Treating both busi-ness models equally would lead to more balanced incentives. In fact, given thepositive externalities of more widely shared prosperity, a case could be made fortreating wage income more favorably than capital income, for instance by expand-ing the earned income tax credit.

This spiral of marginalization can grow because concentration of econom-ic power often begets concentration of political power. In the words attributedto Louis Brandeis: “We may have democracy, or we may have wealth concen-trated in the hands of a few, but we can’t have both.” In contrast, when humansare indispensable to value creation, economic power will tend to be more decen-tralized.

The distributive effects of AI depend on whether it is primarily used to aug-ment human labor or automate it. When AI augments human capabilities, en-abling people to do things they never could before, then humans and machinesare complements. Complementarity implies that people remain indispensable forvalue creation and retain bargaining power in labor markets and in political deci-sion-making. In contrast, when AI replicates and automates existing human ca-pabilities, machines become better substitutes for human labor and workers loseeconomic and political bargaining power.

As machines becomebetter substitutes for human labor, workers lose economic and political bargainingpower and become increasingly dependent on those who control the technology. Incontrast, when AI is focused on augmenting humans rather than mimicking them,humans retain the power to insist on a share of the value created. What is more,augmentation creates new capabilities and new products and services, ultimatelygenerating far more value than merely human-like AI. While both types of AI canbe enormously beneficial, there are currently excess incentives for automation rath-er than augmentation among technologists, business executives, and policy-makers.

278Dædalus, the Journal of the American Academy of Arts & SciencesThe Turing Trap: The Promise & Peril of Human-Like Artificial IntelligenceIn 1988, robotics researcher Hans Moravec noted that “it is comparatively easyto make computers exhibit adult level performance on intelligence tests or play-ing checkers, and difficult or impossible to give them the skills of a one-year-oldwhen it comes to perception and mobility.”33 But I would argue that in many do-mains, Moravec was not nearly ambitious enough. It is often comparatively easierfor a machine to achieve superhuman performance in new domains than to matchordinary humans in the tasks they do regularly.Humans have evolved over millions of years to be able to comfort a baby, nav-igate a cluttered forest, or pluck the ripest blueberry from a bush. These tasksare difficult if not impossible for current machines. But machines excel when itcomes to seeing X-rays, etching millions of transistors on a fragment of silicon, orscanning billions of webpages to find the most relevant one. Imagine how feebleand limited our technology would be if past engineers set their sights on merelymatching human-levels of perception, actuation, and cognition.Augmenting humans with technology opens an endless frontier of new abili-ties and opportunities. The set of tasks that humans and machines can do togetheris undoubtedly much larger than those humans can do alone (Figure 1). Machinescan perceive things that are imperceptible to humans, they can act on objects inways that no human can, and, most intriguingly, they can comprehend things thatare incomprehensible to the human brain. As Demis Hassabis, CEO of DeepMind,put it, the AI system “doesn’t play like a human, and it doesn’t play like a program.It plays in a third, almost alien, way . . . it’s like chess from another dimension.”34Computer scientist Jonathan Schaeffer explains the source of its superiority: “I’mabsolutely convinced it’s because it hasn’t learned from humans.”35 More funda-mentally, inventing tools that augment the process of invention itself promises toexpand not only our collective abilities, but to accelerate the rate of expansion ofthose abilities.What about businesspeople? They often find that substituting machinery forhuman labor is the low-hanging fruit of innovation. The simplest approach is toimplement plug-and-play automation: swap in a piece of machinery for each taska human is currently doing. That mindset reduces the need for more radical chang-es to business processes.36 Task-level automation reduces the need to understandsubtle interdependencies and creates easy A-B tests, by focusing on a known taskwith easily measurable performance improvement.Similarly, because labor costs are the biggest line item in almost every company’sbudget, automating jobs is a popular strategy for managers. Cutting costs–whichcan be an internally coordinated effort–is often easier than expanding