Ai, Automation, and the Future of Work: Ten Things to Solve For

AI, AUTOMATION, AND THE FUTURE OF WORK: TEN THINGS TO SOLVE FOR

BRIEFING NOTE PREPARED FOR THE TECH4GOOD SUMMIT, ORGANIZED BY THE FRENCH PRESIDENCY JUNE 2018

Automation and artificial intelligence (AI) are 1. Accelerating progress in AI and transforming businesses and will contribute to automation is creating opportunities for economic growth via contributions to productivity. businesses, the economy, and society They will also help address “moonshot” societal Automation and AI are not new, but recent challenges in areas from health to climate change. technological progress is pushing the frontier of At the same time, these technologies will transform what machines can do. Our research suggests the nature of work and the workplace itself—which that society needs these improvements to provide is the focus of this briefing note. Machines will value for businesses, contribute to economic be able to carry out more of the tasks done by growth, and make once unimaginable progress humans, complement the work that humans do, on some of our most difficult societal challenges.1 and even perform some tasks that go beyond what In summary: humans can do. As a result, some occupations Rapid technological progress will decline, others will grow, and many more will Beyond traditional industrial automation and change. While we believe there will be enough advanced robots, new generations of more work to go around (barring extreme scenarios), capable autonomous systems are appearing in society will need to grapple with significant environments ranging from autonomous vehicles workforce transitions and dislocation. Workers on roads to automated check-outs in grocery will need to acquire new skills and adapt to the stores.2 Much of this progress has been driven increasingly capable machines alongside them by improvements in systems and components, in the workplace. They may have to move from including mechanics, sensors, and software. declining occupations to growing and, in some AI has made especially large strides in recent cases, new occupations. This briefing note, which years, as machine-learning algorithms have draws on the latest research from the McKinsey become more sophisticated and made use of Global Institute, examines both the promise huge increases in computing power and of the and the challenge of automation and AI in the exponential growth in data available to train workplace and outlines some of the critical issues algorithms. Spectacular breakthroughs are that policy makers, companies, and individuals will making headlines, many involving beyond-human need to solve for. capabilities in computer vision, natural language Challenges remain before these technologies processing, and complex games such as Go. can live up to their potential for the good of the economy and society Potential to transform businesses and AI and automation still face challenges. The contribute to economic growth limitations are partly technical, such as the need for These technologies are already generating value massive training data and difficulties “generalizing” in various products and services, and companies algorithms across use cases. Recent innovations across sectors use them in an array of processes are just starting to address these issues. Other to personalize product recommendations, find challenges are in the use of AI techniques. For anomalies in production, identify fraudulent example, explaining decisions made by machine transactions, and more. The latest generation of learning algorithms is technically challenging, AI advances, including techniques that address which particularly matters for use cases involving classification, estimation, and clustering problems, financial lending or legal applications. Potential promises significantly more value still. An analysis bias in the training data and algorithms, as well as we conducted of several hundred AI use cases data privacy, malicious use, and security are all found that the most advanced deep learning issues that must addressed.8 Europe is leading techniques deploying artificial neural networks with the new General Data Protection Regulation, could account for as much as $3.5 trillion to which codifies more rights for users over data $5.8 trillion in annual value, or 40 percent of the collection and usage. A different sort of challenge value created by all analytics techniques.3 concerns the ability of organizations to adopt Deployment of AI and automation technologies these technologies, where people, data availability, can do much to lift the global economy and technology, and process readiness often make increase global prosperity, at a time when aging it difficult. Adoption is already uneven across and falling birth rates are acting as a drag on sectors and countries. The finance, automotive, growth. Labor productivity growth, a key driver of and telecommunications sectors lead AI adoption. economic growth, has slowed in many economies, Among countries, US investment in AI ranked first dropping to an average of 0.5 percent in 2010–14 at $15 billion to $23 billion in 2016, followed by from 2.4 percent a decade earlier in the United Asia’s investments of $8 billion to $12 billion, with States and major European economies, in the Europe lagging at $3 billion to $4 billion.9 aftermath of the 2008 financial crisis after a 2. How AI and automation will affect work previous productivity boom had waned. AI and Even as AI and automation bring benefits to automation have the potential to reverse that business and society, we will need to prepare for decline: productivity growth could potentially major disruptions to work. reach 2 percent annually over the next decade, with 60 percent of this increase from digital About half of the activities (not jobs) carried out opportunities.4 by workers could be automated Our analysis of more than 2000 work activities Potential to help tackle several societal across more than 800 occupations shows that “moonshot” challenges certain categories of activities are more easily AI is also being used in areas ranging from material automatable than others. They include physical science to medical research and climate science. activities in highly predictable and structured Application of the technologies in these and other environments, as well as data collection and data disciplines could help tackle societal “moonshot” processing. These account for roughly half of the challenges.5 For example, researchers at Geisinger activities that people do across all sectors. The have developed an algorithm that could reduce least susceptible categories include managing diagnostic times for intracranial hemorrhaging others, providing expertise, and interfacing by up to 96 percent.6 Researchers at George with stakeholders. Washington University, meanwhile, are using machine learning to more accurately weight the climate models used by the Intergovernmental Panel on Climate Change.7

