Chapter 1. The challenge computers pose to work and education

This chapter sets out the context for carrying out an exploratory project on the challenge that computer capabilities will pose for work and education in the future. The project uses the OECD’s Survey of Adult Skills (PIAAC) to carry out this exploration. An overview of existing research on past and future trends in skill demand is initially discussed. Examples are drawn mainly from economics, with additional perspectives from education and computer science. Detailing how this study aims to build upon existing research while offering a new approach, this initial chapter offers a roadmap for subsequent sections of the report.


Computer scientists are working on reproducing all human skills. The development of these computer capabilities will have far-reaching implications for work and education.1 In order to respond, the structure of the economy and the skills of the workforce will need to be radically transformed over the 21st century. Although we cannot know exactly how this transformation will proceed, we can make significant progress in understanding its shape over the next few decades by assessing the full extent of current computer capabilities. In many cases, these capabilities have not yet been widely applied.

By knowing which computer capabilities are now available and how they relate to human skills, we can better understand which work tasks can potentially be automated in the near future. This understanding can provide the basis for constructing realistic scenarios about the ways that jobs and skill demand will be redefined in the next few decades. This will help policymakers understand how the education system needs to be shaped in turn to prepare today’s students for those possible futures.

This report describes an exploratory project to understand computer capabilities with respect to one set of human skills in the context of work and education. The project used the OECD’s Survey of Adult Skills, derived from the Programme for the International Assessment of Adult Competencies (PIAAC), as a tool for understanding the implications of growing computer capabilities.

PIAAC measures a set of general cognitive skills – literacy, numeracy, and problem solving with computers2 – that receive extensive development during compulsory education.3 Countries invest in developing these skills because they are widely used by adults both at work and in their personal lives: they are “necessary for fully integrating and participating in the labour market, education and training, and social and civic life” and “relevant to many social contexts and work situations” (OECD 2016, p.16). PIAAC measures these skills precisely because of their acknowledged importance both as outputs of the education system and as inputs in the workplace. This report looks at changes in the use of these skills over the past decades and explores the implications of computers for further changes in the future. The exploratory analysis described in this report provides the first step towards constructing an ongoing and comprehensive programme to assess the capabilities of computers and their implications for work and education.

The project outlined in this report draws upon extensive research in economics, with additional perspectives from education and computer science. This first chapter situates this report within the context of these literatures and also describes the structure of the report.

Past employment trends in skill demand

The transformation of work skills has been a key aspect of global economic development over the past centuries. The nature of this transformation is well known, involving the long-term shift of employment out of agriculture into manufacturing initially, and then into services. The shift is accompanied by large increases in educational attainment. It is helpful to remember the scale of this transformation: in the United Kingdom for example, employment in services increased from 41% of the workforce in 1890 to 72% in 1998 (Maddison, 2003, Table 2-24), while average educational attainment increased from 4.8 years to 13.1 years (Van Zanden et al., 2014, Table 5.5). Different countries are at different stages in this transformation of education and work skills, but the transformation is occurring worldwide and has continuing implications for government policies related to human capital.

The overall shift in employment and increasing demand for education are related to technological change, with new technology during the 20th century tending on average to increase the demand for higher skills and decrease the demand for lower skills. This basic historical relationship between technology and education suggests the metaphor of a “race” between technology and education (Goldin and Katz, 2008; Tinbergen, 1974). The general conclusion that technological change is driving the economy towards ever-increasing demands for education is widely accepted. However, the metaphor does not apply to the 19th century, when the most salient effect of technological change was to decrease the demand for skills by replacing skilled craft workers with unskilled workers in factories (Acemoglu, 2002).

In the late 20th century, the change in skill demand became more complex, with the emergence for several decades of a pattern of “polarisation”. This term refers to increasing employment for workers with higher and lower skills and decreasing employment for workers with mid-level skills. This pattern of job change has been found for the United States, many European countries and Japan (Autor, Katz and Kearney, 2006; Goos, Manning and Salomons, 2009; Ikenaga and Kambayashi, 2010). However, these trends will not necessarily continue. Already, there are questions about whether the pattern of polarised employment changes has continued in recent years, especially with respect to findings of weakness in the demand for workers with higher skills since 2000 (Autor, 2015; Beaudry, Green and Sand, 2016).

Although there are several possible reasons for the trend towards polarisation in the labour market, the strongest explanation is that technology is increasingly being used to perform routine tasks that were previously performed by workers (Goos, Manning and Salomons, 2014; Michaels, Natraj and Van Reenen, 2014). According to this explanation, jobs involving routine cognitive tasks typically occur in the middle of the skill distribution and are susceptible to substitution by technology. On the other hand, jobs involving non-routine tasks that cannot yet be carried out by technology occur either at the low or high end of the skill distribution, depending on whether the non-routine tasks require physical or cognitive skills (Autor, Levy and Murnane, 2003).