markets.Moreover, many investors prefer “scalable” business models, which is often a syn-onym for a business that can grow without hiring and the complexities that entails.But here again, when businesspeople focus on automation, they often set outto achieve a task that is both less ambitious and more difficult than it need be.151 (2) Spring 2022279Erik BrynjolfssonTo understand the limits of substitution-oriented automation, consider a thoughtexperiment. Imagine that our old friend Dædalus had at his disposal an extreme-ly talented team of engineers 3,500 years ago and built human-like machines thatfully automated every work-related task that his fellow Greeks were doing.9 Herding sheep? Automated.9 Making clay pottery? Automated.9 Weaving tunics? Automated.9 Repairing horse-drawn carts? Automated.9 Incense and chanting for victims of disease? Automated.The good news is that labor productivity would soar, freeing the ancientGreeks for a life of leisure. The bad news is that their living standards and healthoutcomes would come nowhere near matching ours. After all, there is only somuch value one can get from clay pots and horse-drawn carts, even with unlimit-ed quantities and zero prices.In contrast, most of the value that our economy has created since ancient timescomes from new goods and services that not even the kings of ancient empireshad, not from cheaper versions of existing goods.37 In turn, myriad new tasks areFigure 1Opportunities for Augmenting Humans Are Far Greater thanOpportunities to Automate Existing TasksNew Tasks ThatHumans Can Do withthe Help of MachinesTasks ThatHumans Can DoHuman TasksThat MachinesCould Automate280Dædalus, the Journal of the American Academy of Arts & SciencesThe Turing Trap: The Promise & Peril of Human-Like Artificial Intelligencerequired: fully 60 percent of people are now employed in occupations that did notexist in 1940. 38 In short, automating labor ultimately unlocks less value than aug-menting it to create something new.At the same time, automating a whole job is often brutally difficult. Every jobinvolves multiple different tasks, including some that are extremely challengingto automate, even with the cleverest technologies. For example, AI may be able toread mammograms better than a human radiologist, but it is not very good at theother twenty-six tasks associated with the job, according to O-NET, such as com-forting a concerned patient or coordinating on a care plan with other doctors.39My work with Tom Mitchell and Daniel Rock on the suitability for machine learn-ing analyzed 950 distinct occupations. We found that machines could perform atleast some tasks in most occupations, but zero in which machine learning coulddo 100 percent of the tasks.40The same principle applies to the more complex production systems that in-volve multiple people working together.41 To be successful, firms typically need toadopt a new technology as part of a system of mutually reinforcing organizationalchanges. 42 Consider another thought experiment: Imagine if Jeff Bezos had “au-tomated” existing bookstores by simply replacing all the human cashiers with ro-bot cashiers. That might have cut costs a bit, but the total impact would have beenmuted. Instead, Amazon reinvented the concept of a bookstore by combining hu-mans and machines in a novel way. As a result, they offer vastly greater productselection, ratings, reviews, and advice, and enable 24/7 retail access from the com-fort of customers’ homes. The power of the technology was not in automating thework of humans in the existing retail bookstore concept but in reinventing andaugmenting how customers find, assess, purchase, and receive books and, in turn,other retail goods.Third, policy-makers have also often tilted the playing field toward automat-ing human labor rather than augmenting it. For instance, the U.S. tax code cur-rently encourages capital investment over investment in labor through effectivetax rates that are much higher on labor than on plants and equipment.43Consider a third thought experiment: Two potential ventures each use AI tocreate $1 billion of profits. If one of them achieves this by augmenting and em-ploying a thousand workers, the firm will owe corporate and payroll taxes, whilethe employees will pay income taxes, payroll taxes, and other taxes. If the secondbusiness has no employees, the government may collect the same corporate taxes,but no payroll taxes and no taxes paid by workers. As a result, the second businessmodel pays far less in total taxes.This disparity is amplified because the tax code treats labor income moreharshly than capital income. In 1986, top tax rates on capital income and laborincome were equalized in the United States, but since then, successive changeshave created a large disparity, with the 2021 top marginal federal tax rates on labor151 (2) Spring 2022281Erik Brynjolfssonincome of 37 percent, while long capital gains have a variety of favorable rules, in-cluding a lower statutory tax rate of 20 percent, the deferral of taxes until capitalgains are realized, and the “step-up basis” rule that resets capital gains to zero,wiping out the associated taxes, when assets are inherited.The first rule of tax policy is simple: you tend to get less of whatever you tax.Thus, a tax code that treats income that uses labor less favorably than income de-rived from capital will favor automation over augmentation. Treating both busi-ness models equally would lead to more balanced incentives. In fact, given thepositive externalities of more widely shared prosperity, a case could be made fortreating wage income more favorably than capital income, for instance by expand-ing the earned income tax credit.44 It is unlikely that any government official candefine in advance exactly which technologies and innovations augment humansrather than merely substitute for them; indeed, most technologies have elementsof each and the outcome depends a great deal on how they are deployed. Thus,rather than prescribe or proscribe specific technologies, a broad-based set of in-centives can gently nudge technologists and managers toward augmentation onthe margin, much as carbon taxes encourage myriad types of cleaner energy orresearch and development tax credits encourage greater investments in research.Government policy in other areas could also do more to steer the economy clearof the Turing Trap. The growing use of AI, even if only for complementing work-ers, and the further reinvention of organizations around this new general-purposetechnology imply a great need for worker training or retraining. In fact, for eachdollar spent on machine learning technology, companies may need to spend ninedollars on intangible human capital.45 However, education and training sufferfrom a serious externality issue: companies that incur the costs to train or retrainworkers may reap only a fraction of the benefits of those investments, with therest potentially going to other companies, including competitors, as these work-ers are free to bring their skills to their new employers. At the same time, work-ers are often cash- and credit-constrained, limiting their ability to invest in theirown skills development. 46 This implies that government policy should directlyprovide education and training or provide incentives for corporate training thatoffset the externalities created by labor mobility. 47In sum, the risks of the Turing Trap are increased not by just one group in oursociety, but by the misaligned incentives of technologists, businesspeople, andpolicy-makers.T he future is not preordained. We control the extent to which AI either ex-pands human opportunity through augmentation or replaces humansthrough automation. We can work on challenges that are easy for ma-chines and hard for humans, rather than hard for machines and easy for humans.The first option offers the opportunity of growing and sharing the economic pie282Dædalus, the Journal of the American Academy of Arts & SciencesThe Turing Trap: The Promise & Peril of Human-Like Artificial Intelligenceby augmenting the workforce with tools and platforms. The second option risksdividing the economic pie among an ever-smaller number of people by creatingautomation that displaces ever-more types of workers.While both approaches can and do contribute to productivity and progress,technologists, businesspeople, and policy-makers have each been putting a fingeron the scales in favor of replacement. Moreover, the tendency of a greater concen-tration of technological and economic power to beget a greater concentration ofpolitical power risks trapping a powerless majority into an unhappy equilibrium:the Turing Trap.The backlash against free trade offers a cautionary tale. Economists have longargued that free trade and globalization tend to grow the economic pie through thepower of comparative advantage and specialization. They have also acknowledgedthat market forces alone do not ensure that every person in every country willcome out ahead. So they proposed a grand bargain: maximize free trade to max-imize wealth creation and then distribute the benefits broadly to compensate anyinjured occupations, industries, and regions. It has not worked as they had hoped.As the economic winners gained power, they reneged on the second part of the bar-gain, leaving many workers worse off than before.48 The result helped fuel a popu-list backlash that led to import tariffs and other barriers to free trade. Economistswept.Some of the same dynamics are already underway with AI. More and moreAmericans, and indeed workers around the world, believe that while the technolo-gy may be creating a new billionaire class, it is not working for