2 McKinsey Global Institute AI, automation, and the future of work:Ten things to solve for Nearly all occupations will be affected by including rising incomes, increased spending automation, but only about 5 percent of on healthcare, and continuing or stepped- occupations could be fully automated by up investment in infrastructure, energy, and currently demonstrated technologies. Many more technology development and deployment. These occupations have portions of their constituent scenarios showed a range of additional labor activities that are automatable: we find that about demand of between 21 percent to 33 percent of 30 percent of the activities in 60 percent of all the global workforce (555 million and 890 million occupations could be automated. This means that jobs) to 2030, more than offsetting the numbers most workers—from welders to mortgage brokers of jobs lost. Some of the largest gains will be to CEOs—will work alongside rapidly evolving in emerging economies such as India, where machines. The nature of these occupations will the working-age population is already growing likely change as a result. rapidly.11

Jobs lost: Some occupations will see significant Additional economic growth, including from declines by 2030 business dynamism and rising productivity Automation will displace some workers. We growth, will also continue to create jobs. Many have found that around 15 percent of the global other new occupations that we cannot currently workforce, or about 400 million workers, could be imagine will also emerge and may account for as displaced by automation in the period 2016–30. much as 10 percent of jobs created by 2030, if This reflects our mid-point scenario in projecting history is a guide. Moreover, technology itself has the pace and scope of adoption. Under the fastest historically been a net job creator. For example, scenario we have modeled, that figure rises to the introduction of the personal computer in the 30 percent, or 800 million workers. In our slowest 1970s and 1980s created millions of jobs not just adoption scenario, only about 10 million people for semiconductor makers, but also for software would be displaced, close to zero percent of the and app developers of all types, customer service global workforce.10 representatives, and information analysts.

The wide range underscores the multiple factors Jobs changed: More jobs than those lost that will impact the pace and scope of AI and or gained will be changed as machines automation adoption. Technical feasibility of complement human labor in the workplace automation is only the first influencing factor. Partial automation will become more prevalent as Other factors include the cost of deployment; machines complement human labor. For example, labor-market dynamics, including labor supply AI algorithms that can read diagnostic scans with a quantity, quality, and the associated wages; the high degree of accuracy will help doctors diagnose benefits beyond labor substitution that contribute patient cases and identify suitable treatment. In to business cases for adoption; and, finally, social other fields, jobs with repetitive tasks could shift norms and acceptance. Adoption will continue toward a model of managing and troubleshooting to vary significantly across countries and sectors automated systems. At retailer Amazon, because of differences in the above factors, employees who previously lifted and stacked especially labor-market dynamics: in advanced objects are becoming robot operators, monitoring economies with relatively high wage levels, such as the automated arms and resolving issues such as France, Japan, and the United States, automation an interruption in the flow of objects.12 could displace 20 to 25 percent of the workforce 3. Key workforce transitions and challenges by 2030, in a midpoint adoption scenario, more While we expect there will be enough work to than double the rate in India. ensure full employment in 2030 based on most of Jobs gained: In the same period, jobs will also our scenarios, the transitions that will accompany be created automation and AI adoption will be significant. The Even as workers are displaced, there will be mix of occupations will change, as will skill and growth in demand for work and consequently educational requirements. Work will need to be jobs. We developed scenarios for labor demand to redesigned to ensure that humans work alongside 2030 from several catalysts of demand for work, machines most effectively.

