THE AUTOMATION ADVANTAGE How Australia can seize a $2 trillion opportunity from automation and create millions of safer, more meaningful and more valuable jobs. This report was commissioned by Google and prepared by AlphaBeta. The information contained in the report has been obtained from third party sources and proprietary research. 2 CONTENTS EXECUTIVE SUMMARY 4 1 AUTOMATION IS CHANGING THE WAY AUSTRALIANS WORK 7 1.1 Over the next 15 years, the average Australian worker will spend 2 hours per week less on manual and routine tasks 9 1.2 Automation will change the jobs we do, but it will mostly change the way we do our jobs 10 2 AUTOMATION IS A $2.2 TRILLION OPPORTUNITY FOR AUSTRALIA—IF WE GET IT RIGHT 12 2.1 Australia would gain $1.2 trillion from transitioning workers affected by automation 14 2.2 Australia would gain $1 trillion by accelerating the rate of automation 17 3 AUTOMATION WILL MAKE AUSTRALIAN JOBS SAFER, MORE SATISFYING AND MORE VALUABLE 19 3.1 Safer jobs, as machines take on dangerous physical tasks 20 3.2 More satisfying jobs, as machines take on routine tasks 21 3.3 More valuable jobs, as machines take on the least valuable tasks within each job 22 4 HOW AUTOMATION CAN BECOME A SUCCESS STORY IN AUSTRALIA 24 4.1 Policy should be targeted at different groups affected by automation 24 4.2 Lessons from abroad: how other countries are responding to automation 27 5 APPENDICES 31 5.1 Appendix A: Estimating timeshares of tasks in the economy 31 5.2 Appendix B: The impact of automation on work quality 33 5.3 Appendix C: Evaluating the potential gains from automation 36 5.4 Appendix D: Evaluating the impact of automation for different groups of workers 38 3 3 1 EXECUTIVE SUMMARY Technological change has long been a source of anxiety for workers. Today, improvements in communication technology, robotics, and machine intelligence are rekindling age-old concerns that technology will soon force millions of people out of work. This report provides a fresh perspective. Automation is, at its core, an opportunity for Australia to harness the power of machines to improve human lives. If we get it right, automation could significantly boost Australia’s productivity and national income—potentially adding up to 2.2 trillion Australian dollars in value to our economy by 2030. But this opportunity will not land in Australia’s lap. To unlock the benefits of automation we must be bold enough to lead changes. This means embracing technology’s potential to make our workplaces more productive, while taking steps to prevent Australia’s most vulnerable workers from sliding into unemployment. This report outlines how Australia can turn the trend of automation into a national economic success story. To understand the impact of automation on Australia’s economy this report analyses how automation changes the working life of every Australian. The use of machines is changing what jobs we do. Strenuous physical jobs are disappearing on factory floors, and routine administrative jobs can increasingly be done without human workers. On the flipside, more jobs are being created in community, personal and business services, and other specialised professions that rely on uniquely human skills such as thinking creatively and being able to understand other people’s emotions. What’s more, automation is changing the way we do our jobs. This report gives a comprehensive picture of the impact of automation on Australian workers by digging below the job level and analysing how technology is affecting the time that we spend on different work tasks within our jobs. In detail: every one of the 20 billion hours that Australians worked last year was assigned to one of more than 2,000 work tasks, creating a complete picture of how much time Australians have spent on each work task over a 15 year period. The results provide remarkable insights and allow us to understand likely future work patterns. Technology is already changing the nature of human job tasks. For example, retail workers are spending less time ringing up items at the register and more time helping customers; bank employees are spending less time counting banknotes and more time giving financial advice; teachers are spending less time recording test scores and more time assisting students; factory workers are spending less time on the assembly line and more time optimising production and training other workers. 4 Over the past 15 years alone, Australians have reduced the amount of time spent on physical and routine tasks by 2 hours each week. Most of that change isn’t coming from the loss of physical and routine jobs. Rather, it comes from workers doing the same jobs but switching to different tasks, as machines take over an increasing load of the repetitive routine work. Automation isn’t a force we can stop. But Australia’s economy has a lot to gain if we manage to avert the employment risks that come with growing machine use. To unlock the full amount of gains, two conditions need to be fulfilled. First, Australia requires a strong policy framework to ensure workers at risk of being displaced are redeployed. There is no reason why this should not be the case. History shows that past waves of technological disruption have ultimately led to increased prosperity, productivity and employment. Over centuries, machines have progressively replaced labour in agriculture, manufacturing, administration and professional services. Yet, humans always find work to do—partly because technology creates new opportunities for workers and partly because humans are infinitely capable of redefining what we mean by work. Today, there is a myriad of occupations that no one ever heard of a few decades ago: think of social media manager, software engineer, ride share driver, well-being coach, website builder or Zumba instructor. In response to the claim that ‘robots will take all our jobs’, economist Milton Friedman noted that “human wants and needs are infinite, and so there will always be new industries, there will always be new professions.” Centuries of economic progress confirm this view. This is not to say automation cannot cause unemployment, especially for older and vulnerable workers who lose their jobs and are unable to find a new one quickly. If automation in Australia proceeds at its historic pace, it will deliver a significant economic dividend of around $1.2 trillion over the next 15 years, but this gain is entirely predicated on our ability to redeploy the workers that are displaced by machines into new forms of work. Second, Australia must encourage more firms to intensify their automation efforts. Currently, Australian companies are behind leading global peers in embracing automation. Only 9 per cent of Australia’s listed companies are making sustained investments in automation, compared with more than 20 per cent in the United States and nearly 14 per cent in leading automation nations globally. This low rate of investment in automation technology acts as a handbrake on our productivity growth that will ultimately reduce our national income. If Australia accelerated its automation uptake, it would stand to gain up to another $1 trillion over the next 15 years. Both scenarios together—successfully moving all workers affected by automation into new employment ($1.2 trillion) and accelerating the rate of automation ($1 trillion)— represent a $2.2 trillion opportunity for Australia over the next 15 years. This economic dividend, however large, is only part of the benefit that automation can bring to Australia. Perhaps most importantly, automation has the potential to improve the work lives of every single Australian in a very tangible way. This report shows that the tasks lost to automation are typically the most dangerous, least enjoyable and the least likely to be associated with high pay. As automation shifts these dangerous, tedious and less valuable tasks from people to machines, work injuries are set to fall and work satisfaction levels bound to rise as workers can focus on more creative and interpersonal activities. As a result, human work will become safer, more meaningful and more valuable. In other words: machines will make human work more “human”. 5 Automation is changing the way we work 2 HOURS of weekly routine and manual work replaced with interpersonal, creative, and cognitive tasks over the next 15 years $2.2 boost to Australia’s national income between 2015 and 2030 from productivity gains TRILLION $1tr $1.2tr from accelerating the rate of automation from transitioning the workforce As automation reduces routine and manual work, our jobs will become... 11% 62% 20% Safer Improved Satifaction More valuable fall in work place injuries as dangerous manual tasks are automated 62% of low skill workers will experience improved satisfaction higher wages for nonautomatable tasks 1 AUTOMATION IS CHANGING THE WAY AUSTRALIANS WORK This report looks at the impact of automation on Australian work. It goes beyond a mere analysis of how automation is changing what jobs we do. Rather, it investigates the way automation is changing how we do those jobs by analysing how the use of machines shifts the amount of time spent on different work tasks. For example, anyone who has walked into a bank branch in the last 20 years can see that automation has had a transformative effect. For one thing, there are far fewer tellers standing behind the counter. Automation, through the rise of automatic teller machines (ATMs) and more recently through the growth of online and mobile banking, has reduced the need for staff to dispense cash and process routine transactions. But the replacement of administrative staff with automated processes doesn’t tell the full picture of how the working lives of bank employees have changed. Banks still have tens of thousands of workers in branches across the country, but instead of calling them “tellers”, these people are now often called “service consultants”. They spend far less time, if any, counting notes and far more time engaging with customers, such as providing complex advice on financial planning or home loans. To understand the full impact of automation on the way Australians work, we have to dig beneath the occupational level of Australia’s 12 million workers to understand not just what jobs they do, but how they spend their time at work. This report analyses how Australians spend a total of 20 billion work hours each year, assigning each of those hours to one of more than 2,000 different work tasks and then bundling these work tasks into six “task groups”: • Interpersonal tasks: These tasks primarily relate to activities that involve directly engaging with other people. A shop assistant selling products to customers would be a typical interpersonal task, as well as a manager training staff or a teacher helping students solve a complex maths question. 8 • Creative & decision-making tasks: These tasks relate to activities involving a large amount of creativity and out-ofthe-box thinking. Typical examples include a painter creating an artwork, a software developer writing a new computer program, and a manager considering a firm’s future strategic direction. • Information synthesis: These tasks relate to activities requiring workers to interpret information or extract meaning from simple data points. An analyst making sense of an industry trend and writing a commentary to provide context around this trend would be a typical example. • Information analysis: These tasks relate to the gathering and processing of information. Typical examples include a meteorologist measuring rainfall, or a cashier calculating daily sales values. • Physical predictable tasks: These tasks include repetitive and routine physical work, such as assembly line workers packaging equipment, or agricultural workers picking fruit. • Physical unpredictable tasks: These tasks relate to a wider array of physical work that is not happening on a routine basis. A car mechanic repairing different types of defects would undertake a physical, yet unpredictable task. The same applies to a doctor performing various types of surgery. Activities in the first three tasks groups—interpersonal, creative & decision making, and information synthesis—are generally the least likely to be rapidly replaced by machines. However, activities in the last three tasks groups—information analysis, physical predictable and physical unpredictable— are expected to experience workplace change driven by automation in the near future. Methodology: Understanding the “tasks” undertaken in every Australian job This report analyses how people from all walks of life—teachers and tradesmen, computer programmers and priests— have been spending their time at work since the year 2000. The observed historical trends are then used to draw conclusions on likely work patterns until 2030. We use a detailed occupational database (O*NET) which breaks down every job into more than 2000 tasks.1 The database reports the frequency with which every occupation performs each task. To convert this into the number of hours spent on each task, the tasks and frequencies were fitted to match the total weekly work hours for each occupation (See appendix A). This approach has two significant benefits over approaches which use judgement or survey data to analyse the time spent on tasks. The first advantage is that this approach removes human error from the equation; human judgement is often subject to biases and crowding effects as can be seen from notable failed “predictions” from the past. The second advantage is that the approach is repeatable over time: whilst surveys can’t be conducted in the past, the approach used in this report can be taken to historical data. The methodology utilised in this report can thus be used to discover and interpret historical trends that surveys cannot measure. EXHIBIT 1 The impact of automation is best understood by breaking the economy down into “tasks” 350+ “Occupations” Non-exhaustive examples: Sales assistant 2000+ “Activities” Six “Task groups” Non-exhaustive examples: Review documents Interpersonal Assess products Assist customers Creative & decision-making Operate equipment Factory worker Monitor facilities Perform manual tasks Assist students Manager Evaluate processes Supervise others Design lesson plans Teacher Maintain hardware Monitor environment Difficult to automate Information synthesis Information analysis Unpredictable physical Easier to automate Predictable physical SOURCE: O’NET, AlphaBeta analysis 1. The US government’s occupational data base O*NET contains detailed information on more than 2,000 work-related activities in almost 1,000 US occupations. 