In general, the research looking at skill polarisation uses indirect measures that are available in economic datasets, with nothing like the skill assessments that are used in education research. Typically, “skill” is inferred indirectly from wages, education or occupation. For example, Autor, Katz and Kearney (2006) use three different measures of skill by occupation: mean level of education by occupation; median level of hourly wages by occupation; and data on tasks taken from brief occupational descriptions. Such measures can be useful as rough indicators of skill in the context of economic analyses. However they are far removed from direct measures of skill.

As technology advances, automation of certain skills is raising questions about the changes in the quantity of jobs that will be needed in the future, not just the skill distribution required across the workforce. Historically, there have been periodic waves of concern about unemployment resulting from new technology. This stretches back at least two centuries to the early industrial revolution (Mokyr, Vickers and Ziebarth, 2015). However, unemployment resulting from technological displacement in the past has always been temporary, with increased productivity leading to decreased prices, which lead in turn to increased demand by consumers for both old and new goods, and then to increased demand by employers for workers (Autor, 2015). This adjustment process may not necessarily take place as smoothly as suggested by economic theory (Autor, Dorn and Hanson, 2016), leaving room for policy interventions to help move displaced workers to new occupations. However, the overall historical experience is one of large-scale and successful redeployment of the workforce as technology shifts the skills needed in different production processes.

In response to several high-profile studies suggesting the possibility of substantial job displacement resulting from computers (Brynjolfsson and McAfee, 2014; Frey and Osborne, 2013), a number of economists have recently looked directly at the employment implications of these technologies. Several studies have specifically looked at the impact of recent applications of computers on employment resulting from applications of computers (Bessen, 2015; Falk and Biagi, 2015; Graetz and Michaels, 2015). In general, these studies have found familiar conclusions about the job effects of this technology. In short, firms, occupations and industries that use higher levels of computers in production experience higher productivity and employment. On the other hand, a recent study looking at the effects of trade and offshoring on jobs in the United States found that it was computer use in occupations, not trade or offshoring that led to increased risk of unemployment (Ebenstein, Harrison and McMillan, 2015).

Projected future trends in skill demand

Looking forwards, several recent economic studies have developed theoretical frameworks to explore the potential effects of computers on employment, wages and productivity in the future. These studies extend earlier theoretical work from a period when the technology was substantially less advanced (e.g., Elliott, 1998; Simon, 1977; Zeira, 1998). Acemoglu and Restrepo (2016) develop a model where automation can displace workers in performing older tasks, but where there is endless creation of new and more complex tasks that workers can perform better than machines. Benzell, Kotlikoff, Lagarda and Sachs (2015) develop a model where computers can automate analytical tasks, but not empathetic or interpersonal tasks. Sachs and colleagues (2015) compare a two-task model, where computers can automate only one of the tasks, with a one-task model where the only task in the economy can be performed by both people and machines. These different models project a variety of results for workers, some positive and some negative, with the results depending on assumptions about the fundamental relationship between computer capabilities and the skills needed to perform tasks in the economy.

Several studies have used information from computer science to understand the relationship between computer capabilities and the skills and tasks in the economy. The most widely cited study is by Frey and Osborne (2013), which estimates that 47% of employment in the United States is at a high risk of automation over the next several decades. The Frey and Osborne analysis involves four steps. First, a group of computer scientists classified 70 occupations as either automatable or not, based on a set of job descriptions and their knowledge of current computer capabilities.4 The occupations chosen were those where the group was most confident in making this judgment. Second, the authors identified job tasks that were most likely to be barriers to computerisation – perception and manipulation, creative intelligence, social intelligence – based on the current state of computer science. Third, occupational data on these hard-to-automate tasks were used to develop a model to predict the “automatability” classification of the 70 occupations from their tasks. This model was then used to predict the automatability of all occupations in the United States economy. Finally, jobs above a predicted automatability of 70% were defined as “high risk”. Figures of employment by occupation were then used to derive the overall estimate of 47%. Although the model was originally developed for the United States, it has been applied to a variety of other countries by substituting different figures for occupational employment in the last step (e.g., Frey, Osborne and Holmes, 2016; Pajarinen, Rouvinen, and Ekeland, 2015).