McKinsey Global Institute AI, automation, and the future of work:Ten things to solve for 3 Digital platforms, the gig economy, and the changes workforce skills, such platforms are rise of tech-enabled independent work becoming part of an essential suite of HR recruiting The rise of digital talent platforms, the gig tools. To harness them, companies will need to economy, and tech-enabled independent work are take a more strategic look at their talent needs, and also affecting the future of work. They are already adapt their human resources function to align it having a transformative effect on some sectors, more clearly with the CEO agenda.2 and they have the potential to help address some Digital platforms can also give a boost to of the labor markets’ challenges in matching independent work. MGI research finds that 20 to jobs to workers and in signaling information to 30 percent of the working age population in the prospective employers. At the same time, they United States and the European Union is engaged challenge some entrenched ways of working and, in independent work, with 70 percent of those in some countries, the workings of social systems. doing so out of preference. While only about Digital talent platforms create transparency and 15 percent of independent work is conducted efficiency in labor markets. Surveys by LinkedIn on digital platforms now, that proportion is find that workers using digital platforms are eight growing rapidly. Independent workers span all times more likely to be at the same company demographic groups: about half of senior earners after two years and 11 percent more satisfied have participated in independent work, and youth than in their previous jobs. By improving worker make up about a quarter of the independent satisfaction across the economy, these platforms workforce.3 While those who pursue independent can drive productivity. About 40 percent of work (digitally enabled or not) out of preference respondents to the surveys said digital platforms are generally satisfied, those who pursue it out of helped them secure a job they would not have necessity are unsatisfied with the income variability otherwise found. By drawing more people into and the lack of benefits typically associated with more formal employment, these platforms can traditional work. Policy makers and innovators will raise labor-force participation. MGI estimates that need to grapple with solutions to these challenges. these effects together could contribute $2.7 trillion to global GDP annually by 2025.1 As automation 2 Ram Charan, Dominic Barton, and Dennis Carey, Talent wins; The new playbook for putting people first, Harvard Business Review Press, 2018. 1 A labor market that works: Connecting talent with 3 Independent work: Choice, necessity, and the gig opportunity in the digital age, June 2015. economy, October 2016.

Workers will need different skills to thrive in the Many workers will likely need to workplace of the future change occupations Automation will accelerate the shift in required Our research suggests that, in a mid-point workforce skills we have seen over the past scenario, around 3 percent of the global workforce 15 years. Demand for advanced technological will need to change occupational categories by skills such as programming will grow rapidly. 2030, though scenarios range from about 0 to Social, emotional, and higher cognitive skills, 14 percent. Some of these shifts will happen such as creativity, critical thinking, and complex within companies and sectors, but many will information processing, will also see growing occur across sectors and even geographies. demand. Basic digital skills demand has been Occupations made up of physical activities increasing and that trend will continue and in highly structured environments or in data accelerate. Demand for physical and manual processing or collection will see declines. Growing skills will decline, but will remain the single largest occupations will include those with difficult to category of workforce skills in 2030 in many automate activities such as managers, and those countries.13 This will put additional pressure on the in unpredictable physical environments such already existing workforce skills challenge, as well as plumbers. Other occupations that will see as the need for new credentialing systems. While increasing demand for work include teachers, some innovative solutions are emerging, solutions nursing aides, and tech and other professionals. that can match the scale of the challenge will be needed.