them. The more tech-nology is used to replace rather than augment labor, the worse the disparity may be-come, and the greater the resentments that feed destructive political instincts andactions. More fundamentally, the moral imperative of treating people as ends, andnot merely as means, calls for everyone to share in the gains of automation.The solution is not to slow down technology, but rather to eliminate or reversethe excess incentives for automation over augmentation. A good start would be toreplace the Turing Test, and the mindset it embodies, with a new set of practicalbenchmarks that steer progress toward AI-powered systems that exceed anythingthat could be done by humans alone. In concert, we must build political and eco-nomic institutions that are robust in the face of the growing power of AI. We canreverse the growing tech backlash by creating the kind of prosperous society thatinspires discovery, boosts living standards, and offers political inclusion for ev-eryone. By redirecting our efforts, we can avoid the Turing Trap and create pros-perity for the many, not just the few.151 (2) Spring 2022283Erik Brynjolfssonauthor’s noteThe core ideas in this essay were inspired by a series of conversations with JamesManyika and Andrew McAfee. I am grateful for valuable comments and sugges-tions on this work from Matt Beane, Seth Benzell, Avi Goldfarb, Katya Klinova, Ale-na Kykalova, Gary Marcus, Andrea Meyer, Dana Meyer, and numerous participantsat seminars at the Stanford Digital Economy Lab and the University of TorontoCreative Destruction Lab, but they should not be held responsible for any errors oropinions in the essay.about the authorErik Brynjolfsson is the Jerry Yang and Akiko Yamazaki Professor and SeniorFellow at the Institute for Human-Centered AI and Director of the Digital Econ-omy Lab at Stanford University. He is also the Ralph Landau Senior Fellow at theInstitute for Economic Policy Research and Professor by Courtesy at the Gradu-ate School of Business and Department of Economics at Stanford University; and aResearch Associate at the National Bureau of Economic Research. He is the authoror coauthor of seven books, including Machine, Platform, Crowd: Harnessing Our Digi-tal Future (2017), The Second Machine Age: Work, Progress, and Prosperity in a Time of Bril-liant Technologies (2014), and Race against the Machine: How the Digital Revolution Is Acceler-ating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Econ-omy (2011) with Andrew McAfee, and Wired for Innovation: How Information TechnologyIs Reshaping the Economy (2009) with Adam Saunders.endnotes1 Alan Turing, “Computing Machinery and Intelligence,” Mind 59 (236): 433–460, https://doi.org/10.1093/mind/LIX.236.433. An earlier articulation of this test comes from Des-cartes in The Discourse, in which he wrote,If there were machines which bore a resemblance to our bodies and imitated ouractions as closely as possible for all practical purposes, we should still have twovery certain means of recognizing that they were not real men. The first is thatthey could never use words, or put together signs, as we do in order to declare ourthoughts to others. . . . Secondly, even though some machines might do some thingsas well as we do them, or perhaps even better, they would inevitably fail in others,which would reveal that they are acting not from understanding.2 Carolyn Price, “Plato, Opinions and the Statues of Daedalus,” OpenLearn, updatedJune 19, 2019, https://www.open.edu/openlearn/history-the-arts/philosophy/plato-opinions-and-the-statues-daedalus; and Andrew Stewart, “The Archaic Period,” PerseusDigital Library, http://www.perseus.tufts.edu/hopper/text?doc=Perseus:text:1999.04.0008:part=2:chapter=1&highlight=daedalus.3 “The Origin of the Word ‘Robot,’” Science Friday, April 22, 2011, https://www.sciencefriday.com/segments/the-origin-of-the-word-robot/.4 Millions of people are now working alongside robots. For a recent survey on the diffusionof robots, AI, and other advanced technologies in the United States, see Nikolas Zolas,284Dædalus, the Journal of the American Academy of Arts & SciencesThe Turing Trap: The Promise & Peril of Human-Like Artificial IntelligenceZachary Kroff, Erik Brynjolfsson, et al., “Advanced Technologies Adoption and Useby U.S. Firms: Evidence from the Annual Business Survey,” NBER Working Paper No.28290 (Cambridge, Mass.: National Bureau of Economic Research, 2020).5 Apologies to Arthur C. Clarke.6 See, for example, Daniel Zhang, Saurabh Mishra, Erik Brynjolfsson, et al., “The AI Index2021 Annual Report,” arXiv (2021), esp. chap. 2, https://arxiv.org/abs/2103.06312. Inregard to image recognition, see, for instance, the success of image recognition systemsin Olga Russakovsky, Jia Deng, Hao Su, et al., “Imagenet Large Scale Visual Recogni-tion Challenge,” International Journal of Computer Vision 115 (3) (2015): 211–252. A broadarray of business application is discussed in Erik Brynjolfsson and Andrew McAfee,“The Business of Artificial Intelligence,” Harvard Business Review (2017): 3–11.7 See, for example, Hubert Dreyfus, What Computers Can’t Do (Cambridge, Mass.: MIT Press,1972); Nils J. Nilsson, “Human-Level Artificial Intelligence? Be Serious!” AI Magazine26 (4) (2005): 68; and Gary Marcus, Francesca Rossi, and Manuela Veloso, “Beyondthe Turing Test,” AI Magazine 37 (1) (2016): 3–4.8 Nilsson, “Human-Level Artificial Intelligence?” 