Further work supported by the OECD has developed an alternative way of extending Frey and Osborne’s automatability judgments for the original 70 occupations to the full economy (Arntz, Gregory and Zierahn, 2016). This research suggests an estimated 9% of jobs in OECD countries are highly automatable, a dramatically lower figure from that estimated by Frey and Osborne from the same starting point. Unlike Frey and Osborne, who use job tasks that are hard to automate as the basis for extending the automatability rating beyond the original 70 occupations, the analysis by Arntz, Gregory and Zierahn uses information on a wide range of job tasks, job characteristics, worker skills and worker characteristics from the OECD’s Survey of Adult Skills (PIAAC). This information includes factors that are not directly related to job tasks such as gender, age, education, proficiency in literacy and numeracy, firm size and income. In addition, the job task information includes not only factors that are similar to the hard-to-automate tasks identified by Frey and Osborne, but also other tasks that may not be hard to automate, such as filling forms or calculating percentages. Arntz and colleagues attribute their differing results to a “task-based approach” that acknowledges the variation of job tasks within an occupation, using the job task information in PIAAC. However, they do not provide any analyses to prove that it is the variation within occupation by tasks that explains the substantial difference in their results from those of Frey and Osborne. Another plausible explanation of their result is that it comes from using a very different set of job features that includes many things that are not job tasks at all. Their report does not include any results for models that use job tasks alone as the basis for the extrapolation.

The McKinsey Global Institute recently issued a report analysing work automatability using an approach that focuses on judgments related to 18 capabilities that are mapped to more than 2 000 work activities in the United States and other countries (Manyika et al., 2017). The report estimates that 49% of the activities at work could be automated with current technology. The report does not describe how the judgments related to the 18 capabilities were obtained. The capability judgments were mapped to work activities by a process using the key words in the titles of the work activities.

Another approach to predicting job automatability focuses on worker skills rather than job tasks or activities. Although these two approaches should be complementary, a focus on skills may be more meaningful to the education community. Elliott (2017) uses a sample of recent articles from the computer science literature to identify computer capabilities in four rough skill areas – language, reasoning, movement and vision – that can be mapped to a set of worker descriptors in occupational data for the United States (O*NET). The descriptions of computer capabilities from the research literature are compared to the anchoring tasks on the O*NET scales. The resulting analysis suggests that 82% of current United States employment is potentially automatable, based upon the types of capabilities discussed in contemporary computer science literature.

All of these studies making projections about the potential automatability of current jobs using current technology include caveats about the various economic, institutional and social factors that affect the application of technology. Extensive literature explores the factors that influence the diffusion of innovation (Rogers, 1995). When diffusion does occur it can often take several decades or more (Comin and Hobijn, 2010; Griliches, 1957; Mansfield, 1961), even though diffusion speed has increased in recent years. None of the studies of potential automatability focus on an exact timeframe. All of them refer loosely to computer applications that could potentially happen over a period of several decades. Of course, it is quite possible that some potential applications of current technology will not occur over this period. It is also possible that even more advanced technology will be developed. All of the studies note that projection of automatability of a percentage of jobs, occupations or tasks over several decades does not mean that the people who currently perform those activities will become unemployed, or even that they will change jobs. Instead, the crucial question relates to the ways in which their activities and required skills will be redefined, whether or not their job or occupation changes over this period.

Separately within computer science, there is also work to assess the capabilities of current computer techniques with standardised tests developed for people. Early versions of this work go back several decades (O’Neil and Baker, 1994). This research encompasses several different types of tests, including elementary and secondary school tests in science and mathematics (Clark and Etzioni, 2016), verbal IQ tests for young children (Ohlsson et al., 2015) and university entrance exams (Arai and Matsuzaki, 2014). To work towards the goal of a broader development and assessment of artificial intelligence, computer scientists have proposed formal assessments in social-emotional intelligence (Jarrold and Yeh, 2016), physical perception and action (Ortiz, 2016), visual interpretation (Zitnick et al., 2016) and common sense reasoning (Davis, 2016).

Plan for the study and report

This study builds on this prior research related to past and future trends in work skills by using the OECD’s Survey of Adult Skills (PIAAC) to look at the use of general cognitive skills in the workplace. Chapter 2 looks to history, using PIAAC to describe the distribution of proficiency in the workforce, the use of skill and the ways these have changed over the past two decades. Chapters 3 and 4 outline the approach of using PIAAC to measure computer capabilities against workforce skills, with Chapter 3 describing the development and Chapter 4 the results. These two chapters provide a way of understanding a technology that could change skill demand in the decades ahead. Chapter 5 discusses the implications of computer capabilities for the future of the skill changes discussed in Chapter 2 and considers the policy implications for education.