9 AUTOMATION IS CHANGING THE WAY AUSTRALIANS WORK 1.1 OVER THE NEXT 15 YEARS, THE AVERAGE AUSTRALIAN WORKER WILL SPEND 2 HOURS PER WEEK LESS ON MANUAL AND ROUTINE TASKS By analysing all the hours Australians workers allocate across more than 2,000 tasks, we get a remarkable picture of how the real working lives of Australians have been changing over the last 15 years. The results of the analysis, summarised in Exhibit 2, paint the picture of a workforce that is changing rapidly. Automation is causing Australian workers to rely on their brains and personalities more than physical labour. Workers have been able to spend less time on routine and manual tasks and more time on complex activities that require a high degree of creative thinking, decision-making, problem-solving, interpretation of information and personal interaction. The arrival of OVC services has helped screen content leap out of the living room and into computers, tablets and mobile phones. Viewers carry screen content with them on their daily commutes and they share it with colleagues at work, look to it in classrooms and for DIY at home – integrating screen content into every part of their daily lives. Exhibit 3 demonstrates the increasing share of viewership occurring on internet-enabled mobile and tablet devices. From just over two hours a month five years ago, Australians now spend approximately 7.3 hours a month on tablets and mobiles viewing screen content. As data packages get cheaper and screens get bigger, this trend looks set to increase. Viewership on personal computers has increased by 45% in the same time, while the use of television sets (including time-shifted viewing) has decreased slightly. EXHIBIT 2 Automation is changing the way we work, reducing the amount of time a worker will spend on routine tasks by up to 2 hours a week Change in types of tasks performed by Australian workers Average share of time spent on work activity Interpersonal Creative & decision-making 31% 25% 4% 8% Unpredictable physical 14% 18% SOURCE: O’NET, AlphaBeta analysis 10 39% +1 hour 20 minutes 2 additional hours a week spent on nonautomatable tasks. Information synthesis Information analysis Predictable physical 35% 2015-2030 change in average work week 26% 27% 6% 7% 11% 16% 7% 6% 9% 13% +20 minutes +20 minutes -20 minutes -50 minutes -50 minutes 2 fewer hours a week spent on automatable tasks. In 2000, automatable activities—a baker cleaning his trays, a warehouse worker driving a forklift, a doctor sifting through piles of scanned images to detect a tumour—used to take up 14 hours (40 per cent) of a typical 35-hour week. Since then, the share of automatable tasks has declined to 11.9 hours (34 per cent) per week in 2015. On the flipside, Australian workers have begun to fill their days with a growing number of tasks that require interpersonal skills. For example, they are spending more time talking to patients, negotiating with clients or conferencing with colleagues. The relevance of these social interactions at work has risen steadily from consuming 10.9 hours (31 per cent) of a typical 35-hour week in 2000 to 12.2 hours (35 per cent) per week in 2015. Exhibit 2 also shows a forecast for what new work Australians will carry out over the next 15 years. This forecast is based purely on the historical trend (note that in a later section, we discuss the implication of this trend accelerating). In this scenario, it is estimated that the average Australian will use another one hour and 20 minutes of work time for job-related activities involving interpersonal skills by 2030, leading their total share to rise to 13.7 hours (39 per cent) per week. Tasks requiring creative and complex cognitive thinking will also become more important. In all, Australians will spend on average 2 hours per week more on interpersonal, creative and synthesis tasks; and less time on routine and manual tasks. 1.2 AUTOMATION WILL CHANGE THE JOBS WE DO, BUT IT WILL MOSTLY CHANGE THE WAY WE DO OUR JOBS Most of the media commentary on automation focuses on the impact at a jobs level—on jobs destroyed or created. This focus is misplaced. Exhibit 2 shows that machines are expected to take over an additional 2 hours of routine and manual work in an average Australian work week by 2030. But most of this change won’t come from people changing jobs as manual and routine work disappears. In fact, 1 hour and 25 minutes—more than two thirds—of the total expected reduction in work time will come from people doing the same job, but completing fewer manual and routine tasks on the job (Exhibit 3). A much smaller part of the automation-driven workplace change will involve workers changing jobs. These workers are indeed at risk of unemployment, but it is important to understand that they only account for one third of the total expected change. Workers who perform a high share of automatable work, such as construction workers or machine operators may need to find new jobs as a result of automation. A detailed analysis of how Australian workers have been spending their time in recent years, as seen in Exhibit 4, reveals a substantial shift away from monotonous tasks. Between 2000 and 2015, the average Australian full-time salesperson has spent nine hours per week less on scanning barcodes and other automatable tasks and instead used that time to assist customers. Similarly, new online education programs and interactive learning software are freeing up teachers to spend more time interacting with their students. Since 2000, an average Australian full-time teacher has been able to delegate eight hours of dull weekly routine work—such as recording test scores—to computers. Occupational data show that teachers have used the newly gained spare time for tasks requiring creativity, interpersonal skills and strategic problem-solving abilities, such as helping special-needs students. The roles of employees in the manufacturing sector, by far the biggest user of automation technology according to the International Federation of Robotics, are rapidly changing. As industrial robots and other process-automation technologies are increasingly shouldering the physical part of factory work, labourers doing routine manual work, such as packers or assembly-line workers, have spent eight hours more per fulltime work week on training and other non-automatable tasks between 2000 and 2015. Even managers, which are commonly thought of as being immune to the impact of automation, have gained one hour of work time per week since 2000 to spend on non-routine activities thanks to technology. New management automation software helps them collect huge amounts of complex data and speed up office workflows, allowing them to focus more on creative and interpersonal tasks, such as strategic planning and keeping customers and staff happy. 11 AUTOMATION IS CHANGING THE WAY AUSTRALIANS WORK EXHIBIT 3 Two-thirds of the shift away from automatable tasks will be driven by people changing the way they work, not changing jobs Source of change in types of tasks performed by Australian workers Expected fall in average fulltime weekly work hours spent on automatable tasks from 2015-2030 Reduction due to changing jobs 29% (35 minutes) 71% Less than one-third of the reduction in automation work is due to people changing jobs (with job loss being one part of this) (1 hour 25 minutes) Reduction due to changing activities within jobs More than two-thirds of the reduction in automatable work comes from a change in the way jobs are carried out SOURCE: O’NET, ABS, AlphaBeta analysis A detailed analysis of how Australian workers have been spending their time in recent years, as seen in Exhibit 4, reveals a substantial shift away from monotonous tasks. Between 2000 and 2015, the average Australian full-time salesperson has spent nine hours per week less on scanning barcodes and other automatable tasks and instead used that time to assist customers. Similarly, new online education programs and interactive learning software are freeing up 12 teachers to spend more time interacting with their students. Since 2000, an average Australian full-time teacher has been able to delegate eight hours of dull weekly routine work—such as recording test scores—to computers. Occupational data show that teachers have used the newly gained spare time for tasks requiring creativity, interpersonal skills and strategic problem-solving abilities, such as helping special-needs students. The roles of employees in the manufacturing sector, by far the biggest user of automation technology according to the International Federation of Robotics, are rapidly changing.2 As industrial robots and other process-automation technologies are increasingly shouldering the physical part of factory work, labourers doing routine manual work, such as packers or assembly-line workers, have spent eight hours more per fulltime work week on training and other non-automatable tasks between 2000 and 2015. Even managers, which are commonly thought of as being immune to the impact of automation, have gained one hour of work time per week since 2000 to spend on non-routine activities thanks to technology. New management automation software helps them collect huge amounts of complex data and speed up office workflows, allowing them to focus more on creative and interpersonal tasks, such as strategic planning and keeping customers and staff happy. EXHIBIT 4 Automation will free up time for workers to focus on higher value tasks Non-exhaustive examples: Task composition of work Fulltime hours per week Automatable 2000 Sales assistant 28 Non-Automatable 2015 37 9 hours 12 Factory worker1 6 14 • Less time on an assembly line 8 hours 34 26 35 36 1 hour 5 4 27 35 5 • More time training other workers • Less time collecting data • More time on strategic planning 8 hours 13 • Less time scanning items • More time assisting customers 3 Manager2 Teacher3 Time saved on automatable tasks Reduction in weekly hours spend on automatable tasks • Less time recording test scores • More time assisting higher need students Notes: Assumes a full-time worker works 40 hours per week, figures rounded to nearest hour 1 Time shares show are for illustrative examples of a factory worker 2 Unweighted average of ANSZSCO 1 digit code used to estimate manager timeshares (excluing farmers and CEOs) 3 Example based on high-school teacher SOURCE: O’NET, ABS, AlphaBeta analysis 13 2 AUTOMATION IS A $2.2 TRILLION OPPORTUNITY FOR AUSTRALIA —IF WE GET IT RIGHT The growing use of machines presents a substantial opportunity for the Australian economy. The previous section spelled out the nature of that opportunity in the simplest terms: at the current rate of automation, the average Australian worker will need 2 hours less each week to do their job by 2030 because machines will liberate them from a range of automatable tasks. In these terms, automation is a quintessential positive productivity shock. This section quantifies the opportunity for Australia to capitalise on this positive productivity shock and turn the growing trend to automation into a national economic success story (Exhibit 5). First, there is an immense potential gain from having strong labour-market and education policies in place to ensure that work hours displaced by machines are reinvested in other work or new employment for the minority of displaced workers. To put this in numerical terms: if every Australian was able to spend the extra 2 hours of weekly work time that machines are expected to shoulder over the next 15 years on higher-value activities (rather than simply reduce their work time by 2 hours per week), it could boost Australia’s economy by up to $1.2 trillion in value over that timeframe.3 EXHIBIT 5 Automation could deliver a $2.2 trillion dividend to Australia if workers are transitioned successfully and the uptake of automation is accelerated Incremental gains from automation Net present value of GDP increment, $ trillion 1.0 2.2 1.2 Gains from transitioning workers Scenario: Difference in value between automation displacing workers vs workers transitioned to new activities 3. Measured in net present value (NPV) terms. 