68.9 John Searle was the first to use the terms strong AI and weak AI, writing that with weak AI,“the principal value of the computer . . . is that it gives us a very powerful tool,” whilestrong AI “really is a mind.” Ed Feigenbaum has argued that creating such intelligenceis the “manifest destiny” of computer science. John R. Searle, “Minds, Brains, and Pro-grams,” Behavioral and Brain Sciences 3 (3) (1980): 417–457.10 However, this does not necessarily mean living standards would rise without bound.In fact, if working hours fall faster than productivity rises, it is theoretically possible,though empirically unlikely, that output and consumption (other than leisure time)would fall.11 See, for example, Robert M. Solow, “A Contribution to the Theory of Economic Growth,”The Quarterly Journal of Economics 70 (1) (1956): 65–94.12 See, for example, Daron Acemoglu, “Directed Technical Change,” Review of EconomicStudies 69 (4) (2002): 781–809.13 See, for instance, Erik Brynjolfsson and Andrew McAfee, Race Against the Machine: Howthe Digital Revolution Is Accelerating Innovation, Driving Productivity, and Irreversibly TransformingEmployment and the Economy (Lexington, Mass.: Digital Frontier Press, 2011); and DaronAcemoglu and Pascual Restrepo, “The Race Between Machine and Man: Implicationsof Technology for Growth, Factor Shares, and Employment,” American Economic Review108 (6) (2018): 1488–1542.14 For instance, the real wage of a building laborer in Great Britain is estimated to havegrown from sixteen times the amount needed for subsistence in 1820 to 167 times thatlevel by the year 2000, according to Jan Luiten Van Zanden, Joerg Baten, Marco Mirad’Ercole, et al., eds., How Was Life? Global Well-Being since 1820 (Paris: OECD Publishing,2014).15 For instance, a majority of aircraft on U.S. Navy aircraft carriers are likely to be un-manned. See Oriana Pawlyk, “Future Navy Carriers Could Have More Drones ThanManned Aircraft, Admiral Says,” Military.com, March 30, 2021. Similarly, companieslike Kittyhawk have developed pilotless aircraft (“flying cars”) for civilian passengers.151 (2) Spring 2022285Erik Brynjolfsson16 Loukas Karabarbounis and Brent Neiman, “The Global Decline of the Labor Share,” TheQuarterly Journal of Economics 129 (1) (2014): 61–103; and David Autor, “Work of the Past,Work of the Future,” NBER Working Paper No. 25588 (Cambridge, Mass.: National Bu-reau of Economic Research, 2019). For a broader survey, see Morgan R. Frank, DavidAutor, James E. Bessen, et al., “Toward Understanding the Impact of Artificial Intelli-gence on Labor,” Proceedings of the National Academy of Sciences 116 (14) (2019): 6531–6539.17 Daron Acemoglu and David Autor, “Skills, Tasks and Technologies: Implications forEmployment and Earnings,” Handbook of Labor Economics 4 (2011): 1043–1171.18 Seth G. Benzell and Erik Brynjolfsson, “Digital Abundance and Scarce Architects:Implications for Wages, Interest Rates, and Growth,” NBER Working Paper No. 25585(Cambridge, Mass.: National Bureau of Economic Research, 2021).19 Prasanna Tambe, Lorin Hitt, Daniel Rock, and Erik Brynjolfsson, “Digital Capital andSuperstar Firms,” Hutchins Center Working Paper #73 (Washington, D.C.: HutchinsCenter at Brookings, 2021), https://www.brookings.edu/research/digital-capital-and-superstar-firms.20 There is some evidence that capital is already becoming an increasingly good substitutefor labor. See, for instance, the discussion in Michael Knoblach and Fabian Stöckl,“What Determines the Elasticity of Substitution between Capital and Labor? A Litera-ture Review,” Journal of Economic Surveys 34 (4) (2020): 852.21 See, for example, Tyler Cowen, Average Is Over: Powering America beyond the Age of the GreatStagnation (New York: Penguin, 2013). Or more provocatively, Yuval Noah Harari,“The Rise of the Useless Class,” Ted Talk, February 24, 2017, https://ideas.ted.com/the-rise-of-the-useless-class/.22 Anton Korinek and Joseph E. Stiglitz, “Artificial Intelligence and Its Implications for In-come Distribution and Unemployment,” in The Economics of Artificial Intelligence, ed. AjayAgrawal, Joshua Gans, and Avi Goldfarb (Chicago: University of Chicago Press, 2019),349–390.23 Erik Brynjolfsson and Andrew McAfee, “Artificial Intelligence, for Real,” Harvard BusinessReview, August 