Acemoglu, D., (2002), “Technical Change, Inequality and the Labor Market”, Journal of Economic Literature, Vol. 40/1, American Economic Association, pp. 7-72.

Acemoglu, D., and P. Restrepo (2016), “The Race Between Man and Machine: Implications of Technology for Growth, Factor Shares and Employment”, National Bureau of Economic Research, Working Paper No. 22252.

Arai, N.H., and T. Matsuzaki (2014), “The Impact of AI on Education – Can a Robot Get Into the University of Tokyo?”, in C.-C. Liu et al., eds., Proceedings of the 22nd International Conference on Computers in Education, Japan: Asia-Pacific Society for Computers in Education.

Arntz, M., T. Gregory, and U. Zierahn (2016), “The Risk of Automation for Jobs in OECD Countries: A Comparative Analysis”, OECD Social, Employment and Migration Working Papers, No. 189, OECD Publishing, Paris.

Autor, D.H. (2015), “Why Are There Still So Many Jobs? The History and Future of Workplace Automation”, Journal of Economic Perspectives, Vol. 29/3, pp. 3-30.

Autor, D.H., D. Dorn, and G.H. Hanson (2016), “The China Shock: Learning from Labor Market Adjustment to Large Changes in Trade”, National Bureau of Economic Research, Working Paper No. 21906.

Autor, D.H., L.F. Katz and M.S. Kearney (2006), “The Polarization of the U.S. Labor Market”, American Economic Review, Vol. 9/2, pp.189-194.

Autor, D.H., F. Levy, and R.J. Murnane (2003), “The Skill Content of Recent Technological Change: An Empirical Exploration”, Quarterly Journal of Economics, Vol. 118/4, pp. 1279-1333.

Beaudry, P., D.A. Green, and B.M. Sand (2016), “The Great Reversal in the Demand for Skill and Cognitive Tasks”, Journal of Labor Economics, Vol. 34/1, pp. S199-S247.

Benzell, S.G. et al. (2015), “Robots are Us: Some Economics of Human Replacement”, National Bureau of Economic Research, Working Paper No. 20941.

Bessen, J.E. (2015), “How Computer Automation Affects Occupations: Technology, Jobs and Skills”, Boston University School of Law, Law and Economics Research Paper No. 15-49, available at

Brynjolfsson, E., and A. McAfee (2014), The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies, W.W. Norton & Company, New York.

Clark, P., and O. Etzioni (2016), “My Computer Is an Honor Student – But How Intelligent Is It? Standardized Tests as a Measure of AI”, AI Magazine, Vol. 37/1, Association for the Advancement of Artificial Intelligence, pp. 5-12.

Comin, D., and B. Hobijn (2010), “An Exploration of Technology Diffusion”, American Economic Review, Vol. 100/5, pp. 2031-59.

Davis, E. (2016), “How to Write Science Questions That Are Easy for People and Hard for Computers”, AI Magazine, Vol. 37/1, Association for the Advancement of Artificial Intelligence, pp. 13-22.

Ebenstein, A., A. Harrison and M. McMillan (2015), “Why are American Workers Getting Poorer? China, Trade and Offshoring”, National Bureau of Economic Research, Working Paper No. 21027.

Elliott, S.W. (2017), “Projecting the Impact of Information Technology on Work and Skills in the 2030s”, in J. Buchanan, D. Finegold, K. Mayhew and C. Warhurst, eds., The Oxford Handbook of Skills and Training, Oxford University Press.

Elliott, S.W. (1998), “Computer Technology, Human Labor, and Long-Run Economic Growth”, Heinz School Working Paper, No. 98-23, Carnegie Mellon University.

Falk, M., and F. Biagi (2015), “Empirical Studies on the Impacts of ICT Usage in Europe”, Joint Research Centre of the European Commission, Institute for Prospective Technological Studies, Digital Economy Working Paper, No. 2015/14, JRC98693.

Frey, C.B., and M.A. Osborne (2013), “The Future of Employment: How Susceptible are Jobs to Computerization?”, Oxford Martin School, Working Paper, September, University of Oxford.

Frey, C.B., M.A. Osborne, and C. Holmes (2016), “Technology at Work v2.0: The Future is Not What It Used to Be”, Citi GPS: Global Perspectives and Solutions and Oxford Martin School, Working Paper, January, Citi and University of Oxford.

Goldin, C., and L. Katz (2008), The Race Between Education and Technology, Harvard University Press, Cambridge, MA.

