is that shortlisting software may exacerbate biases if it is trained on data that reflects previous hiring decisions. A Carnegie Mellon study looking at the use of algorithms in job adverts found that men were significantly more likely than women to be shown adverts for highly paid jobs when browsing Google’s internet search engine. On the other hand, algorithms could eliminate prejudice if they are tuned to give weight only to the qualifications and experience of candidates, rather than their age, gender or class. One AI tool called RAI is expressly designed to help employers reach their diversity targets.

To focus solely on traditional HR practices, however, would be to ignore the way AI is changing the very meaning of recruitment and what it means to be employed. The expansion of the gig economy — where people find atomised tasks through online platforms and apps — has only been made possible thanks to increasingly sophisticated algorithms. Uber, for example, relies on AI to predict hotspots of passenger demand, while Deliveroo depends on it to orchestrate the complex pick up and delivery routes of its riders. The rapid rise of gig working patterns, which the RSA estimates 3 percent of the UK workforce are now involved in, has been a major point of contention, with many fearing that workers are being exploited (Good Gigs: a fairer future for the UK’s gig economy). This may well be true for some, but we should also recognise how these platforms and the algorithms underpinning them have made it easier to access work and on hours of people’s choosing.

Pay

What about the impact of AI and robotics on pay? To the extent that automation leads to job losses, it will of course wipe out people’s pay packets. But it may also reduce the wages of those who remain in work. In a survey undertaken earlier this year, a third of US experts agreed that ‘IT and automation’ are a central reason why median wages have been stagnant in the US over the past decade. Only 20 percent disagreed. Technological alarmists point to the basic laws of supply and demand in setting wages. AI and robotics, they say, will flood the labour market with a cheap supply of mechanical labour, which will in turn reduce the bargaining power of human workers. American mathematician and philosopher Norbert Wiener observed as much in the 1950s:

“Let us remember that the automatic machine is the precise economic equivalent of slave labour. Any labour which competes with slave labour must accept the consequences of slave labour”

Furthermore, new machines may de-skill occupations, thereby lowering the barriers to entry and reducing the negotiating position of workers in existing jobs. Higher skilled professionals are likely to bear the brunt of this disruption. For example, deep learning algorithms capable of detecting cancers may enable lower skilled nurse practitioners to complete diagnoses that usually take radiologists a decade to train for, with the latter losing out as a result. Yet AI and robotics may also serve to de-skill already low-skilled jobs. In its recent investigation of the warehousing industry, the LA Times reported that Amazon workers were now surrounded by machines giving them precise instructions for every manner of task, reducing the scope and need for initiative. This includes scanners that tell workers how big a size of box to use, and small machines that produce exactly the right amount of tape for packing.

On the other hand, there is evidence that AI and robotics could lead to a boost in wages — not least because of productivity growth, which generates more absolute wealth that can be shared with workers. A 2015 study looking at the use of robots across 17 countries found they raised labour productivity by 0.36 percentage points annually over the period 1993–2007. They also lifted wages and total factor productivity. Overall, this productivity boost was equivalent to the contribution of steam technology between 1850 and 1910. Looking forward, McKinsey estimate that automation from AI and robotics could raise productivity growth globally by 0.8 to 1.4 percent annually. Another consultancy, Accenture, believes that AI could increase labour productivity in the UK by 25 percent by 2035 (Why Artificial Intelligence is the future of growth, PDF).

There is good reason to believe low-skilled workers will benefit just as much as high-skilled workers. Indeed, what makes the latest advances in AI and robotics distinct from previous innovations is the sectoral breadth of their application, going beyond manufacturing (where productivity has been rising for decades) to low-skilled sectors that have historically suffered sluggish productivity growth, such as care and retail. A recent IPPR investigation found that low skilled sectors in the UK — including retail, accommodation, food and admin services — are 29 percent less productive than the economy as a whole, and are also less productive than their equivalents in Western Europe. Assuming new technologies can raise productivity in these laggard sectors, whether through robots in social care or self-service checkouts in retail, then low-skilled workers will be set to benefit, so long as employers share the gains and there is significant support for in-work learning.

A final consideration when thinking about the impact of new technology on pay is Baumol’s cost disease. This refers to the phenomenon whereby productivity and wage rises that occur in one sector of the economy can lead to wage rises in another, even if the second sector has experienced no equivalent productivity growth. The reason is because wages have to rise across the economy to prevent workers leaving their jobs for the lead sectors, and partly because workers in the lead sectors have greater spending power to channel elsewhere. This is why the pay of teachers and hairdressers has risen throughout the post-war period, despite the former teaching roughly the same number of students and the latter serving the same number of clients as fifty years ago. Thus, even if low-skilled sectors do not see dizzying productivity gains as a result of AI and robotic adoption, continued productivity growth in high-skilled sectors like advanced manufacturing and finance should lift wages across the board.