7, 2017.24 Robert D. Putnam, Our Kids: The American Dream in Crisis (New York: Simon and Schuster,2016) describes the negative effects of joblessness, while Anne Case and Angus Deaton,Deaths of Despair and the Future of Capitalism (Princeton, N.J.: Princeton University Press,2021) documents the sharp decline in life expectancy among many of the same people.25 Simon Smith Kuznets, Economic Growth and Structure: Selected Essays (New York: W. W.Norton & Co., 1965).26 Friedrich August Hayek, “The Use of Knowledge in Society,” The American Economic Review35 (4) (1945): 519–530.27 Erik Brynjolfsson, “Information Assets, Technology and Organization,” ManagementScience 40 (12) (1994): 1645–1662, https://doi.org/10.1287/mnsc.40.12.1645.28 For instance, in the year 2000, an estimated 85 billion (mostly analog) photos were tak-en, but by 2020, that had grown nearly twenty-fold to 1.4 trillion (almost all digital)photos.286Dædalus, the Journal of the American Academy of Arts & SciencesThe Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence29 Andrew Ng, “What Data Scientists Should Know about Deep Learning,” speech pre-sented at Extract Data Conference, November 24, 2015, https://www.slideshare.net/ExtractConf/andrew-ng-chief-scientist-at-baidu (accessed September 9, 2021).30 Sanford J. Grossman and Oliver D. Hart, “The Costs and Benefits of Ownership: A The-ory of Vertical and Lateral Integration,” Journal of Political Economy 94 (4) (1986): 691–719; and Oliver D. Hart and John Moore, “Property Rights and the Nature of the Firm,”Journal of Political Economy 98 (6) (1990): 1119–1158.31 Erik Brynjolfsson and Andrew Ng, “Big AI Can Centralize Decisionmaking and Power.And That’s a Problem,” MILA-UNESCO Working Paper (Montreal: MILA-UNESCO,2021).32 “Simon Electronic Brain–Complete History of the Simon Computer,” History Com-puter, January 4, 2021, https://history-computer.com/simon-electronic-brain-complete-history-of-the-simon-computer/.33 Hans Moravec, Mind Children: The Future of Robot and Human Intelligence (Cambridge,Mass.: Harvard University Press, 1988).34 Will Knight, “Alpha Zero’s ‘Alien’ Chess Shows the Power, and the Peculiarity, of AI,”Technology Review, December 2017.35 Richard Waters, “Techmate: How AI Rewrote the Rules of Chess,” Financial Times, Janu-ary 12, 2018.36 Matt Beane and Erik Brynjolfsson, “Working with Robots in a Post-Pandemic World,”MIT Sloan Management Review 62 (1) (2020): 1–5.37 Timothy Bresnahan and Robert J. Gordon, “Introduction,” The Economics of New Goods(Chicago: University of Chicago Press, 1996).38 David Autor, Anna Salomons, and Bryan Seegmiller, “New Frontiers: The Origins andContent of New Work, 1940–2018,” NBER Preprint, July 26, 2021.39 David Killock, “AI Outperforms Radiologists in Mammographic Screening,” NatureReviews Clinical Oncology 17 (134) (2020), https://doi.org/10.1038/s41571-020-0329-7.40 Erik Brynjolfsson, Tom Mitchell, and Daniel Rock, “What Can Machines Learn, andWhat Does It Mean for Occupations and the Economy?” AEA Papers and Proceedings(2018): 43–47.41 Erik Brynjolfsson, Daniel Rock, and Prasanna Tambe, “How Will Machine LearningTransform the Labor Market?” Governance in an Emerging New World (619) (2019), https://www.hoover.org/research/how-will-machine-learning-transform-labor-market.42 Paul Milgrom and John Roberts, “The Economics of Modern Manufacturing: Technol-ogy, Strategy, and Organization,” American Economic Review 80 (3) (1990): 511–528.43 See Daron Acemoglu, Andrea Manera, and Pascual Restrepo, “Does the U.S. Tax CodeFavor Automation?” Brookings Papers on Economic Activity (Spring 2020); and Daron Ace-moglu, ed., Redesigning AI (Cambridge, Mass.: MIT Press, 2021).44 This reverses the classic result suggesting that taxes on capital should be lower than taxeson labor. Christophe Chamley, “Optimal Taxation of Capital Income in General Equi-librium with Infinite Lives,” Econometrica 54 (3) (1986): 607–622; and Kenneth L. Judd,“Redistributive Taxation in a Simple Perfect Foresight Model,” Journal of Public Econom-ics 28 (1) (1985): 59–83.151 (2) Spring 2022287Erik Brynjolfsson45 Tambe et al., “Digital Capital and Superstar Firms.”46 Katherine S. Newman, Chutes and Ladders: Navigating the Low-Wage Labor Market (Cam-bridge, Mass.: Harvard University Press, 2006).47 While the distinction between complements and substitutes is clear in economic theory,it can be trickier in practice. Part of the appeal of broad training and/or tax incentives,rather than specific technology mandates or prohibitions, is that they allow technol-ogies, entrepreneurs, and, ultimately, the market to reward approaches that augmentlabor rather than replace it.48 See David H. Autor, David Dorn, and Gordon H. Hanson, “The China Shock: Learningfrom Labor-Market Adjustment to Large Changes in Trade,” Annual Review of Economics8 (2016): 205–240.