14 Gains from accelerating automation Total Difference in additional productivity value between historic rate of automation and increasing automation to the US rate Sum of transition and acceleration EXHIBIT 6 The impact of automation is not new, but where automation is having an impact has changed Job losses due to productivity improvement by sector % of employment lost each year, US data Agriculture Manufacturing Construction Mining Utilities Services 0.8 Historical job losses have been concentrated in highly physical industries such as agriculture and manufacturing 0.7 0.6 Service industries have been less impacted throughout history, but this is beginning to change 0.5 0.4 0.3 0.2 0.1 0.0 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 Note: 2011 onwards based on the linear trend for each industry since 1990 SOURCE: Groningen Growth and evelopment Centre. World-KLEMS database Second, there is significant scope to increase the gains from automation if Australian firms deepened their investments in productivity-enhancing technologies. If historic trends continue, automation will improve Australia’s labour productivity by 8 per cent over the next 15 years. This means automation would drive around one third of the total expected increase in labour productivity in Australia by 2030.4 But Australian companies lag behind global peers in embracing automation. If Australian firms were to accelerate their automation investments to match automation rates in leading global countries such as the US, they could add around $1 trillion to Australia’s economic output over the next 15 years.5 2.1 AUSTRALIA WOULD GAIN $1.2 TRILLION FROM TRANSITIONING WORKERS AFFECTED BY AUTOMATION For Australia to make automation an economic success, we must ensure that the labour freed up by machines taking on routine and manual tasks is redeployed, not left idle. For the most part labour will be readily available to redeploy within existing jobs, however in some instances automation can lead to higher unemployment or reduced work hours. If overall employment is reduced, rather than output increased, then the potential economic gains of automation could evaporate. 4. Labour productivity measured as real GDP per hour worked was $84 in 2015 and is expected to rise to $103 by 2030, $7 of the change can be attributed to automation based on historic trends with the remaining change due to all other factors. 5. Measured in NPV terms. 15 AUTOMATION IS A $2.2 TRILLION OPPORTUNITY FOR AUSTRALIA—IF WE GET IT RIGHT History suggests that there is no reason for automation to spark widespread and persistent unemployment. Past waves of technological disruption have ultimately led to increased prosperity, productivity and employment. Over centuries, machines have progressively replaced labour in agriculture, manufacturing, administration and professional services without causing mass unemployment. Exhibit 6 shows that automation is not a new phenomenon. In the 1950s, automation caused a large number of agriculture workers to lose their jobs. In the 1990s, automation primarily affected manufacturing workers. As modern machines are increasingly capable of undertaking routine cognitive labour, the impact of automation is widening. Advances in artificial intelligence—with machine learning techniques like deep neural networks allowing us to realise outcomes including optical-character recognition and language processing—mean computers are now able to drive cars, trade stocks, detect fraud, and recognise speech to answer basic questions. Exhibit 6 shows that the services sector, traditionally shielded from automation-related job losses, is fast becoming a prime target for technology-driven productivity reforms.6 Over the past decade, between 1 and 1.5 per cent of services jobs have disappeared due to technological change and other productivity improvements. While the data illustrates productivity-related job losses in the US, a similar trend can be observed in Australia and other developed nations. In tourism, for example, the share of Australian holidaymakers who used official agents to receive travel advice in 2013 had fallen to 37 per cent—15 per cent less than in the year 2000. The culprit? The internet, which is emerging as the top source of information for travellers.7 Increasingly, automation is transforming the way we choose to receive services. Such a disruption, although painful for individual workers displaced by technology, does not necessarily cause widespread mass unemployment, as the history of the US agricultural sector illustrates. When new farm equipment such as tractors and combine-harvesters started boosting the productivity of American farms over much of the 1950s and 1960s and forced a substantial share of unskilled labourers out of work each year, the nationwide unemployment rate barely budged. The US jobless rate even retreated from a high of 6.8 per cent in 1958 to 3.5 per cent in 1969, indicating laid-off agriculture workers successfully found other jobs elsewhere in the economy.8 To be sure, there are widespread concerns that this time around the impact of technological change on employment will be much more profound, as advances in artificial intelligence are now enabling computers to take over a growing number of cognitive tasks, rather than simple physical activities.9 However, whilst automation has indeed begun to impact many “white collar” occupations, which were long shielded from the impact of computerisation, computers are still primarily replacing predictable work.10 The cognitive tasks modern computers are performing—data entry, predictable calculations, or even statistical analysis using machine learning—are still reliant on well-defined rules and structural data. 6. Groningen growth and development centre, World KLEMS database data for US employment and productivity. While not all productivity gains are due to automation, productivity gains that result in job losses are more likely to be driven by machines replacing human labour than factors such as improved education. 7. Roy Morgan Research, (2013): Available from: http://www. roymorgan.com/findings/travel-agents-overseas-holidays-201302270608 8. US Department of Labor. Historical labour force statistics available at: https://data.bls.gov/timeseries/LNU04000000?years_option=all_ years&periods_option=specific_periods&periods=Annual+Data 9. Jason Furman (2016), Is this time different? The opportunities and challenges of artificial intelligence. Available at: https://obamawhitehouse. archives.gov/sites/default/files/page/files/20160707_cea_ai_furman.pdf 10. Frank Levy and Richard Murnane (2013), Dancing with robots, human skills for computerised work. Available from: http://content.thirdway.org/ publications/714/Dancing-With-Robots.pdf 16 EXHIBIT 7 Australia’s economic gain from full transition of affected workers rather than full displacement is $1.2 trillion 2030 scenario Workers displaced scenario Workers transitioned scenario Key assumptions None of the time savings are reinvested: • Work hours are reduced involuntarily • Workers are displaced without being reabsorbed into other jobs • All time savings are reinvested to generate more output • Hours worked per capita remains unchanged Workers transitioned $2.6 trillion 2015-2030 GDP, real 2015 A$ trillion Outcomes $ Workers displaced $2.3 trillion $1.6 trillion 2015 2020 Year 2025 2030 Note: all scenarios are based on automation replacing <8 minutes of the average work week p.a. to 2030 SOURCE: O’NET, ABS, AlphaBeta analysis This means humans are still indispensable. The algorithm sifting through piles of big data cannot function without a human mind that specifies the rules it must follow, and data used to train deep neural networks must be labelled so the algorithm can learn. Computers are still far inferior to humans in handling unpredictable situations that require out-of-thebox thinking, empathy and understanding other humans. While today’s automation technologies differ from those employed in the 1960s, historic developments remain as good a guide as possible over how automation will impact work. The experience of the 20th century indicates that it is unreasonable to assume that the current wave of technological progress will displace millions of workers. This is not to say that automation cannot cause any unemployment, especially for older and vulnerable workers who lose their jobs and lack the flexibility to find a new one quickly. Depending on the economic climate, and the skills and mobility of individual workers, many might struggle to find new work in their region. In a negative scenario of future automation in Australia, the nationwide unemployment rate could rise, even as workers remain in short supply in pockets of the economy. A skills mismatch and lack of re-training opportunities would hinder laid-off workers from being hired in expanding industries with high labour demand. The result: stalling economic growth, as the productivity boost from automation would be offset by workers sliding into unemployment. Exhibit 7 describes two scenarios for the employment consequences of automation. In the “workers displaced scenario” the time saved by automation is not reinvested into other activities. In this scenario, Australian productivity would rise but GDP growth would be limited. In the “workers transitioned scenario” the time saved by automation is redeployed and workers are transitioned into new activities. In GDP terms, the net present value of these two scenarios differs by $1.2 trillion over 15 years. The ultimate outcome is likely to lie between these two worlds. 17 AUTOMATION IS A $2.2 TRILLION OPPORTUNITY FOR AUSTRALIA—IF WE GET IT RIGHT EXHIBIT 8 Australia lags behind global leaders in automation Global automation uptake1 % publicly listed firms engaging in automation, 2010-2015 Switzerland 25.1% USA 20.3% United Kingdom 12.3% Finland 12.0% Netherlands 10.8% Germany 10.7% France 10.0% Sweden 9.6% Australia 9.1% Hong Kong 7.5% Belgium Norway 50% fewer Australian firms are engaged in automation compared to leading countries2 6.5% 5.4% Leading countries average2=13.9% Note: Only countries with detailed firm level data for over 100 publicly listed firms included 1. Automation uptake measure as the fraction of publicly liste firms with 5% growth in capital expenditure and labour prouctivity over 5 years 2. Includes all countries in sample with higher automation rates than Australia SOURCE: Compustart data, AlphaBeta analysis 2.2 AUSTRALIA WOULD GAIN $1 TRILLION BY ACCELERATING THE RATE OF AUTOMATION If Australia embraced automation more strongly, rather than fighting it, the economic benefits of using machines at work could be even greater. Compared to other advanced economies, Australian firms appear to underinvest in automation technology. Data on publicly listed firms, comprised in Exhibit 8, show that Australian companies lag behind global peers in investing in robotics and other productivity-boosting technologies.11 In Switzerland, for example, more than 25 per cent of publicly listed companies appear heavily engaged in automation. These “automation leaders” have been making continuous capital investments in anything from smarter machines to automation software between 2010 and 2015, which has helped boost the productivity of their workers by at least 5 per cent over that period.12 In Australia, the automation uptake among publicly listed companies is around 9 per cent. While this is comparable to the degree of automation engagement among listed firms in Sweden, it is close to three times lower than the automation rate among listed companies in Switzerland and less than half the automation rate of listed companies in the US (20.3 per cent). In some of the most progressive countries globally, as seen in Exhibit 8, at least 10 per cent of listed companies appear strongly committed to automation. 