Hollowing out and progression

Most economists agree that recent waves of technology, combined with the offshoring of manufacturing activity to East Asia, has led to a hollowing out of the UK labour market, with middle-skilled jobs such as machinists, secretaries and administrators falling as a proportion of the workforce. Most also agree that there has been a subsequent ‘filling in’ thanks to jobs growth elsewhere.
However, there is a risk that within individual occupations and sectors, the automation of middle-tier jobs could remove rungs on progression ladders. Both the retail and accommodation/food services industries already have limited progression prospects, owing to a large number of entry level jobs and small number of management positions. Further research is required to understand the impact of automation on occupational mobility.

Experience

Worker experience is another domain often overlooked in popular commentary on automation. Yet the quality of work is no less important than the quantity of it. For some, AI and robotics will lead to a kind of ‘digital Taylorism’, with employers using new tools to relieve staff of responsibilities and control the minutiae of their day to day tasks. Recall the story of the warehouse workers whose initiative had been compromised by machines that automate microscopic decisions, down to the types of boxes they use. In logistics, predictive algorithms are being used to direct the routes of delivery drivers metre by metre, while in retail, a Silicon Valley startup called Percolata is using a combination of shop sensors and sophisticated algorithms to calculate the performance metrics of individual workers and apply this is to create store schedules with an optimal mix of staff. The company says its algorithm can boost sales by 10–30 percent.

Artificial intelligence could also lead to an unhealthy degree of monitoring in the workplace. Existing technology already allows for a degree of surveillance, for example with GPS systems in cars and RFID tags on worker clothing that can be used to track the whereabouts of staff. But AI may take this to another level of intensity. It was reported earlier this year that ‘sociometric badges’ powered by machine learning are being used by employers to analyse the speed, tone and volume — but not the content — of their employees’ voices, with a view to analysing workplace interactions. Another start-up, Veriato, has developed software to log staff behaviour on office computers, including browsing history, email messages, keystrokes and document use. This data is then crunched by an AI system to create a productivity baseline for the company and flag where individuals may be performing poorly.

Yet just as with our discussion on recruitment and pay, AI and robotics could quite as easily be a boon for worker experience. First and foremost, these technologies could humanise jobs, phasing out mundane tasks and opening space for more intellectually stimulating work. McKinsey estimate that only 2 percent of the average worker’s time is spent on creative tasks, and 9 percent on social and emotional reasoning. In contrast, 67 percent is spent on ‘recognising known patterns’, which is hardly the makings of a fulfilling job. Indeed, while we must take seriously the risk of technological unemployment and disruption, there are some types of work that we should not mourn the loss of. As the CEO of one robotics company put it, “Does anyone write on their resume that they’re skilled at walking down hallways without bumping into things and they know how to ride elevators?”

Leslie Willcocks at the LSE has been studying the impact of automation at a firm level for several years, and his findings are worth heeding. In one investigation, he reported on how Associated Press had deployed new software to automate corporate earnings reports, allowing the company to produce 4,700 reports per quarter, up from 300 when humans wrote them. But rather than feeling threatened by these machines, the company’s journalists were ‘positive about the reframing of their job responsibilities’ away from mundane, highly-structured assignments. In another study, Willcocks looked at the experience of a major gas and electricity utility, which had installed software to verify meter readings submitted by household residents. This led to a quarter of back office admin being automated, with humans left to work on the ‘really unusual’ reading cases that required more investigation.

While Willcocks focused on white collar workplaces, the scope for AI and robotics to humanise low-skilled jobs may be just as expansive. Algorithms in healthcare could allow entry-level nurses to play a more active role in diagnosis, semi-autonomous trucks could lower accident rates for HGV drivers (assuming humans are still behind the wheel), and robots in social care could allow caring staff to spend more time comforting patients and less time lifting them and preparing their meals. Over the years our labour market has shifted away from agriculture towards manufacturing and then onto services, and at each point more workers have been relieved of the three ‘ds’ of dull, dirty and dangerous jobs. There is every reason to believe this trend will continue with further advances in AI and robotics.

Finally, it is worth considering how new machines might lend greater agency to workers, in the sense of having more control over how they access and manage work. Two promising uses of AI stand out in this regard. The first is French-based Bob Emploi, a new AI platform that uses anonymised public employment data from the French government to deliver custom recommendations to job seekers so they can improve their job search strategy. The platform is planning to install a new ‘skills recommendation’ feature that will recommend which skills are likely to help job seekers land roles in specific industries and occupations. The other platform is WorkIT, which uses the power of IBM’s Watson computer to help workers of Walmart find out about their rights and the policies of the supermarket. The Resolution Trust and Bethnal Green Ventures’ partnership on ‘WorkerTech’ is exploring similar solutions to empower UK workers.

Consumer power

The fifth dimension in the landscape of automation is consumer power. AI and robotics will not only affect people at work but also in the home — as customers, patients, learners and political citizens. The experience of history tells us that technological advances more often than not supercharge living standards.

