Volume 106

Colonial Virginia: Incubator of Judicial Review

What is the historical origin of judicial review in the United States? Although scholars have acknowledged that British imperial “disallowance” of colonial law was an influential antecedent, the extant historical scholarship devoted to the mechanics of disallowance is sparse. This limited exploration is surprising. Not unlike modern judicial review, the guiding question imperial overseers considered when disallowing colonial legislation was whether it was ‘repugnant’ to the laws of England. In response, this Note’s first contribution is to explain the process by which the so-called repugnancy principle was enforced against inferior colonial law. Even fewer scholars have attempted to connect the ultimate repugnancy assessment to the historical context surrounding disallowed colonial laws. This Note’s second contribution is thus to augment existing literature by exploring colonial Virginia’s specific experience under imperial supervision.

Among the scholars that have explored the connection between colonial disallowance and the origins of judicial review, some have documented the link between imperial legislative review of colonial legislation and James Madison’s proposed constitutional solution to the problem of unrestrained state legislatures in the aftermath of independence. What remains to be explored, however, is how Madison explicitly drew on the history of imperial review of colonial Virginia’s laws as he argued at the Constitutional Convention for a federal power to “negative” state laws. Accordingly, this Note’s third contribution is to reveal that the historical practice of imperial review in Madison’s native Virginia animated his proposed solution to check the unrestrained popular will of state legislators. Although his proposed solution was ultimately rejected at the Convention, that rejection was conditioned on the judiciary possessing the power of judicial review. By exposing this hidden link, this Note demonstrates that colonial Virginia rightly may be regarded as the intellectual incubator of judicial review.

Introduction

During the British imperial era, the supreme laws of England trumped conflicting inferior colonial law. Colonial assemblies—by the terms of their colonial charters—were prohibited from enacting legislation repugnant to the laws of England. The British monarch, to both monitor the colonial assemblies and to ensure compliance with the superior laws of England, empowered the Board of Trade (“Board”) and the Privy Council with the duty to enforce the so-called repugnancy principle. That principle required the Privy Council and the Board to compare colonial legislation to English law. If the colonial legislation was, upon that comparison, deemed repugnant to the laws of England, then the law was disallowed.1 1.Disallowance was the term used to proclaim that colonial law was legally inoperative as it diverged from the laws of England. See Dudley Odell McGovney, The British Privy Council’s Power to Restrain the Legislatures of Colonial America: Power to Disallow Statutes: Power to Veto, 94 U. Pa. L. Rev. 59, 81 (1946); see also Robin L. Einhorn, American Taxation, American Slavery 15 (2006) (equating disallowance to a “veto”); Robert J. Steinfeld, The Rejection of Horizontal Judicial Review During America’s Colonial Period, 2 Critical Analysis L. 214, 218 n.19 (2015) (“[D]isallowance operated as a ‘repeal’ of the statute.”).Show More The historical record suggests that the imperial power of legislative review was not one the Privy Council and the Board were hesitant to exercise. Indeed, from 1696, when the Board of Trade was established, to 1776, when the United States declared its independence, scholars have estimated that more than 8,500 colonial laws were reviewed,2 2.Mary S. Bilder, The Corporate Origins of Judicial Review, 116 Yale L.J. 502, 538 (2006).Show More and over 400 colonial laws were disallowed for being repugnant to the laws of England.3 3.Jonathan R.T. Hughes, Social Control in the Colonial Economy 13 n.12 (1976); see also Leon T. David, Councillors and the Law Officers in the Colonies in America, 12 Am. U. L. Rev. 23, 32 (1963) (“Of some 8,563 acts submitted for approval, it disallowed 469.”); Sharon Hamby O’Connor & Mary Sarah Bilder, Appeals to the Privy Council Before American Independence:An Annotated Digital Catalogue, 104 Law Libr. J. 83, 85 (2012) (“The Council could disallow a law; approximately 8563 were sent for review and 469 (5.5%) disallowed.”).Show More This historical system of oversight and disallowance echoes a similar, more modern institution: American judicial review. The similarity between British imperial oversight and modern judicial review has not gone unnoticed. In the words of one historian, the Privy Council and the Board subjected colonial “provincial laws to a kind of constitutional test.”4 4.Oliver Morton Dickerson, American Colonial Government 1696–1765, at 234 (1912).Show More