11. Data source from S&P Compustat: Global and North America fundamentals databases. Note: Given the lack of available global automation benchmarks, an index measuring change in investment and productivity was constructed instead where sufficient data was available 12. Automation leaders, are defined as publicly listed firms whose investment and labour productivity have increased by at least 5 per cent between 2010 and 2015. Japan not included in dataset due to lack of granularity for employee data 18 13. Deloitte (2016), Global Manufacturing Competitiveness Index. Available at: https:// www2.deloitte.com/global/en/pages/ manufacturing/articles/global-manufacturing- Historic productivity growth EXHIBIT 9 10-year rolling average If Australian firms embrace automation to the same extent as peer economies, prouctivity growth could increase by over 50% Historical rate continues Labour productivity growth % of Year on year growth, historic and projected productivity growth Australia catches up to the US2 10 9 8 7 6 5 4 3 2 1 0 -1 -2 -3 Catching up to the US could reduce automatable work by over 4 hours a week for the average worker, as opposed to 2 hours under historical trends Australia catches up to leading rate economies1, 2 Catching up to the US rate of automation uptake by 2020 would increase annual productivity growth by 51% 2.2 1.4 1990 1995 2000 2005 2010 2015 2020 2025 +51% 2030 SOURCE: ABS, Compusial, O’NET, Alpha Beta analysis 1 Leading peer economies is defined as an equally weighted average of all countries ahead of Australia in automation uptake 2 Scenario relates to the proportion of firms embracing automation as defined on Exhibit 8. Scenario assumes Australia catches up to the benchmark rate by 2020 and maintains that rate out to 2030, resulting in an uplift in productivity Many people, led by fears that robots could trigger mass unemployment, may be considering this relatively low level of corporate automation uptake a boon for Australia. Yet slowing down the pace of automation, rather than accelerating it, may do more harm than good, depriving Australia of the resulting productivity benefits, and potentially reducing the global competitiveness of local industries. The US manufacturing industry may serve as an example: US manufacturing firms have invested heavily in automation technologies in recent years to remain competitive against foreign low-cost rivals. This has led to job losses, but also enabled a large number of workers to move into better-quality roles and remain employed in an industry that many considered precarious. The result: by 2020, the US manufacturing industry is expected to be more globally competitive than China’s.13 If local firms were as committed to automation as their US peers, which would require the share of automation-focused firms in Australia to roughly double in coming years, Australia’s productivity growth could increase by more than 50 per cent to 2.2 per cent annually by 2030, as seen in Exhibit 9. Higher productivity growth means we can produce the same output with less work. More specifically: doubling the pace of automation to match the US uptake would allow Australian workers to save twice as much time spent on dull and dangerous tasks. It would relieve the average Australian worker of more than 4 hours of repetitive weekly routine work by 2030, as opposed to a saving of just 2 hours per week if past automation rates continued. Accelerating the pace of automation could provide an even stronger catalyst for Australia’s economic growth, provided there are policies and opportunities in place to help affected workers develop in-demand skills and remain productively engaged in the workforce. Smart retraining opportunities can go a long way to supporting workers to stay productive, as discussed in further detail in Section 4. If all workers affected by automation remain employed, increasing the rate of automation in Australia to US levels could add another A$1 trillion to Australia’s economic output over the years between 2015 and 2030. 19 3 AUTOMATION WILL MAKE AUSTRALIAN JOBS SAFER, MORE SATISFYING AND MORE VALUABLE Seizing Australia’s $2.2 trillion automation opportunity isn’t only about the strength of the economy as a whole. It will have real, tangible benefits for every worker in Australia. Machines have a huge potential to change the daily work routine of millions of people for the better. This section provides detailed insights on how machines are advancing human work. Think of a butcher using robotic meatcutting machines instead of handling sharp knives himself, think of a mining worker getting a good night’s sleep while autonomous haulage trucks are doing the long tiring drive across a mine site. An analysis of recent trends in worker satisfaction, workplace injuries and pay levels for tens of thousands of Australians confirms what may seem like an intuitive finding: machines are shouldering our riskiest, least enjoyable and least valuable tasks within a job, allowing humans to focus on more creative and interpersonal tasks. In short: machines enable humans to be more “human” at work. The benefits of automation for Australian workers are quantifiable. For one, allowing robots to take on more manual work will deliver a particularly strong gain for anyone involved in painstaking, physical labour, which is currently responsible for the bulk of workplace injuries. Assuming past automation trends continue, the amount of sick days due to accidents involving physical work in Australia could be 11 per cent lower by 2030. Second, work satisfaction is bound to increase, as machines take over a greater share of dull routine tasks. This analysis shows that the monotonous, automatable tasks performed by typically low-skilled workers are also the least satisfying tasks to perform. If current automation trends persist, low skill workers will take on more stimulating and satisfying human tasks at work, and as many as 62 per cent of them would be happier in their jobs by 2030 compared with today. Third, allowing more people to focus on tasks that are more difficult to automate has a clear financial benefit. Easily automatable tasks are among the worst paid. In contrast, work activities that are difficult for robots to take over because they require a large amount of creative thinking, human logic and emotional intelligence earn almost 20 per cent more than automatable tasks. 3.1 SAFER JOBS, AS MACHINES TAKE ON DANGEROUS PHYSICAL TASKS As we use machines to automate physical tasks, workplaces become safer. This is because the activities that are easiest to automate, such as painstaking physical work, are typically among the most dangerous. Considering that physical tasks use up only around one quarter (27 per cent) of all work hours in our economy, as seen in Exhibit 2, they cause an outsized number of work accidents. In 2015, they were responsible for more than half (57 per cent) of all sick days workers took to recover from injuries sustained on the job, Exhibit 10 shows. Robots have the potential to substantially lower the amount of workplace accidents by taking over tasks that often lead to injuries, such as lifting heavy objects or operating dangerous machinery. The use of self-driving trucks, for example has been resoundingly successful for mining company Rio Tinto. Today, 69 fully automated trucks are moving around its remote iron-ore mine sites in the Australian Pilbara desert, making Rio Tinto the world’s largest owner and operator of autonomous haulage systems.14 The company’s safety record has improved noticeably since introducing self-driving trucks, with injury rates falling from 1.21 accidents per 200,000 hours worked in 2007 to 0.44 accidents in 2016.15 14. Rio Tinto (2017), “Mine of the future”. Available at: http://www.riotinto.com/australia/pilbara/mine-of-the-future-9603.aspx 15. Rio Tinto (2016), Sustainable Development Report, page 25. Available at: http://www.riotinto.com/documents/RT_SD2016_our_people.pdf 20 EXHIBIT 10 Workplace injuries will fall by 11% as automation eliminates some of the most dangerous physical tasks in the economy Injury intesnse tasks will be automated... % share of total days lost to injury in the economy ...reducing workplace injuries Millions of work days lost to workplace injury -11% Predictable physical 33% Unpredictable physical Information analysis Information synthesis 1.9 24% 4% 3% Interpersonal Creative & decision-making 20% Some of the most automatable task groups are also the most likely to cause workplace injuries 1.7 15% 2015 2030 SOURCE: ABS, O’NET, Alpha Beta analysis Assuming past trends continue, the total number of work days lost to injuries sustained from physical work in the Australian economy could fall by 11 per cent to 1.7 million in 2030. The productivity gain will likely be even higher, as machines can also improve the safety of jobs involving non-physical tasks. Any worker, not just those performing physical tasks, is at risk of being involved in a car accident. Autonomous driving technology has the potential to reduce such accidents, given crash rates are lower for autonomous vehicles. 16 3.2 MORE SATISFYING JOBS, AS MACHINES TAKE ON ROUTINE TASKS Automation will also increase work satisfaction, particularly for lower skilled workers, who are often required to perform the most dangerous, strenuous and repetitive jobs in an economy. The Household, Income and Labour Dynamics in Australia (HILDA) survey, compiled by researchers at the University of Melbourne, provides valuable information about the personal well-being of different worker groups in Australia.17 To understand which workers are currently most dissatisfied with their jobs and thus stand to gain most if unpleasant work tasks were automated, the latest results of the HILDA survey were applied to the six activity groups identified earlier in the report (for details on the methodology see Appendix B). 16. Myra Blanco, Jon Atwood, Sheldon Russell, Tammy Trimble, Julie McClafferty and Miguel Perez (2016). Automated vehicle crash rate comparison using naturalistic data. Available from: http://www.vtti.vt.edu/featured/?p=422 17. Melbourne Institute of Applied Economic and Social Research, HILDA Survey. Available at: http://melbourneinstitute.unimelb.edu.au/hilda 21 AUTOMATION WILL MAKE AUSTRALIAN JOBS SAFER, MORE SATISFYING AND MORE VALUABLE EXHIBIT 11 Workers that are losing jobs to automation have no other skills. The least satisfying tasks will be automated... Satisfaction ratings scale of 1-10 Predictable physical 7.4 Unpredictable physical 7.4 Information analysis 7.5 Information synthesis 7.5 Interpersonal Creative & decision-making ...Increasing job satisfaction for low skill workers % of workers with improve satisfaction, 2015-2030 Workers currently engaged in more automatable tasks have lower job satisfaction 62% 7.7 7.8 Australia’s lowest skill workers will benefit most as their work moves away from automatable tasks 56% 30% Low skill workers Medium skill workers High skill workers SOURCE: ABS, O’NET, HILDA, AlphaBeta analysis The outcome, illustrated in Exhibit 11, shows that the most easily automatable tasks, such as assembly-line work or data entry, are typically also the least enjoyable. By taking over more and more of these monotonous and tedious activities, automation has the potential to raise the job satisfaction for every worker, albeit to varying degrees. The improvement tends to be strongest for the low-skilled, who typically perform the bulk of automatable work. If current automation trends persist, it is estimated that 62 per cent of low-skilled workers in Australia would be happier in their jobs by 2030 compared with today. High-skilled workers would also benefit: 30 per cent of them would likely report a higher job satisfaction in 2030 if they could swap some of their automatable routine work for more complex and creative tasks. 3.3 MORE VALUABLE JOBS, AS MACHINES TAKE ON THE LEAST VALUABLE TASKS WITHIN EACH JOB There is a clear financial incentive to shift from repetitive routine work to activities that require more complex, creative and interpersonal skills. Australian wage data shows that the least automatable tasks are typically the best paid (see Appendix B for details on the methodology).18 18. ABS 6306.0 DO011_201405 – Employee earnings and hours, Australia May 2014. 22 EXHIBIT 12 The “value” of each hour of work will increase as workers focus on the most valuable tasks in their jobs The tasks less exposed to automation make up 71% of wage income... % share of wage income, 2015 36% Interpersonal ... but these tasks take up 66% of time % share of time, 2015 36% 71% of wages Creative & decision-making 30% Information analysis 5% 9% Unpredictable physical 8% Information synthesis 66% of time 25% 6% 7% 12% 11% 16% Share of wages Share of time Predictable physical SOURCE: ABS, O’NET,, Alpha Beta analysis Exhibit 12 shows that tasks that are difficult for robots to perform earn almost three-quarters (71 per cent) of the total wage income generated in Australia, despite only taking up two-thirds of total work time in Australia. This means that a worker who spends 40 hours a week on nonautomatable tasks—whether teaching students or setting a new business strategy—earns approximately 20 per cent more per hour compared to someone who spends 40 hours a week performing automatable tasks, such as packaging deliveries, preparing food or operating heavy machinery (see Appendix B). Based on these results, the gains from automation could be particularly substantial for low-skilled workers. If such workers could learn to perform more uniquely human tasks and firms also accelerated their rate of automation, real wages for this group could be 10 per cent higher by 2030, an annual income gain of approximately $6,000 per worker. 