4 McKinsey Global Institute AI, automation, and the future of work:Ten things to solve for Workplaces and workflows will change as more remained stagnant for about half a century despite people work alongside machines rising productivity—a phenomenon known as As intelligent machines and software are integrated “Engels’ Pause,” after the German philosopher more deeply into the workplace, workflows and who identified it. workspaces will continue to evolve to enable 4. Ten things to solve for humans and machines to work together. As In the search for appropriate measures and self-checkout machines are introduced in stores, policies to address these challenges, we should for example, cashiers can become checkout not seek to roll back or slow diffusion of the assistance helpers, who can help answer technologies. Companies and governments questions or troubleshoot the machines. More should harness automation and AI to benefit from system-level solutions will prompt rethinking of the enhanced performance and productivity the entire workflow and workspace. Warehouse contributions as well as the societal benefits. design may change significantly as some These technologies will create the economic portions are designed to accommodate primarily surpluses that will help societies manage robots and others to facilitate safe human- workforce transitions. Rather, the focus should be machine interaction. on ways to ensure that the workforce transitions Automation will likely put pressure on average are as smooth as possible. This is likely to require wages in advanced economies more actionable and scalable solutions in several The occupational mix shifts will likely put pressure key areas: on wages. Many of the current middle-wage Ensuring robust economic and productivity jobs in advanced economies are dominated growth. Strong growth is not the magic answer by highly automatable activities, such as in for all the challenges posed by automation, manufacturing or in accounting, which are likely but it is a pre-requisite for job growth and to decline. High-wage jobs will grow significantly, increasing prosperity. Productivity growth is a especially for high-skill medical and tech or other key contributor to economic growth. Therefore, professionals, but a large portion of jobs expected unlocking investment and demand, as well to be created, including teachers and nursing as embracing automation for its productivity aides, typically have lower wage structures. The contributions, is critical. risk is that automation could exacerbate wage polarization, income inequality, and the lack of Fostering business dynamism. income advancement that has characterized the Entrepreneurship and more rapid new business past decade across advanced economies, stoking formation will not only boost productivity, but social, and political tensions.14 also drive job creation. A vibrant environment for small businesses as well as a competitive In the face of these looming challenges, environment for large business fosters business workforce challenges already exist dynamism and, with it, job growth. Accelerating Most countries already face the challenge of the rate of new business formation and the adequately educating and training their workforces growth and competitiveness of businesses, to meet the current requirements of employers. large and small, will require simpler and evolved Across the OECD, spending on worker education regulations, tax and other incentives. and training has been declining over the last two decades. Spending on worker transition Evolving education systems and learning and dislocation assistance has also continued for a changed workplace. Policy makers to shrink as a percentage of GDP. One lesson of working with education providers (traditional the past decade is that while globalization may and non-traditional) and employers themselves have benefited economic growth and people as could do more to improve basic STEM skills consumers, the wage and dislocation effects through the school systems and improved on workers were not adequately addressed. on-the-job training. A new emphasis is needed Most analyses, including our own, suggest that on creativity, critical and systems thinking, and the scale of these issues is likely to grow in the adaptive and life-long learning. There will need coming decades. We have also seen in the past to be solutions at scale. that large-scale workforce transitions can have a Investing in human capital. Reversing the lasting effect on wages; during the 19th century trend of low, and in some countries, declining Industrial Revolution, wages in the United Kingdom public investment in worker training is critical.15

McKinsey Global Institute AI, automation, and the future of work:Ten things to solve for 5 Through tax benefits and other incentives, safety nets are available, and should be policy makers can encourage companies to adopted and adapted, while new approaches invest in human capital, including job creation, should be considered and tested. learning and capability building, and wage Investing in drivers of demand for work. growth, similar to incentives for the private Governments will need to consider stepping up sector to invest in other types of capital, investments that are beneficial in their own right including R&D. and will also contribute to demand for work Improving labor market dynamism. (e.g. infrastructure, climate change adaptation). Information signals that enable matching of These types of jobs, from construction to workers to work, credentialing, could all work rewiring buildings and installing solar panels, better in most economies. Digital platforms are often middle-wage jobs, those most can also help match people with jobs and affected by automation. restore vibrancy to the labor market. When Embracing AI and automation safely. Even more people change jobs, even within a as we capture the productivity benefits of company, evidence suggests that wages these rapidly evolving technologies, we need rise.16 As more varieties of work and income- to actively guard against the risks and mitigate earning opportunities emerge, including the gig any dangers. The use of data must always take economy, we will need to solve for issues such into account concerns, including data security, as portability of benefits, worker classification, privacy, malicious use, and potential issues of and wage variability.17 bias, issues that policy makers, tech and other Redesigning work. Workflow design and firms, and individuals will need to find effective workspace design will need to adapt to a ways to address. new era in which people work more closely ••• with machines. This is both an opportunity and a challenge, in terms of creating a safe There is work for everyone today and there will and productive environment. Organizations be work for everyone tomorrow, even in a future are changing too, as work becomes more with automation. Yet that work will be different, collaborative and companies seek to become requiring new skills, and a far greater adaptability increasingly agile and non-hierarchical. of the workforce than we have seen. Training and retraining both midcareer workers and new Rethinking incomes. If automation (full or generations for the coming challenges will be an partial) does result in a significant reduction imperative. Government, private sector leaders, in employment and/or greater pressure on and innovators all need to work together to better wages, some ideas such as conditional coordinate public and private initiatives, including transfers, support for mobility, universal basic creating the right incentives to invest more in income, and adapted social safety nets could human capital. The future with automation and be considered and tested. The key will be to AI will be challenging, but a much richer one if find solutions that are economically viable we harness the technologies with aplomb—and and incorporate the multiple roles that work mitigate the negative effects. plays for workers, including providing not only income, but also meaning, purpose, This briefing note was written by James Manyika, and dignity. chairman and director of the McKinsey Global Institute and a senior partner at McKinsey & Rethinking transition support and safety Company, based in San Francisco; and Kevin nets for workers affected. As work evolves Sneader, McKinsey & Company’s global managing at higher rates of change between sectors, partner-elect, based in Hong Kong. locations, activities, and skill requirements, many workers will need assistance adjusting. McKinsey Global Institute research reports are Many best practice approaches to transition available on www.mckinsey.com/mgi.