Goos, M., A. Manning, and A. Salomons (2014), “Explaining Job Polarization: Routine-Biased Technological Change and Offshoring”, American Economic Review, Vol. 104/8, pp. 2509-2526.

Goos, M., A. Manning, and A. Salomons (2009), “Job Polarization in Europe”, American Economic Review: Papers & Proceedings, Vol. 99/2, American Economic Association, pp. 58-63.

Graetz, G., and G. Michaels (2015), “Robots at Work”, CEP Discussion Paper, No. 1335, Centre for Economic Performance, London.

Griliches, Z. (1957), “Hybrid Corn: An Exploration in the Economics of Technological Change”, Econometrica, Vol. 25/4, the Econometric Society, New York, pp. 501-522.

Ikenaga, T., and R. Kambayashi (2010), “Long-term Trends in the Polarization of the Japanese Labour Market: The Increase of Non-routine Task Input and Its Valuation in the Labour Market”, Institute of Economic Research Working Paper, Hitotsubashi University, Tokyo.

Jarrold, W., and P.Z. Yeh (2016), “The Social-Emotional Turing Challenge”, AI Magazine, Vol. 37/1, Association for the Advancement of Artificial Intelligence, pp. 31-38.

Maddison, A. (2003), The World Economy: A Millennial Perspective, OECD Publishing, Paris.

Mansfield, E. (1961), “Technical change and the rate of imitation”, Econometrica, Vol. 29/4, the Econometric Society, New York, pp. 741-66.

Manyika, J., M. et al. (2017), A Future That Works: Automation, Employment, and Productivity, McKinsey Global Institute, (accessed 23 January 2017).

Michaels, G., A. Natraj, and J. Van Reenen (2014), “Has ICT Polarized Skill Demand? Evidence from Eleven Countries Over Twenty-Five Years”, The Review of Economics and Statistics, Vol. 96/1, the President and Fellows of Harvard College and the Massachusetts Institute of Technology, pp. 60-77.

Mokyr, J., C. Vickers, and N.L. Ziebarth (2015), “The History of Technological Anxiety and the Future of Economic Growth: Is This Time Different?”, Journal of Economic Perspectives, Vol. 29/3, American Economic Association, pp. 31-50.

OECD (2016), The Survey of Adult Skills: Reader’s Companion, Second Edition, OECD Skills Studies, OECD Publishing, Paris,

Ohlsson, S. et al. (2015), Measuring an Artificial Intelligence System’s Performance on a Verbal IQ Test for Young Children, University of Illinois, Chicago.

O’Neil, H.F., and E.L. Baker, eds. (1994), Technology Assessment in Software Applications, Lawrence Erlbaum Associates, Hillsdale, New Jersey.

Ortiz, C.L. (2016), “Why We Need a Physically Embodied Turing Test and What It Might Look Like”, AI Magazine, Vol. 37/1, Association for the Advancement of Artificial Intelligence, pp.55-62.

Pajarinen, M., P. Rouvinen, and A. Ekeland (2015), “Computerization and the Future of Jobs in Norway”, Research Institute of the Finnish Economy and Statistics, Discussion Paper, Norway.

Rogers, E. (1995), Diffusion of Innovations, 4th edition, Free Press, Macmillan Publishing, New York.

Sachs, J.D., S.G. Benzell, and G. LaGarda (2015), “Robots: Curse or Blessing? A Basic Framework”, National Bureau of Economic Research Working Paper, No. 21091.

Simon, H.A. (1977), The New Science of Management Decision, revised edition, Prentice-Hall, Englewood Cliffs, New Jersey.

Tinbergen, J. (1974), “Substitution of graduate by other labour”, Kyklos, Vol. 27/2, pp. 217-26.

Van Zanden, J.L., et al., eds (2014), How Was Life?: Global Well-being Since 1820, OECD Publishing, Paris,

Zeira, J. (1998), “Workers, Machines, and Economic Growth”, Quarterly Journal of Economics, Vol. 113/4, pp. 1091-1117.

Zitnick, C.L, et al. (2016), “Measuring Machine Intelligence through Visual Question Answering”, AI Magazine, Vol. 37/1, Association for the Advancement of Artificial Intelligence, pp., 63-72.


← 1. Throughout this report, the term “computers” is used to refer generally to computers, robots, and other types of information and communications technologies.

← 2. The formal name used for the problem solving skill area in PIAAC is “problem solving in technology-rich environments.”

← 3. Most older adults who take PIAAC will not have received instruction related to problem solving with computers during compulsory education.

← 4. Specifically, the assessments analysed in Frey and Osborne (2013) “were based on eyeballing the O*NET tasks and job descriptions of each occupation” (p. 30).