23 4 HOW AUTOMATION CAN BECOME A SUCCESS STORY IN AUSTRALIA Australians have much to gain from embracing the opportunity to offload repetitive routine tasks onto machines. However, these gains cannot be taken for granted. Automation can only become a success story in Australia if policymakers help workers navigate the big shift towards automation. This requires a finely tuned policy response rather than a blanket approach. For example, low-risk workers are expected to require little policy support to remain employable in the automation age, as most of them already perform a large amount of uniquely human tasks. High-risk workers, in contrast, are at risk of sliding into unemployment if policymakers fail to enact targeted retraining and job transition programs. Ignoring the needs of Australia’s most vulnerable workers would come at a large cost for society, potentially driving 20 per cent of them into joblessness. Educating future workers is equally crucial. An additional 6.2 million people are projected to join the Australian workforce in coming years. They will significantly advance our economy if they have the skills to perform the high-value tasks that robots are unable to master. Australia needs a bold and proactive policy approach that treats automation as an economic opportunity, rather than threat. There is value in drawing on the experience of other countries, and this section provides some examples as a starting point. 4.1 POLICY SHOULD BE TARGETED AT DIFFERENT GROUPS AFFECTED BY AUTOMATION Policymakers in Australia play an important role in harnessing Australia’s automation opportunity. They design the framework that allows businesses to take advantage of automation and create safer, more enjoyable and more valuable jobs. But this policy framework also needs to support workers whose jobs are at risk. Different groups of workers have different policy needs. The young, well educated, and highly skilled will likely adapt easily to changes in their workplace. Others, including lower-skilled workers and those near retirement age, will likely struggle more when trying to transition from one job to another. Rather than pursuing a blanket approach, policymakers must meet the needs of three different worker groups when providing support, highlighted in Exhibit 13. • Current high-risk workers: These workers are predominantly low-skilled and perform a large share of automatable tasks. If these workers lose their jobs, they would need a lot of support to find new employment. • Current low-risk workers: For these workers, the benefits from automation will likely outweigh its threat. The majority of them are medium- to high-skilled employees who perform a variety of uniquely human tasks. While parts of their jobs might be prone to automation, there is still plenty of work for them that machines cannot eliminate. • Future workers: These workers have not yet joined the labour force, and their skill level can still be influenced by education and training. This gives policymakers an opportunity to be proactive and design strategies to ensure future workers have the right skills to succeed in an increasingly automated world. 24 EXHIBIT 13 Australia’s policy response to automation will need to be tailored to different groups of people Groups affected by automation Nearing retirement Career stage 1 High-risk current workers: Retrain and transition Currently mid-career Current & future students 2 Low-risk current workers: Accelerate automation and create new opportunities 3 Future workers Educate and prepare for the automated future Low SOURCE: Alpha Beta analysis The policy response to automation must cater for all three worker groups. However, quantitative analysis of the impact of automation on each group shows that the costs for society will be highest if Australia fails to adequately prepare its future workers for the automation age.19 An additional 6.2 million people are projected to join the Australian workforce in coming years, as seen in Exhibit 14. Teaching them the right skills to perform a wide range of non-automatable tasks would generate an economic gain of between A$300 billion and A$600 billion by 2030 (see Appendix D for further details on the methodology). There are already several initiatives in place to equip young Australians with critical skills for the future and boost their employment opportunities, including earlychildhood education programs.20 Medium High Skill level Costs would also be significant if policymakers failed to cater for the estimated 3.5 million workers at high risk of being displaced by automation in coming years. The economic gain of fully keeping these people in the workforce between 2015 and 2030 could amount to between A$200 billion and A$400 billion in net present value. On the flipside, a failure to reskill these most vulnerable workers could drive up to 20 per cent of them into joblessness. 19. Exhibit 14 only illustrates the expected economic gains in a scenario where businesses maintain their current pace of automation and policymakers help ensure that all affected workers remain in employment. This scenario does not include the additional gains, valued at $1 trillion (see Exhibit 6), that Australia can expect if firms accelerate their current rate of automation. 20. More details on the initiatives can be found at: https://www.mychild.gov.au/agenda 25 HOW AUTOMATION CAN BECOME A SUCCESS STORY IN AUSTRALIA Australia already provides such support to laid-off workers. One of the most prominent initiatives is the “Automotive Industry Structural Adjustment Programme”, which helps workers affected by the demise of Australia’s car manufacturing industry with training, resumé-writing advice and other assistance.21 There are signs that the program has indeed helped cushion the automotive industry’s crisis: while car manufacturers have shed tens of thousands of jobs since 2006, only 3 per cent of automotive workers remained unemployed in 2011. More than half of the laid-off workers have found a new role in other parts of the manufacturing sector or other industries.22 Workers at low risk of losing their livelihood due to automation are expected to need only minimal government support. Their relatively high skill level should enable them to switch jobs with ease if employers decide to make their role redundant or use machines for some of the tasks they were hired to perform. A recent initiative by the Australian Bureau of Meteorology (BOM) illustrates how skilled workers can remain relevant in their jobs by shifting to slightly different work activities. When the BOM decided to modernise its regional weather-observation stations and fully automate more than a dozen of them, it moved to redeploy affected staff elsewhere in the organisation—partly by retraining weather observers as technicians. The result: relatively steady employee numbers. At the end of June 2016, the Bureau counted 1,458 permanent and 206 temporary staff, compared with 1,452 permanent and 201 temporary staff a year earlier. If Australia manages to retain all low-risk workers in employment, it could generate an economic gain worth up to A$250 billion by 2030. EXHIBIT 14 The economic gains from transitioning workers successfully are value at A$600 billion to 1.2 trillion dollars GDP gains from successful workforce transition NPV of GDP (2015-2030), $A billion Total: $600 billion to $1.2 trillion Low-risk group: Future workers: • 7.6 million workers • 6.2 million workers • Successful transition is likely to occur without policy intervention • Successfully preparing the workforce of the future could be worth up to $600 billion $300-$600 $100-$250 High-risk group: $200-$400 • 3.5 million workers • Fully transitioning high risk workers could be worth up to A$400 billion and prevent employment falls in this group of 10-20% SOURCE: ABS, O’NET, Alpha Beta analysis 21. Australian Government (2017), “Automotive Industry Structural Adjustment Programme”. Available at: https://www.business.gov.au/assistance/automotiveindustry-structural-adjustment-programme 22. Australian Government (2017), “Worker redeployment and skills development”. Available at: https://industry.gov.au/AboutUs/CorporatePublications/ ReviewofSouthAustralianandVictorianEconomies/Pages/Worker-redeployment-and-skills-development.aspx 26 4.2 LESSONS FROM ABROAD: HOW OTHER COUNTRIES ARE RESPONDING TO AUTOMATION International examples can offer Australia some guidance to best harness the benefits of automation: policies range from supporting the adoption of automation technologies, better preparing future workers, and ensuring that the minority of workers whose jobs are threatened by automation can be redeployed elsewhere in the economy. Exhibit 15 provides an overview of selected initiatives around the world that Australia could potentially learn from. EXHIBIT 15 Many countries have employed a range of strategies that can harness the benefits and minimise the costs of automation 1 Retrain and transition • Community college subsidies: Findings from Washington state suggest that a year of training in community college can increase lifetime earnings by up to 9-13% Retraining & education programme • Active labour market policy: Denmark provides basic literacy & numeracy education, higher education support, and vocational training for unemployed workers • Lifetime learning credits: Singapore’s ‘SkillsFuture’ initiative offers lifetime credits of S$500 for all Singaporean citizens aged over 25 for use on enrolling in Government approved training courses • Union supported learning: Britain's national Trade Union Centre (TUC) founded “Unionlearn” in 2006 to provide UK union members learning and skilling opportunities throughout their careers 2 Accelerate automation and create new opportunities • Indirectly support automation technologies: The US Department of Transportation opened 10 autonomous vehicle testing tracks to accelerate cooperation amongst developers Embrace automation • Directly support automation technologies: Japan is investing in robotics, and deregulating the industry to support a tripling of its robotics market • Directly support automation technologies: Korea seeks to invest $500 million in robotics over the next 5 years • Directly support technologies: As an example, the Swiss postal service has trialed automated delivery robots and drones instead of traditional delivery methods 3 Educate and prepare for the automated future Industry & Educational institutions partnerships Early education initiatives Smarter learning • P-Tech schools: The P-Tech program partners leading companies with high schools to teach students the STEM skills that the future workforce requires • Dual training programs: German/Swiss apprentices split time between study and practical work experience in large German/Swiss firms, with a focus on future-proof skills • Computer programming education: The Estonian “ProgeTiger” programme introduces computer programming in school curriculums from years 1-12 • Massive Online Open Courses (MOOCS): Online learning platforms that offer ‘nano-degrees’ teaching courses tailored to skills that tech companies need SOURCE: Alpha Beta analysis 27 HOW AUTOMATION CAN BECOME A SUCCESS STORY IN AUSTRALIA Assisting high-risk workers: Embracing automation for low-risk workers: A look at policy examples from around the world can help Australia sharpen its own approach to assist the most vulnerable workers. Workers at low risk of being displaced by automation, most of them well educated and already performing a wide range of uniquely human tasks, typically require only minimal help from governments to switch from repetitive routine activities to more valuable ones. Governments can encourage the structural shift to higher value work patterns by encouraging more businesses to engage in automation or by directly investing in automation. • USA - Higher education subsidies for displaced workers: Researchers in Washington State found that one year of training in a community college increases the income prospects of laid-off male workers by 9 per cent and of laid-off female workers by 13 per cent.24 The findings encouraged Washington State to provide targeted assistance to displaced workers by funding their tuition, mapping out education paths, and helping with job searches.25 • Denmark - Active labour market policy: Denmark spends as much as 2 per cent of its GDP on its active labour market policies, ranging from assistance to improve numeracy, literacy and job readiness to funding of tertiary education and vocational training. Such policies can ensure highrisk workers are able to acquire new, relevant skills and become job-ready quickly after losing employment. The effectiveness of the policy is demonstrated by Denmark’s high levels of employment security relative to the rest of Europe.26 • Singapore - Lifetime learning support: Singapore’s “SkillsFuture” initiative grants so-called lifetime credits to all citizens aged 25 and over. These credits can be used to pay for training courses from tertiary education organisations and other approved providers. Older workers benefit particularly, as they can use credits accumulated over a lifetime to upgrade outdated skills.27 • USA - Indirect investment in automation: The U.S. Government has adopted policies that facilitate private businesses’ investments in automation. For example, the US has become one of the most liberal jurisdictions for the use of driverless cars. It has opened 10 test tracks across the country to create shared spaces for high-tech companies to facilitate collaboration in technology development.29 • Switzerland, Japan, South Korea - Direct investment in automation: Switzerland, in a strong endorsement of automation, recently began trialling the use of mail-delivery robots at its national mail service, Swiss Post.30 On a more ambitious level, the governments of Japan and South Korea are both investing large sums of taxpayer money in robotics. 31 • UK - Union supported learning: Britain’s national Trade Union Centre established a program called “Unionlearn” in 2006, which offers on-the-job training for employed union members to ensure their skills remain relevant in a rapidly changing workplace. As many as 87 per cent of employers support the program.28 24. Louis Jacobson, Robert LaLonde, and Daniel Sullivan (2004), Estimating the Returns to Community College Schooling for Displaced workers. Available at: http://repec.iza.org/dp1017.pdf 25. https://www.sbctc.edu/paying-for-college/worker-retraining-student.aspx 26. The Danish National Labour Market Authority (2008), Danish Employment Policy. Available at: https://www.oecd.org/employment/leed/40575308.pdf 27. Singapore Skillsfuture. Available at: http://www.skillsfuture.sg/credit/about#programme1 28. Unionlearn website. Available at: https://www.unionlearn.org.uk/about 29. US Department of Transportation. Available at: https://www.transportation.gov/briefing-room/dot1717 30. Swiss Post (2016), “Swiss Post to test self-driving delivery robots”. Available at: https://www.post.ch/en/about-us/company/media/press-releases/2016/ swiss-post-to-test-self-driving-delivery-robots 31. Japan’s robotics strategy (2015). Availabel at: http://www.meti.go.jp/english/press/2015/0123_01.html ‘Robot Friendly Korea’ (2012). Availabel at: http://www.korea.net/NewsFocus/Sci-Tech/view?articleId=99841 28 Preparing future workers: Australia stands to gain most benefits from policies that help future workers acquire the skills needed to perform highvalue tasks in an automated society. The following examples can provide an incentive for Australia to rethink its current education and training policies and adopt lessons from abroad. • Germany/Switzerland - Develop industry & educational partnerships: Germany and Switzerland have a long history of combining theoretical education and practical workplace training to prepare young people effectively for the reallife job environment. In Germany, about 50 per cent of all school-leavers undergo vocational training provided by companies which consider the dual system the best way to acquire skilled staff.32 Dual-track partnerships could allow the education system to rapidly adapt to disruptive changes arising from automation. • Estonia - early education initiatives: In 2012, Estonia began introducing computer programming as part of its teaching curriculum for school students as young as six, through its “ProgeTiger” program.33 Australia could similarly seek to update its education curriculums to better reflect the skills required in an increasingly automated society. • USA - Smarter learning: In the fast-paced digital world, specialised and rapidly deliverable education solutions are necessary to prepare workers for a dynamic work environment. One new educational approach is the development of “nanodegrees”. In the US, and now around the globe, providers such as Udacity34 and Coursera35 offer programming and STEM courses through online platforms. These typically cost much less than regular university degrees and can offer a narrow focus on crucial skills. Providers of these “Massive Open Online Courses” (MOOCs) can further use machine learning and data analytics based on its student cohorts to determine which courses are suited for which type of applicants. While the concept is still in its infancy, employers are increasingly open to these new models of learning: a survey of 114 human-resources managers in the US found that an overwhelming majority (95 per cent) considered nanodegrees and other “digital badges” a useful asset for applicants to have. 36 The future of automation in Australia depends on policymakers. If they enact a framework that will protect the country’s most vulnerable workers, while allowing the highly skilled to perform more meaningful jobs, automation can become a success story in Australia. Ideally, their policy choice enables Australia to seize its $2.2 trillion automation advantage. 32. Federal Ministry of Education and Research (2017), “The German Vocational Training System.” Available at: https://www.bmbf.de/en/thegerman-vocational-training-system-2129.html 33. www.hitsa.ee/it-education/educational-programmes/progetiger 34. More details at: https://www.udacity.com/ 35. More details at: https://www.coursera.org/ 36. Emily Rimland and Victoria Raish (2016) Employer perceptions of critical information literacy skills and digital badges. Available from: http:// crl.acrl.org/content/early/2015/05/11/crl15-712.full.pdf+html 29 5 APPENDICES 5.1 APPENDIX A: ESTIMATING TIMESHARES OF TASKS IN THE ECONOMY How time spent on tasks has changed across the Australian workforce Measuring the impact of automation on the workplace remains a challenge for researchers. There is no readily available data showing how workers have spent their day at work in the past, compared with today. This report utilises a unique and innovative approach to overcome this challenge. It provides the first published estimates of the impact of automation on the workplace using verifiable historical data. O*NET frequency data covering 964 US occupations was used to measure the time workers spent on various job-related tasks in recent years. The approach, summarised in Exhibit 16, was repeated multiple times between 2006 and 2014 to determine how automation has affected the work activities in different occupations.37 EXHIBIT 16 The analysis begins by assigning each O*NET Detailed Work Activity (DWA) a unique number, and noting that the amount of time a worker in occupation ‘ ’ spends on ‘ ’ DWAs in a work week can be expressed by the following equation: Where is the number of times a week a worker in occupation ‘ ’ performs task ‘ , the same worker to perform task ‘ ’, and ’ the amount of time it takes the total hours worked a week by the worker. O*NET provides survey frequency scores on a scale of 1-7 which are converted to weekly frequency scores by AlphaBeta as presented in table 1. Table 1. Conversion of O*NET frequency scores to weekly frequencies O*NET score O*NET description of frequency AlphaBeta implied weekly frequency 1 Yearly or less 0.02 2 More than yearly 0.12 3 More than Monthly 0.5 4 More than weekly 2 5 Daily 5 6 Several times a day 20 7 Hourly or more 40 It is assumed all workers surveyed work 40 hours a week and that all workers within the same 1 digit SOC code take the same amount of time to perform a task. There are ‘ ’ unknowns in equation 1, namely the different amounts of time taken to perform a task. 37. The US Department of Labor’s O*NET database is one of the world’s richest sources for labour data. The database contains information on the frequency of over 2,000 Detailed Work Activities (DWAs) grouped under 41 Generalised Work Activities (GWAs) across 964 US Standard Occupation Classification (SOC) codes. 30 EXHIBIT 16 (continued) he frequencies with which occupations within a 1-digit USthe SOC perform the time taken to hich ‘𝐽𝐽’ occupations within‘𝐽𝐽’ a 1-digit US SOC perform tasks, time taken totasks, frequencies with which ‘ ’ occupations within a 1-digit US SOC perform tasks, time taken erforms tasks,hours andThe the total weekly hours worked can besystem expressed as by a linear system given by the equation (2). to performs tasks, and total weekly worked can be expressed as a linear given equation (2). the total weekly hours worked can be expressed as a linear system given by equation (2). (2) 𝐹𝐹 ⋯ ⋱ ⋯ ℎ 𝑓𝑓1,1 40 = [⋮] [ ⋮ 𝑓𝑓1,𝐽𝐽 40 ℎ 𝑡𝑡 𝑓𝑓𝑁𝑁,140 = 𝑡𝑡1 𝑓𝑓1,1 × [ ] ⋮ ]⋮ [ ⋮ ][ ⋮ 𝑓𝑓𝑁𝑁,𝐽𝐽40 𝑡𝑡𝑁𝑁 𝑓𝑓1,𝐽𝐽 𝐹𝐹 ⋯ ⋱ ⋯ 𝑡𝑡 𝑓𝑓𝑁𝑁,1 𝑡𝑡1 ⋮ ]× [ ⋮ ] 𝑡𝑡𝑁𝑁 𝑓𝑓𝑁𝑁,𝐽𝐽 he in equation (2)(over has more (over vector 2000 activities under vector ‘𝑡𝑡’) than there are (2)system has more unknowns 2000unknowns activities under ‘𝑡𝑡’)(over than2000 thereactivities are there are equations (‘ ’ must The system inless equation (2) has more unknowns under vector ‘ ’) than quations (‘𝐽𝐽’ must be strictly than 964 occupations). For most scenarios, the system may have infinite strictly less than 964 occupations). For most scenarios, the system may have infinite 38 be strictly less than 964 occupations). For most scenarios, the system may have infinite solutions. To obtain unique 38 To obtain unique values for the vector of time to perform tasks (𝑡𝑡), a regularised least squares olutions. unique values for the vector of time to perform tasks (𝑡𝑡), a regularised least squares values for the vector of time to perform tasks ( ), a regularised least squares solution is used to satisfy equation 3. olution is used fy equation 3. to satisfy equation 3. 2 2 2 − 𝑡𝑡ℎ‖≥ + 𝜇𝜇‖𝑡𝑡‖ 𝑡𝑡𝑖𝑖 ≥ 0𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 ∀𝑖𝑖, 𝜇𝜇 𝑖𝑖𝑖𝑖 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 ‖𝐹𝐹𝐹𝐹 −(3)ℎ‖2 + 𝜇𝜇‖𝑡𝑡‖min‖𝐹𝐹𝐹𝐹 𝑤𝑤ℎ𝑒𝑒𝑒𝑒𝑒𝑒 0 ∀𝑖𝑖, 𝜇𝜇 𝑖𝑖𝑖𝑖 𝑤𝑤ℎ𝑒𝑒𝑒𝑒𝑒𝑒 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 > 0 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 > 0 𝑖𝑖 nesthe above, 𝜇𝜇 penalises large deviations in time to perform tasks. Solving equation (3) provides a large deviations in time taken to perform tasks.taken Solving equation (3) provides a nique ofIntime on activities for each occupation. Lastly DWA tasks. is mapped to equation its the above, penalisesLastly large deviations in is time takeneach to its perform Solving (3) provides a unique spentestimate on activities for spent each occupation. each DWA mapped to Generalised Activity (GWA) which are mapped to tasks groups. The of GWAs to task estimate of timetospent on activities each occupation. Lastly each DWA is mapped to itsgroups Generalised Work Activity vity (GWA)Work which are mapped tasks groups. Theformapping of GWAs to mapping task groups shown in figure (GWA) 1. which are mapped to tasks groups. The mapping of GWAs to task groups is shown in figure 1. Figure 1. Allocation of O*NET GWAs into task groups O*NET GWAs Categorisation Information input 1 Searching for, collecting and receiving information Task groups Interpersonal Creative & decision making Mental Processes 2 Identifying and evaluating relevant information 3 Information and data processing 7 Work output Interacting with others 4 5 6 Reasoning and decision making Performing physical and manual work activities Performing complex and technical activities Activities in (1) Activities in (2) Activities in (3) Activities in (4) Activities in (5) Activities in (6) Activities in (7) mapped to: mapped to: mapped to: mapped to: mapped to: mapped to: mapped to: 100% 25% 50% 100% 50% Information analysis 100% 25% Information synthesis 75% 25% ow we translate results of the data analysis context? into the Australian is the thecan results of theUnpredictable USthe data analysis intoUSthe Australian This is the context? This 50% physical 40% cond challenge this report overcomes. It determines how work patterns have changed in the Predictable physical 50% 10% report overcomes. It determines how work patterns have changed in the Total 100% 100% 100% 100% 100% 100% stralian economy by matching US occupations with their Australian equivalent using concordance y matching US occupations with their Australian equivalent using concordance 100% bles. To complete the picture and determine Australian it combines the he picture and determine Australian workplace trends, itworkplace combines trends, the cupational data with ABS statistics on hours worked by occupation. h ABS statisticsSOURCE: on hours worked by occupation. O’NET, Alpha Beta analysis 38. There are possible permutations which will not have infinite solutions. For example, assume all occupations perform task 1 exactly 40 times a week, is report goes one stepreports further than other reports dealing with automation’s impact onperform. the andother perform no other tasks: the only solution for the system would be that task 1the takes precisely 1 hour to tep further than dealing with automation’s impact on orkforce. It measures how much of the automation-driven change in work patterns is due to s how much of the automation-driven change in work patterns is due to There are possible permutations whichsolutions. will not have infinite solutions. example, assume all occupations ermutations which will not have infinite For example, assume For all occupations erform 1 exactly timesno a week, and perform nosolution other tasks: thesystem only solution for the system would be 40 timestask a week, and 40 perform other tasks: the only for the would be hat task 1 takes precisely 1 hour to perform. ely 1 hour to perform. 31 APPENDICES How can we translate the results of the US data analysis into the Australian context? This is the second challenge this report overcomes. It determines how work patterns have changed in the Australian economy by matching US occupations with their Australian equivalent using concordance tables. To complete the picture and determine Australian workplace trends, it combines the occupational data with ABS statistics on hours worked by occupation. This report goes one step further than other reports dealing with automation’s impact on the workforce. It measures how much of the automation-driven change in work patterns is due to workers changing jobs, and how much is due to workers simply changing the way they work within the same job. This analysis produces a novel finding: automation in recent years has mostly bs, and how much due toinworkers simplynot changing wayExhibit they17 work within led to aischange work activities, a change the in jobs. describes the method of this analysis. nalysis produces a novel finding: automation in recent years has mostly led to a ities, not a change in jobs. Exhibit 17 describes the method of this analysis. EXHIBIT 17 a matrix of timeshares of tasks in to thethe economy where to the timeshare spent on task type meshares of tasks Let in thebeeconomy where 𝑠𝑠𝑖𝑖,𝑗𝑗 corresponds timeshare spent oncorresponds task (interpersonal, creative, physical predictable, etc.) in occupation creative, physical predictable, etc.) in occupation 𝑗𝑗. For all 𝐽𝐽 occupations in the . For all occupations in the Australian economy, the the vector of total vector share of hoursofspent tasks (𝑡𝑡𝑡𝑡)on is tasks given( by:) is given by: of work total share workon hours spent 𝑠𝑠1,1 [ ⋮ 𝑠𝑠6,1 𝑆𝑆 ⋯ ⋱ ⋯ ℎ 𝑡𝑡𝑡𝑡 𝑠𝑠1,𝐽𝐽 𝑡𝑡𝑡𝑡1 ℎ1 ⋮ ]×[ ⋮ ] = [ ⋮ ] 𝑡𝑡𝑡𝑡𝐽𝐽 𝑠𝑠6,𝐽𝐽 ℎ𝐽𝐽 vector of fractions of total work hours for each occupation; the elements of ℎ sum to 1. In the above, is the vector of fractions of total work hours for each occupation; the elements of sum to 1. of 𝑆𝑆 and ℎ at different points in time allows for a detailed analysis of changes within and Taking observations of and at different points in time allows for a detailed analysis of changes within and between mple, by taking the difference between 𝑆𝑆2014 ×ℎ2014 − 𝑆𝑆2006 ×ℎ2006 the total change in the total change in hours spent on each task jobs. For example, by taking the difference between ask is given. The change attributable only to changes within a job, and not from workers is given. The change attributable only to changes within a job, and not from workers changing jobs, is determined by rmined by comparing 𝑆𝑆2014 ×ℎ2006 − 𝑆𝑆2006 ×ℎ2006 . comparing . es are extrapolatedThese to 2015, 2030 and 2000 estimate thetoimpact of automation 15various estimates aretoextrapolated 2015, 2030 and 2000 toinestimate the impact of automation in 15year periods. ix B: The5.2impact of automation on work quality APPENDIX B: THE IMPACT OF AUTOMATION ON WORK QUALITY Estimating theAustralian impact of automation ct of automation on the workforceon the Australian workforce Prior work on automation typically attempts to answer the broad question of how many occupations will be impacted by ation typicallyautomation. attemptsThis to answer theanswers broadthat question of however, how many occupations report also question, it also answers another important, but mostly overlooked question: automation. This also answers thatquality question, however, also answers how report will automation impact the of working lives? Ititmeasures the change in work activity patterns, as detailed in Appendix A,question: and combines results with extensive datathe from ABS and but mostly overlooked howthe will automation impact quality ofHILDA39 surveys. Consequently, this report makes three new contributions: it quantifies howas automation work safer, asures the change in work activity patterns, detailedmakes in Appendix A, more and enjoyable, and more valuable. 39 surveys. with extensive ABS and HILDA Consequently, this report ABSdata data from on workplace injuries reveal that the vast majority of injuries are associated with physical tasks—even when using a as shown inmakes Exhibitwork 18. The impact of automation on safety was measured by estimating how accident ntributions: itconservative quantifies estimate, how automation safer, more enjoyable, numbers would fall if the observed change in work activity patterns continued and the Australian workforce continued to grow at a constant rate. ce injuries reveal that the vast majority of injuries are associated with physical sing a conservative estimate, as shown in Exhibit 18. The impact of automation ured by estimating how accident numbers would fall if the observed change in ns continued and the Australian workforce continued to grow at the a constant rate. 39. Household, Income, and Labour Dynamics in Australia, Melbourne Institute, University of Melbourne 32 EXHIBIT 18 Estimating the impact of automation on workplace injury Steps • ABS data on number of injuries resulting in missed working days, classified by a description of cause, was used • ABS workplace injuries are classified as resulting in 1-4 days of missed work, or 5+ days of missed work • It is assumed that injuries resulting in 1-4 days of missed work lead to an average of 2.5 missed days of work, and injuries resulting in 5+ days of missed work lead to an average of 10 days of missed work • Causes of injury are assigned for task groups as follows ABS description of causes of workplace injuries Assigned to task group1 Lifting, pushing, pulling, or bending Routine/non-routine physical tasks Repetitive movement with low muscle loading Routine/non-routine physical tasks Prolonged standing, working in cramped or unchanging positions Split proportionally across all task groups Vehicle accident Split proportionally across all task groups Hitting or being hit by an object or vehicle Split proportionally across all task groups Fall on same level (including slip or fall) Split proportionally across all task groups Fall from a height Split proportionally across all task groups Exposure to mental stress Split proportionally across all task groups Contact with a chemical or substance Split proportionally across all task groups Other Split proportionally across all task groups • The implied number of injuries associated with a given task is calculated from the above • The expected reduction in injuries is calculated by estimating the number of injuries using 2030 projected timeshares, e.g.: Routine physical injuries 2030 = TS Routine phys 2030 TS Routine phys 2015 x Routine physical injuries 2015 Note: This method holds labour force size constant to control for labour force trend growth in injuries 1 Where there is uncertainty surrounding the "task" associated with an injury cause, causes are assigned proportionally to all tasks in the economy. If causes were assigned in greater detail, the number of physically-caused injuries would be even greater SOURCE: ABS, O*NET, AlphaBeta analysis HILDA data on job satisfaction across Australian occupations was used to assess the impact of automation on work quality. For that purpose, the available data was regressed on time spent on different tasks, as shown in Exhibit 19. Ordinary Least Squares (OLS) regression results show that workers performing a greater share of non-automatable tasks achieved higher satisfaction scores. The next step was to examine changes in timeshares across the different occupations (as detailed in Appendix A) and derive implications for job satisfaction. The result: automation has caused low-skilled work to change most rapidly towards including more satisfying tasks, suggesting that automation will disproportionately improve job satisfaction amongst the lowestskilled workers in Australia. 33 APPENDICES thirdreplacing part of the analysis waswork guided by the qu art of the analysis was guided by the question of The whether automatable with non-automatable work generate more va utomatable work will generate more value and higher wages for workers. ABSwill weekly wage datafor was regressed onand timeshares was regressed on timeshares of different tasks, controlling hours worked age of of differen EXHIBIT 19 workers as shown in Exhibit 20. shown in Exhibit 20. Exhibit 19 Estimating the impact of automation on satisfaction Steps Exhibit 20 • Satisfaction scores on a scale of 1-10 are provided for each 4 digit occupation through HILDA surveys • For each observation of individuals in occupation i, the satisfaction score is assumed to be a function of timeshares on activity types: • In the above, physical task shares are excluded due to multicollinearity • is interpreted as the expected satisfaction of an individual that only engages in physical work, and s as the premium over physical work (or discount if negative) in satisfaction scores of performing a given task at work instead of physical work • An ordinary least squares regression is used to estimate the value of coefficients, and expected satisfaction scores are calculated for 2006 and 2014 for each occupation • The expected change in satisfaction across time is estimated for occupation terciles (grouped by share of automatable tasks performed at work in 2006) • For each tercile of workers, an expected share of workers who could experience higher satisfaction as a result of increased automation is determined by comparing 2014 to 2006 expected satisfaction Note: Since the direction of outcome is measured instead of magnitude, there is no need to separately estimate satisfaction for 2015 and 2030 SOURCE: HILDA, O*NET, AlphaBeta analysis The third part of the analysis was guided by the question of whether replacing automatable work with non-automatable work will generate more value and higher wages for workers. ABS weekly wage data was regressed onanalysis timeshares of different tasks, controlling hours automatable worked andwork age of The third part of the was guided by the question of whetherfor replacing with workers as shown in Exhibit 20. non-automatable work will generate more value and higher wages for workers. ABS weekly wage data was regressed on timeshares of different tasks, controlling for hours worked and age of workers as shown in Exhibit 20. Exhibit 20 35 34 35 The third part of the analysis was guided by the question of whethe The third partautomatable of the analysis was guided by the que e third part of the analysis was guided bywith the question of whether replacing work non-automatable work will generate more value and higher w th non-automatable work will generate more value and higher wages for workers. ABS weekly with non-automatable work will generate more val wage data was regressed on timeshares of different tasks, controlli hird part of the analysis was guided by the question of whether replacing automatable work ge data was regressed on timeshares of different tasks, controlling for hours worked and age of timeshares wage data was regressed on of differen The third of20 the analysis was guided by question of whether automatable work workers asthe shown in higher Exhibit 20.replacing orkers as shown in part Exhibit 20. EXHIBIT nalysis was guided by the question of whether replacing automatable work wages non-automatable work will generate more value and for workers. ABS weekly e third part the analysis wasanalysis guided by the question of whether automatable work ABS workers as shown inworkers. Exhibit 20.weekly Theofthird part of the was guided by the question ofhigher whether replacing automatable work with non-automatable work will generate value andreplacing wages for work will generate more on value and higher wages formore workers. ABS weekly data was regressed timeshares of different tasks, controlling for hours worked andofage of th non-automatable work will generate more valueofand higher wages for workers. ABS weekly with non-automatable workaffect will generate more valuetasks, and higher wages workers. ABSand weekly wage dataofwill was regressed timeshares different forfor hours worked age hibit How automation worker wages Exhibit ed on20 timeshares different tasks,on controlling for20 hours worked andcontrolling age of ge data was on timeshares of different of tasks, controlling hours worked and age of and age of wageregressed data was regressed on20. timeshares different tasks,for controlling for hours worked ers as shown inasExhibit shown in20. Exhibit Exhibit 20 Steps xhibit 20.workers orkers as workers shown inasExhibit shown20. in Exhibit 20. • ABS data on average weekly earnings is provided for each occupation Exhibit 20 it 20 hibit 20 Exhibit 20each occupation i, the expected weekly wage is assumed to be a function of time spent on activity types, average • For hours worked in occupation i, and average age of workers in occupation i: • Physical task shares are excluded due to multicollinearity. In the above can be interpreted as the expected weekly wage (controlling for age and hours worked) of an individual that only engages in physical work, and as the expected premium • (or discount if is negative) of performing a given task at work instead of a physical task • An ordinary least squares regression is used to estimate the value of coefficients and expected wages are calculated for a worker performing 100% of a given task, assuming the worker is of the overall average worker age, working the total average weekly hours, e.g. for interpersonal: • Share of value for each task is calculated for the aggregate economy. E.g. for interpersonal: • In the above performing 100% of a given task is the vector of expected wages for a worker • In the above timeshares for tasks in the economy is the vector of • The results show that on average non-automatable work has a wage premium of 24% compared to automatable work Note: Since the direction of outcome is measured instead of magnitude, there is no need to separately estimate satisfaction for 2015 and 2030 35 SOURCE: ABS, O*NET, AlphaBeta analysis 35 35 35 35 The results show incomeshare shareof ofnon-automatable non-automatable work relative to the totaltotal timetime spentspent on such The results show thatthat thethe income work relative to the on work is substantially higher than the ratio for automatable work, suggesting that non-automatable work such work is substantially higher than the ratio for automatable work, suggesting that nonpays a wage premium of around 20 per cent compared to automatable work. The premium is determined automatable work pays a wage premium of around 20 per cent compared to automatable work. The using the following equation: premium is determined using the following equation: 35 % 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑓𝑓𝑓𝑓𝑓𝑓 𝑛𝑛𝑛𝑛𝑛𝑛-𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 % 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑜𝑜𝑜𝑜 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 = × % 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑓𝑓𝑓𝑓𝑓𝑓 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 % 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑜𝑜𝑜𝑜 𝑛𝑛𝑛𝑛𝑛𝑛-𝑎𝑎𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 5.3 Appendix C: Evaluating the potential gains from automation Estimating the impact of automation on the Australian economy The methodology used in this report differs to existing reports, as it provides a quantifiable estimate of the impact of automation on productivity in recent years. Beginning with the average weekly work hours in 2000, the change in time spent on automatable tasks through to 2015 is converted into a more tangible figure: work hours saved. From this figure, the implied gain in labour productivity can be derived. The analysis also accounts for the fact that some workers will reinvest the time saved through automation by assuming that non-automatable work generates 20 per cent more value (in line with 35 APPENDICES 5.3 APPENDIX C: EVALUATING THE POTENTIAL GAINS FROM AUTOMATION Estimating the impact of automation on the Australian economy The methodology used in this report differs to existing reports, as it provides a quantifiable estimate of the impact of automation on productivity in recent years. Beginning with the average weekly work hours in 2000, the change in time spent on automatable tasks through to 2015 is converted into a more tangible figure: work hours saved. From this figure, the implied gain in labour productivity can be  derived. The analysis also accounts for the fact that some workers will reinvest the time saved through automation by assuming that non-automatable work generates 20 per cent more value (in line with the wage premium). The gains from automation are then compared with total productivity growth as calculated from ABS data of labour hours worked and national GDP. Exhibit 21 shows that as much as 36 per cent of average productivity growth since the year 2000 was driven by automation. Remaining productivity gains are attributable to factors such as improved worker education and other efficiency gains. EXHIBIT 21 If automation continues at historical rates it will account for 36% of average labour productivity growth out to 2030 Expected labour productivity growth at historical rate Output per hour worked, real 2015 A$ 12 Automation is expected to account for 36% of labour productivity growth 103 7 84 2015 GDP per hour worked Productivity gains due to automation1 Productivity gains due education and other factors2 Includes gains due to time saved and reinvestment of labour hours into higher value added activities Productivity gains due to other factors inferred by removing productivity growth attributable to automation from total productivity growth 2000-2015 SOURCE: ABS, O*NET, AlphaBeta analysis 36 2030 GDP per hour worked This method of estimating the impact of automation has a clear advantage: it disentangles the impact of various types of productivity gains on employment. For example, some means to improve productivity, such as educating workers, are unlikely to adversely impact the number of workers employed. However, automation can result in worker displacement with a range of possible outcomes. Exhibit 22 details two cases—one where automation simply displaces workers, the other where workers reinvest their freed-up time elsewhere in the economy. Comparing such scenarios provides a novel way to quantify the potential gains from automation. EXHIBIT 22 GDP growth could be 2.4-3.0% p.a. out to 2030, depending on whether labour is transitioned or displaced Implied real GDP growth rate % p.a. (CAGR) Contribution to real GDP growth from 2016-30 Real 2015 $A billions Workers transitioned scenario 300 1,637 2015 GDP 0 450 30 2,577 160 Labour force growth Productivity growth from nonautomation factors Time savings due to automation 3.0% CAGR Change in hours worked per capita Uplift due to the higher value nature of new work activities 2030 GDP worked -160 -60 2,327 Workers displaced scenario 300 450 1,637 2015 GDP 160 Labour force growth Productivity growth from nonautomation factors Time savings due to automation 2.4% CAGR Change in hours worked per capita Uplift due to the higher value nature of new work activities 2030 GDP worked 1 Inferred from labour force entry rate of 1.9%, assumed annual migration contribution to labour force of 150,000 and labour force exit rate of 1.4% 2 Assumes time savings are reinvested into activities that generate 20% more GVA per hour than automated activities 3 Lost earnings could result due to lower labour earnings (caused by unemployment or involuntary reductions in hours), or due to labour shortages in occupations expected to grow as a result of automation SOURCE: ABS, O*NET, AlphaBeta analysis 37 APPENDICES 5.4 Appendix D: Evaluating the impact of automation for different groups of workers This analysis evaluates the potential impact of automation on three worker groups. The size and output contribution of each group is determined by combining the three parts of the earlier analysis: how automation changes the time spent on work tasks in occupations; whether it changes the value of work as people switch to more non-automatable tasks; and how it affects the size of the workforce and productivity. The process of determining group sizes is detailed in Exhibit 23. EXHIBIT 23 Which workers belong in which group, and how does group size change over time? Which workers belong in which group? The 2015 labour force was divided into occupations in 2015 and ranked in order of share of automatable work. Workers were divided into two groups High-risk current workers • Comprise 1/3 of the workforce • Average 70% of time spent on automatable work • Labour productivity in 2015 equal to 95% of economy wide average productivity1 Low-risk current workers • Comprise 2/3 of the workforce • Average 18% of time spent on automatable work • Labour productivity in 2015 equal to 103% of economy wide average productivity1 Future workers • Baseline estimate of 1/3 high-risk and 2/3 low-risk workers • Proportion of high and low risk workers may change depending on policy choices and exogenous factors How does group size change over time? Current worker are assumed to exit the workforce based on proportion population exceeding working age of 65 Current workers in 2015 Current workers in 20XX Current workers in 2015 Current workers in 2015 Future worker cohort size for 20XX is estimated by taking the overall projected labour force size for 20XX and subtracting current workers in 20XX Two rates of automation were used to measure the value and productivity gain from automation for each group (see Exhibit 24):40 • Constant automation: automation proceeds by historical rates • Accelerated automation: automation is accelerated until 2020 to catch up to global leaders Both automation rates were applied to different scenarios of worker displacement and transition. Modelling for future workers includes changes in the proportion of high- and low-skilled workers. An example of determining a scenario’s NPV is illustrated in Exhibit 25. Differences in the NPV of transition/non-transition scenarios were then used to determine the value of successful policy for each group. This approach breaks down the aggregate trends observed in Appendix C into worker subgroups, while also determining automation’s specific value for each worker group. By analysing which segment of the Australian workforce has most to gain from automation, this report provides a clear picture of how policy actions must be tailored to target the Australian workforce today and in the future, and what the value of assisting different worker group might be. 40. Labour productivity growth rates due to automating existing work are assumed to be the same for both groups, even if initial level of productivity is not. Productivity growth rates differ when labour is reinvested (As high-risk workers have greater potential for reinvesting labour in higher-value added activities. 38 y actions must be tailored to target the Australian workforce today and in the ns must tailoreddifferent to targetworker the Australian workforce value ofbe assisting group might be. today and in the of assisting different worker group might be. EXHIBIT 24 or the year 20XX under no transition is given by the following equation: year 20XX underLabour no transition is given the20XX following productivity for theby year under noequation: transition is given by the following equation: 𝑃𝑃𝑖𝑖,2015 𝑃𝑃𝑖𝑖,20𝑋𝑋𝑋𝑋𝑃𝑃𝑖𝑖,2015 = 1 − ̅̅̅̅̅̅̅̅̅ 𝐹𝐹20𝑋𝑋𝑋𝑋,𝑗𝑗 𝑃𝑃𝑖𝑖,20𝑋𝑋𝑋𝑋 = 1 − ̅̅̅̅̅̅̅̅̅ 𝐹𝐹20𝑋𝑋𝑋𝑋,𝑗𝑗 ̅̅̅̅̅̅̅̅̅ or high-risk, and 𝐹𝐹 20𝑋𝑋𝑋𝑋,𝑗𝑗 is the economy wide average fraction of automated hours in ̅̅̅̅̅̅̅̅̅ h-risk, and 𝐹𝐹 is the economy wide average fraction of automated hours in 20𝑋𝑋𝑋𝑋,𝑗𝑗 nario 𝑗𝑗. Where is either low or high-risk, and is the economy wide average fraction of automated hours in year 20XX under scenario . kers are able to reinvest their lost hours into new work, and labour productivity is given e able to reinvestUnder theirtransition, lost hours into new work, and labour given workers are able to reinvest theirproductivity lost hours intoisnew work, and labour productivity is given by: 𝑃𝑃𝑖𝑖,2015 (1 + 𝑣𝑣 ∗ 𝐹𝐹𝑖𝑖,20𝑋𝑋𝑋𝑋,𝑗𝑗 ) 𝑃𝑃𝑖𝑖,20𝑋𝑋𝑋𝑋 = (1 + 𝑣𝑣 ∗ 𝐹𝐹𝑖𝑖,20𝑋𝑋𝑋𝑋,𝑗𝑗 ) 𝑃𝑃𝑖𝑖,2015 1 − ̅̅̅̅̅̅̅̅̅ 𝐹𝐹20𝑋𝑋𝑋𝑋,𝑗𝑗 𝑃𝑃𝑖𝑖,20𝑋𝑋𝑋𝑋 = 1 − ̅̅̅̅̅̅̅̅̅ 𝐹𝐹20𝑋𝑋𝑋𝑋,𝑗𝑗 Where is the expected premium of non-automated work overisautomated work, and ed premium of non-automated work over automated work, and 𝐹𝐹 the fraction 𝑖𝑖,20𝑋𝑋𝑋𝑋,𝑗𝑗 mium of non-automated work over automated work, and 𝐹𝐹 is the fraction group in the 𝑗𝑗year under the year 20XX automated under scenario for20XX group 𝑖𝑖 scenario for𝑖𝑖,20𝑋𝑋𝑋𝑋,𝑗𝑗 ar 20XX under scenario 𝑗𝑗 for group 𝑖𝑖 is the fraction of hours EXHIBIT 25 How is the Net Present Value of output of each group calculated? The NPV for each group is calculated by estimating the output for each group of workers through to 2030. An example for low risk workers under no worker transition is shown below. Estimating the NPV of output for low risk workers under constant automation and displacement Constant automation and worker displacement: low risk output NPV 2016-2030 2016 GDP NPV 20XX GDP NPV 2030 GDP NPV Hours worked in 2016 by Low risk workers 2016 GDP per hour of low risk workers Discount factor 2016 Hours worked in 2030 by Low risk workers 2030 GDP per hour of low risk workers Discount factor 2030 Is equal to number of low risk workers in 2016, multiplied by 2015 hours worker per capita multiplied by (1-F)1 Is equal to 2015 average GDP per hour worked by low risk worker productivity index in 2016 under no transition Equal to 97% raised to the power of years elapsed since 2015 Is equal to number of low risk workers in 2030, multiplied by 2015 hours workd per capita multiplied by (1-F)1 Is equal to 2015 average GDP per hour worked by low risk worker productivity index in 2030 under no transition Equal to 97% raised to the power of years elapsed since 2015 Note: The above process is carried out for each year separately 2016-2030, and for each group of workers. 1 "F" is the fraction of work hours automated for a given group in a given year since 2015 SOURCE: AlphaBeta analysis 40 40 39 alphaBeta Strategy xeconomics