6 McKinsey Global Institute AI, automation, and the future of work:Ten things to solve for Further reading Endnotes 1 David Autor, “Why are there still so many jobs? Recent MGI reports on automation and the future of work, including A future that works: Automation, employment, The history and future of workplace automation,” and productivity, January 2017 and Jobs lost, jobs Journal of Economic Perspectives, Summer 2015. gained: Workforce transitions in a time of automation, December 2017. David Autor and Anna Salomons, “Does 2 Disruptive technologies: Advances that will transform life, productivity growth threaten employment?” business, and the global economy, May 2013. working paper prepared for ECB Forum on Central 3 Notes from the AI frontier: Insights from hundreds of use Banking, June 2017. cases, April 2018. 4 Solving the productivity puzzle: The role of demand and the Erik Brynjolffson and Andrew McAfee, The second promise of digitization, February 2018. machine age: Work, progress, and prosperity in a 5 Nicola Nosengo, “Can artificial intelligence create the next time of brilliant technologies, WW Norton, 2014. wonder material?” Nature, May 4, 2016. Jason Furman “Should we be reassured if 6 Mohammad R. Arbabshirani et al., “Advanced machine learning in action: identification of intracranial hemorrhage automation in the future looks like automation in the on computed tomography scans of the head with clinical past?” in NBER book The Economics of Artificial workflow integration,” npj Digital Medicine, volume 1, article Intelligence: An Agenda, Ajay K. Agrawal, Joshua 9, April 2018. Gans, and Avi Goldfarb, ed., NBER, forthcoming. 7 Nicola Jones, “How machine learning could help to improve climate forecasts,” Nature, August 23, 2017. Terry Gregory, Anna Salomons, and Ulrich 8 Michael Chui, James Manyika, and Mehdi Miremadi, “What Zierhahn, Racing with or against the machine? AI can and can’t do (yet) for your business,” McKinsey Evidence from Europe, Centre for European Quarterly, January 2018. Economic Research, discussion paper 16-053, 9 Artificial intelligence: The next digital frontier? June 2017. July 2016. 10 Jobs lost, jobs gained: Workforce transitions in a time of automation, December 2017. William R. Kerr, Allison Ciechanover, and Jeff 11 Ibid. Huizinga, Managing the future of work, Harvard 12 Nick Wingfield, “As Amazon pushes forward with robots, Business School, May 2018. workers find new roles,” The New York Times, September 10, 2017. Ljubice Nedelkoska and Glenda Quintini, 13 Skill shift: automation and the future of the workforce, Automation, skills use and training, OECD social, May 2018. employment and migration working papers, 14 Poorer than their parents? Flat or falling incomes in number 202, March 2018. advanced economies, July 2016. 15 Public spending on labour markets, OECD. 16 A labor market that works: Connecting talent with opportunity in the digital age, June 2015. 17 Independent work: Choice, necessity, and the gig economy, October 2016.