In the last 250 years, global income per head has grown ten-fold, while in the most advanced economies it’s closer to a twenty times increase. If we take into account technology’s impact on the quality of the goods we consume, real income per head is estimated to have grown by anything between 40 and 190 times. Technology’s effects on living standards were particularly noticeable in the postwar period, when television ownership went from 19 percent of households in 1955 to 96 percent in 1975, and when washing machine ownership jumped from 18 percent to 70 percent over the same period.

AI and robotics are almost certain to sustain this trend. For example, according to Boston Consulting Group, the operating cost of a robot welder in the car industry has plummeted to $8 an hour, versus $25 an hour for human welders. As the industry uses more robotic welders to produce cars, these savings are likely to be passed onto consumers in the form of lower prices. The Bank of America estimates that advanced robotics and AI could cut labour costs by 18–33% across all industries by 2025. Recall also the bricklaying robot (SAM) which can lay up to 4 times as many bricks as the average human bricklayer. Deployed in the right way, this could speed up home building and possibly reduce prices for homebuyers.

In other cases, AI and robotics will open up goods to people that were once out of reach. Robo-advisory services in finance, for example, are cheap enough to be used by most high street savers, unlike traditional financial advice which comes with an average price tag of £150 per hour.

AI and robotics could be equally transformative for the delivery of public services, with gains being felt both in cost savings and quality improvements. Whether it is DeepMind’s partnership with Moorfields Eye Hospital to improve detection of common eye diseases, or IBM Watson’s work with cancer centres to provide more tailored drug treatments, the scope for new machines to transform health outcomes appears vast.

In education, too, AI promises to amplify the work of teachers and trainers. Knewton is a new tool that helps teachers create tailor-made lessons for every student, by monitoring how they respond to different content and learning materials — be it games, videos or literature (see Knewton). Local authorities are also set to gain. Aylesbury Vale Distict Council is today trialling Amazon’s Alexa personal AI assistant as a new way for local residents to make requests, such as setting up council tax payments or applying for business permits.

Then there is the prospect of greater leisure time. The first household machines emancipated people (predominantly women) from home-based duties. One estimate suggests that the time spent on household chores fell from 58 hours per week in 1900 to 18 in 1975. It may be that the latest domestic robots reduce this figure even further. According to the International Federation of Robotics, 3.7 million such machines were sold in 2015, including for vacuum cleaning, lawn-mowing and window cleaning.

But it is in terms of working hours where the greatest opportunity for time-saving lies. Contrary to popular belief, the number of annual hours worked per employee has fallen in the UK since 2000, as it has in most developed countries (see the OECD’s statistics on average annual hours actually worked per hour). While AI and robotics may not bring the working week down to 15 hours — as John Maynard Keynes once speculated — it holds out the hope of workers gaining at least some extra leisure time, assuming we make the right choices as a society.

A matter of choices

In this article we have reviewed the potential impact of AI and robotics on the workforce, with a particular slant towards the low-skilled. Our findings suggest that, while a significant proportion of jobs could be fully displaced by new machines (15 percent of private sector jobs over the next 10 years, according to our YouGov poll), grim predictions of mass automation and widespread economic strife do not stand up to scrutiny.

Machines are still incapable of performing many tasks, and very few can comprehensively automate whole jobs. Occupations are more likely to evolve than be eliminated, and new ones will emerge in the long-run. Low-skilled workers will probably face the greatest disruption, but sectors vary significantly in their automation potential and we are likely to see a continued growth in ‘human-centric’ roles in health care, social care and education.

What is less clear is how AI and robotics will modify the quality of work, in terms of the other dimensions we considered. As we have seen, these technologies could lead to greater productivity, open up the door to higher wages, phase out mundane work in favour of more intellectually stimulating vocations, and create a level playing field in terms of recruitment — for the low-skilled as much as anyone else. Yet these technologies could just as easily be used to de-skill jobs, strip workers of their bargaining power, put downward pressure on wages, amplify monitoring and standardisation of work, and bake biases into recruitment. In this article we have documented the experiences of Amazon warehouse workers, Deliveroo riders, Associated Press journalists and Walmart supermarket staff — each of whom has engaged with technology on different terms and with different outcomes.

The point is that technology is not predetermined to achieve a particular result. Algorithms and robots do not have objectives of their own, but are directed by humans. Indeed, the sense of technology being a passive tool to be wielded as its owners see fit is possibly one reason why a high proportion of business leaders in our survey said they neither agreed nor disagreed that technology would lead to particular consequences; they rightfully conclude that nothing is guaranteed (see below, although note that business leaders were asked about technology overall, which includes but is not limited to AI and robotics).

The good news, therefore, is that as a society we have a choice in how to apply AI and robotics and manage their effects. There are choices to be made by developers and engineers in terms of the functionality they imbue in machines, there are choices to be made by employers as to which technologies they purchase, there are choices to be made by HR teams as to whether and how they help staff evolve into new roles as machines take on certain tasks, and there are choices to be made by policymakers about the kind of regulatory, welfare and tax system that can maximise the upsides of disruption and minimise the downsides.

Business leader attitudes about the impact of technology on work: