The Technology Trap

by Carl Benedikt Frey

  53. L. Nedelkoska and G. Quintini, 2018, “Automation, Skills Use and Training” (OECD Social, Employment and Migration Working Paper 202, Organisation of Economic Co-operation and Development, Paris).

  54. One study by researchers at the University of Mannheim suggests that only 9 percent of jobs are exposed to automation. See M. Arntz, T. Gregory, and U. Zierahn, 2016, “The Risk of Automation for Jobs in OECD Countries” (OECD Social, Employment and Migration Working Paper 189, Organisation of Economic Co-operation and Development, Paris). And more recently, a study by the OECD estimates that 14 percent of jobs are at risk of being replaced. See L. Nedelkoska and G. Quintini, 2018, “Automation, Skills Use and Training” (OECD Social, Employment and Migration Working Paper 202, Organisation of Economic Co-operation and Development, Paris). The intuition behind these studies and ours is that we can infer the automatability of jobs by analyzing the tasks they entail. However, instead of relying primarily on tasks, the Mannheim study also incorporated demographic variables such as sex, education, age, and income. Because women and college-educated people, for example, tend to work in occupations that are less exposed to automation, their approach means that a female taxi driver with a PhD is less likely to be displaced by autonomous vehicles than a man who has been driving a taxi for many decades. In practice, however, this seems unlikely to be true. Aware of this problem, the authors of the OECD study followed our approach and relied on tasks rather than worker characteristics. But like the Mannheim study, the OECD study used individual-level data from the Programme for the International Assessment of Adult Competencies (PIAAC) survey instead of occupational averages. This approach allows the authors to distinguish between workers within occupations who might perform slightly different tasks. The drawback is that they have to rely on broader occupational categories, lumping many different occupations together, which means that valuable information is lost, as the OECD study rightly points out (Nedelkoska and Quintini, 2018, “Automation, Skills Use and Training”). What’s more, that study regrettably does not provide any detail on any within-occupation variation, which suggests that other things are likely to be more relevant in explaining the differences between their results and ours. Indeed, it is hard to believe that the tasks performed by different truck drivers (or workers in other occupations) vary that greatly. In the end, the only reasonable way to check whether their model or ours is preferable is how well they perform on that training set (the OECD study also used our training data set). A frequently used metric to assess this is the area under the curve (AUC), and by this measure the nonlinear model in our study is much more accurate than their linear model. For a detailed discussion of how and why these estimates differ, see also C. B. Frey and M. Osborne, 2018, “Automation and the Future of Work—Understanding the Numbers,” Oxford Martin School, https://www.oxfordmartin.ox.ac.uk/opinion/view/404.

  55. See, for example, Arntz, Gregory, and Zierahn, 2016, “The Risk of Automation for Jobs in OECD Countries,” table 5.

  56. Council of Economic Advisers, 2016, “2016 Economic Report of the President,” chapter 5, https://obamawhitehouse.archives.gov/sites/default/files/docs/ERP_2016_Chapter_5.pdf.

  57. J. Furman, forthcoming, “Should We Be Reassured If Automation in the Future Looks Like Automation in the Past?,” in Economics of Artificial Intelligence, ed. Ajay K. Agrawal, Joshua Gans, and Avi Goldfarb (Chicago: University of Chicago Press), 8.

  58. M. Ford, 2015. Rise of the Robots: Technology and the Threat of a Jobless Future (New York: Basic Books), introduction, Kindle.

  59. D. Remus and F. Levy, 2017, “Can Robots Be Lawyers? Computers, Lawyers, and the Practice of Law,” Georgetown Journal Legal Ethics 30 (3): 526.

  60. As we made clear, “we focus on estimating the share of employment that can potentially be substituted by computer capital, from a technological capabilities point of view, over some unspecified number of years. We make no attempt to estimate how many jobs will actually be automated. The actual extent and pace of computerisation will depend on several additional factors which were left unaccounted for” (C. B. Frey and Osborne, 2017, “The Future of Employment,” 268).

  61. See also D. H. Autor, 2014, “Skills, Education, and the Rise of Earnings Inequality among the ‘Other 99 Percent,’ ” Science 344 (6186): 843–51.

  62. W. K. Blodgett, 1918, “Doing Farm Work by Motor Tractor,” New York Times, January 6.

  63. D. P. Gross, 2018, “Scale Versus Scope in the Diffusion of New Technology: Evidence from the Farm Tractor,” RAND Journal of Economics 49 (2): 449.

  64. “17,000,000 Horses on Farms,” 1921, New York Times, December 30.

  65. T. Sorensen, P. Fishback, S. Kantor, and P. Rhode, 2008, “The New Deal and the Diffusion of Tractors in the 1930s” (Working paper, University of Arizona, Tucson).

  66. R. Solow, 1987, “We’d Better Watch Out,” New York Times Book Review, July 12; H. Gilman, 1987, “The Age of Caution: Companies Slow the Move to Automation,” Wall Street Journal, June 12.

  67. Quoted in ibid.

  68. See, for example, T. F. Bresnahan, E. Brynjolfsson, and L. M. Hitt, 2002, “Information Technology, Workplace Organization, and the Demand for Skilled Labor: Firm-Level Evidence,” Quarterly Journal of Economics 117 (1): 339–76; E. Brynjolfsson, L. M. Hitt, and S. Yang, 2002, “Intangible Assets: Computers and Organizational Capital,” Brookings Papers on Economic Activity 2002 (1): 137–81; E. Brynjolfsson and L. M. Hitt, 2000, “Beyond Computation: Information Technology, Organizational Transformation and Business Performance,” Journal of Economic Perspectives 14 (4): 23–48.

  69. M. Hammer, 1990, “Reengineering Work: Don’t Automate, Obliterate,” Harvard Business Review 68 (4): 104–12.

  70. On companies reengineering plans, see J. Rifkin, 1995, The End of Work: The Decline of the Global Labor Force and the Dawn of the Post-market Era (New York: G. P. Putnam’s Sons).

  71. P. A. David, 1990, “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox,” American Economic Review 80 (2): 355–61.

  72. For a detailed discussion, see R. J. Gordon, 2005, “The 1920s and the 1990s in Mutual Reflection” (Working Paper 11778, National Bureau of Economic Research, Cambridge, MA).

  73. S. D. Oliner and D. E. Sichel, 2000, “The Resurgence of Growth in the Late 1990s: Is Information Technology the Story?,” Journal of Economic Perspectives 14 (4): 3–22.

  74. W. D. Nordhaus, 2005, “The Sources of the Productivity Rebound and the Manufacturing Employment Puzzle” (Working Paper 11354, National Bureau of Economic Research, Cambridge, MA).

  75. In the period 1993–2007, robots are estimated to have accounted for bit more than one-tenth of overall growth in gross domestic product (GDP) across seventeen countries. See G. Graetz and G. Michaels, forthcoming, “Robots at Work,” Review of Economics and Statistics.

  76. J. Bughin et al., 2017, “How Artificial Intelligence Can Deliver Real Value to Companies,” McKinsey Global Institute, https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/how-artificial-intelligence-can-deliver-real-value-to-companies.

  77. It is true that many of the benefits brought by technology are unmeasured, which could in principle account for some of the productivity slowdown. In a recent study, the economists Austan Goolsbee and Peter Klenow used a novel approach to measure the value of internet-based technologies, examining the time people spend on the internet. Building on the intuition that consumption involves expenditure of both income and time, they estimated that the internet-related consumer surplus could be up to 3 percent (or $3,000 annually for the median person). See A. Goolsbee and P. Klenow, 2006, “Valuing Consumer Products by the Time Spent Using Them: An Application to the Internet,” American Economic Review 96 (2): 108–13. Chad Syverson recently extended their value-of-time analysis, using the American Time Use Survey and data on personal disposable income. Applying the 3 percent estimate of Goolsbee and Klenow, he calculated an internet-related consumer surplus of around
$3,900 per capita for 2105 (2017, “Challenges to Mismeasurement Explanations for the US Productivity Slowdown,” Journal of Economic Perspectives 31 [2]: 165–86). All the same, it is not clear that mismeasurement has become greater in the computer era. Indeed, the Boskin Commission, appointed by the U.S. Senate in 1995, also found evidence of substantial unmeasured quality improvements. The question of whether the recent productivity slowdown is an artifact of mismeasurement is thus not a question of whether mismeasurement exists but one of whether it has gotten larger in recent years. Economists have shown that the answer is no. While there is surely mismeasurement, it seems to have gotten smaller, not larger. Mismeasurement associated with prices of computer hardware and related services as well as intangible assets (such as patents, trademarks, and advertising expenditures) only make the productivity slowdown worse. The decline in domestic production of computer-related goods and services since the period 1995–2004 means that despite mismeasurement’s having worsened for some digital technologies, the mismeasurement problem was greater then than it is now. Together, these adjustments add 0.5 percentage point to the labor productivity numbers published for 1995–2004, but only 0.2 percentage point for 2004–14 (see D. M. Byrne, J. G. Fernald, and M. B. Reinsdorf, 2016, “Does the United States Have a Productivity Slowdown or a Measurement Problem?,” Brookings Papers on Economic Activity, 2016 [1]: 109–82). Even if we factor in high-end estimates of consumer benefits from free services like those of Wikipedia, Google, Facebook, and so on, this can account for only about a third of the slowdown. Syverson has calculated that if the productivity deceleration had not happened, measured GDP would have been 16 percent higher in 2015, adding $2.9 trillion to the U.S. economy. This amounts to $9,100 for every citizen or $23,400 for every household (2017, “Challenges to Mismeasurement Explanations for the US Productivity Slowdown”). The bottom line is that mismeasurement might be large, but it is not sufficiently large to account for the productivity slowdown. The productivity slowdown appears structural and real.

  78. Brynjolfsson, Rock, and Syverson, forthcoming, “Artificial Intelligence and the Modern Productivity Paradox,” 25.

  79. C. F. Kerry and J. Karsten, 2017, “Gauging Investment in Self-Driving Cars,” Brookings Institution, October 16. https://www.brookings.edu/research/gauging-investment-in-self-driving-cars/.

  80. Brynjolfsson, Rock, and Syverson, forthcoming, “Artificial Intelligence and the Modern Productivity Paradox,” 25.

  81. N. F. Crafts and T. C. Mills, 2017, “Trend TFP Growth in the United States: Forecasts versus Outcomes” (Discussion Paper 12029, Centre for Economic Policy Research, London). Their findings are consistent with the observation of Eric Bartelsman that productivity forecasts perform “horribly, with forecast standard errors being larger than ranges that would be useful for policy purposes” (2013, “ICT, Reallocation and Productivity” [Brussels: European Commission, Directorate-General for Economic and Financial Affairs]).

  82. H. Jerome, 1934, “Mechanization in Industry” (Working Paper 27, National Bureau of Economic Research, New York), 19.

  83. H. R. Varian, forthcoming, “Artificial Intelligence, Economics, and Industrial Organization,” in The Economics of Artificial Intelligence: An Agenda, ed. Ajay K. Agrawal, Joshua Gans, and Avi Goldfarb (Chicago: University of Chicago Press), 1.

  84. Ibid., 15.

  85. Brynjolfsson, Rock, and Syverson, forthcoming, “Artificial Intelligence and the Modern Productivity Paradox.”

  86. N. F. Crafts, 2004, “Steam as a General Purpose Technology: A Growth Accounting Perspective,” Economic Journal 114 (495): 338–51.

  87. Quoted in J. L. Simon, 2000, The Great Breakthrough and Its Cause (Ann Arbor: University of Michigan Press), 108.

  88. P. Colquhoun, 1815, A Treatise on the Wealth, Power, and Resources of the British Empire, Johnson Reprint Corporation), 68–69. See also J. Mokyr, 2011, The Enlightened Economy; Britain and the Industrial Revolution, 1700–1850 (London: Penguin), chapter 5, Kindle. I am indebted to Joel Mokyr for pointing me to this reference.

  89. Malthus, [1798] 2013, An Essay on the Principle of Population, 179.

  90. R. Henderson, 2017, comment on “Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics, by E. Brynjolfsson, D. Rock and C. Syverson,” National Bureau of Economic Research, http://www.nber.org/chapters/c14020.pdf.

  91. J. M. Keynes, [1930] 2010, “Economic Possibilities for Our Grandchildren,” in Essays in Persuasion (London: Palgrave Macmillan), 321–32.

  92. V. A. Ramey and N. Francis, 2009, “A Century of Work and Leisure,” American Economic Journal: Macroeconomics 1 (2): 189–224.

  93. W. A. Sundstrom, 2006, “Hours and Working Conditions,” in Historical Statistics of the United States, Earliest Times to the Present: Millennial Edition Online, ed. S. B. Carter et al. (New York: Cambridge University Press).

  94. Ramey and Francis, 2009, “A Century of Work and Leisure.”

  95. These estimates are based on the age-year specific leisure measures and survival probabilities. See ibid.

  96. These results depart somewhat from the estimates of Mark Aguiar and Erik Hurst, who find a larger increase in leisure for the period after 1965. The main reason is that they classify child care as leisure rather than home production. See M. Aguiar and E. Hurst, 2007, “Measuring Trends in Leisure: The Allocation of Time Over Five Decades,” Quarterly Journal of Economics 122 (3): 969–1006. Ramey and Francis also classify activities like talking to and playing with children as leisure, but they classify other child care tasks as home production. Given that people report much lower levels of enjoyment associated with such activities, this seems reasonable. See J. Robinson and G. Godbey, 2010, Time for Life: The Surprising Ways Americans Use Their Time (Philadelphia: Penn State University Press.)

  97. Keynes, [1930] 2010, “Economic Possibilities for Our Grandchildren,” 322.

  98. R. L. Heilbroner, 1966, “Where Do We Go from Here?,” New York Review of Books, March 17, https://www.nybooks.com/articles/1966/03/17/where-do-we-go-from-here/.

  99. D. H. Autor, 2015, “Why Are There Still So Many Jobs? The History and Future of Workplace Automation,” Journal of Economic Perspectives 29 (3): 8.

  100. Heilbroner, 1966, “Where Do We Go from Here?”

  101. B. Stevenson and J. Wolfers, 2013, “Subjective Well-Being and Income: Is There Any Evidence of Satiation?,” American Economic Review 103 (3): 598–604.

  102. H. Simon, 1966, “Automation,” New York Review of Books, March 26, https://www.nybooks.com/articles/1966/05/26/automation-3/.

  103. C. Stewart, 1960, “Social Implications of Technological Progress,” in Impact of Automation: A Collection of 20 Articles about Technological Change, from the Monthly Labor Review (Washington, DC: Bureau of Labor Statistics), 12.

  104. H. Voth, 2000, Time and Work in England 1750–1830 (Oxford: Clarendon Press of Oxford University Press).

  105. Aguiar and Hurst, 2007, “Measuring Trends in Leisure Measuring Trends in Leisure.”

  106. Quoted in C. Curtis, 1983, “Machines vs. Workers.” New York Times, February 8.

  107. F. Bastiat, 1850, “That Which Is Seen, and That Which Is Not Seen,” https://mises.org/library/which-seen-and-which-not-seen.

  108. D. Acemoglu and P. Restrepo, 2018b, “The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment,” American Economic Review 108 (6): 1488–542.

  109. T. Berger and C. B. Frey, 2017a, “Industrial Renewal in the 21st Century: Evidence from US Cities,” Regional Studies 51 (3): 404–13.

  110. Brynjolfsson and McAfee, 2014, The Second Machine Age, 11.

  111. A. Goolsbee, 2018, “Public Policy in an AI Economy” (Working Paper 24653, National Bureau of Economic Research, Cambridge, MA).

  Chapter 13

  1. See D. S. Landes, 1969, The Unbound Prometheus: Technological Change and Development in Western Europe from 1750 to the Present (Cambridge: Cambridge Uni
versity Press), 4.

  2. A. H. Hansen, 1939, “Economic Progress and Declining Population Growth,” American Economic Review 29 (1): 10–11.

  3. R. J. Gordon, 2016, The Rise and Fall of American Growth: The U.S. Standard of Living Since the Civil War (Princeton, NJ: Princeton University Press).

  4. Landes, 1969, The Unbound Prometheus, 4.

  5. F. Fukuyama, 2014, Political Order and Political Decay: From the Industrial Revolution to the Globalization of Democracy (New York: Farrar, Straus and Giroux), 450.

  6. On workers rationally opposing replacing technologies, see A. Korinek and J. E. Stiglitz, 2017, “Artificial Intelligence and Its Implications for Income Distribution and Unemployment” (Working Paper 24174, National Bureau of Economic Research, Cambridge, MA).

  7. A. Greif and M. Iyigun, 2013, “Social Organizations, Violence, and Modern Growth,” American Economic Review 103 (3): 534–38.

  8. Quoted in A. Greif and M.Iyigun, 2012, “Social Institutions, Violence and Innovations: Did the Old Poor Law Matter?” (Working paper, Stanford University, Stanford, CA), 4.

  9. Malthus wrote: “To remedy the frequent distresses of the common people, the poor laws of England have been instituted; but it is to be feared, that though they may have alleviated a little the intensity of individual misfortune, they have spread the general evil over a much larger surface.… The poor laws of England tend to depress the general condition of the poor in these two ways. Their first obvious tendency is to increase [the] population without increasing the food for its support.… Secondly, the quantity of provisions consumed in workhouses upon a part of the society that cannot in general be considered as the most valuable part diminishes the shares that would otherwise belong to more industrious and more worthy members, and thus in the same manner forces more to become dependent” ([1798] 2013, An Essay on the Principle of Population, 55 and 62–63, Digireads. com, Kindle). In similar fashion, Ricardo argued that “the clear and direct tendency of the poor laws, is in direct opposition to these obvious principles: it is not, as the legislature benevolently intended, to amend the condition of the poor, but to deteriorate the condition of both poor and rich.… This pernicious tendency of these laws is no longer a mystery, since it has been fully developed by the able hand of Mr. Malthus; and every friend to the poor must ardently wish for their abolition” ([1817] 1911, The Principles of Political Economy and Taxation. Reprint. London: Dent, 33).