Within the last decade, Mary Bilder and Alison LaCroix have explored the connection between the disallowance of colonial legislation and the origin of judicial review.5 5.See Mary Sarah Bilder, The Transatlantic Constitution: Colonial Legal Culture and the Empire (2004); Alison L. LaCroix, The Authority for Federalism: Madison’s Negative and the Origins of Federal Ideology, 28 Law & Hist. Rev. 451, 466–69 (2010). But see Philip Hamburger, A Tale of Two Paradigms: Judicial Review and Judicial Duty, 78 Geo. Wash. L. Rev. 1162, 1174–75 n.38 (2010). His research shows that “judges had for centuries done their duty by holding government acts unlawful and void. They had done this as to sovereign acts of the king and even as to legislation, other than acts of Parliament. As a result, early American judges did not need to establish precedents for a power of judicial review.” Id. Although Professor Hamburger offers a compelling alternative account, he overlooks the fact that even though crown officials “consistently recognized the assemblies’ authority to pass laws, they always insisted that those bodies were subordinate institutions.” Jack P. Greene, Law and Origins of the American Revolution in The Cambridge History of Law in America 447, 449 (Michael Grossberg & Christopher Tomlins eds., 2008). The insubordination of colonial assemblies beneath the British imperial apparatus thus also provides a historical antecedent from which Americans, like James Madison, could derive intellectual inspiration for American judicial review.Show More The argument is that “recurrent administrative testing of colonial statutes against a ‘constitutional’ standard exemplified in the laws of England helped pave the way for acceptance of the doctrine of judicial review in the new nation.”6 6.Joseph H. Smith, Administrative Control of the Courts of the American Plantations, 61 Colum. L. Rev. 1210, 1253 (1961).Show More Yet the extant historical scholarship devoted to this striking similarity hardly touches upon the mechanics of imperial disallowance.7 7.Astonishingly, Oliver Morton Dickerson’s American Colonial Government, which was published in 1912, remains the authoritative source on the mechanics of imperial disallowance.Show More In this respect, this Note’s first contribution is to explain the mechanics by which the repugnancy principle was enforced against inferior colonial law.

By a similar token, even fewer scholars have attempted to connect colonial legislation and the law’s surrounding historical context to the Board and the Privy Council’s ultimate repugnancy assessment.8 8.See Mary Sarah Bilder, Colonial Constitutionalism and Constitutional Law, in Transformations in American Legal History: Essays in Honor of Professor Morton J. Horwitz 28, 43 (Daniel W. Hamilton & Alfred L. Brophy eds., 2009).Show More The reason for the dearth of scholarly literature linking together these narratives is that there exists “no comprehensive list of disallowed acts.”9 9.Id.Show More This lacuna in source material also explains why “comparably little study has been given to the topic” of imperial review of colonial law in general.10 10.Id.Show More In response, this Note’s second contribution is to augment the existing literature by exploring the colonial experience under imperial supervision, specifically in the Colony of Virginia.

Colonial Virginia, after all, “had the largest population of any colony in North America,” possessed an influential economic and legal system, and “produced great leaders,” many of whom would go onto shape the Constitution’s structural framework.11 11.William E. Nelson, The Law of Colonial Maryland: Virginia Without Its Grandeur, 54 Am. J. Legal Hist. 168, 198–99 (2014).Show More Virginia was, on balance, “the jewel in the crown” of Britain’s overseas empire.12 12.Mary Carroll Johansen, The Relationship Between the Board of Trade and Plantations and the Colonial Government of Virginia, 1696–1775, at 38 (1992) (unpublished M.A. thesis, The College of William & Mary) (on file with The College of William & Mary Libraries).Show More This fact alone makes the absence of a thorough analysis of colonial Virginia’s interaction with the Privy Council remarkable. And this historical gap is only compounded by the fact that the “father of the Constitution,” James Madison, was himself a son of colonial Virginia.13 13.Michael P. Zuckert, Judicial Review and the Incomplete Constitution: A Madisonian Perspective on the Supreme Court and the Idea of Constitutionalism, in The Supreme Court and the Idea of Constitutionalism 53, 55 (Steven Kautz et al. eds., 2009).Show More In modern times, Madison is rightly memorialized for his profound influence on the Federal Constitution’s structure and for “laying the foundations of the Republic.”14 14.Charles Evans Hughes, James Madison, 18 A.B.A. J. 854, 854 (1932) (referring to Madison as the “Father of the Constitution”); see also Daniel J. Hulsebosch, Being Seen Like a State: How Americans (and Britons) Built the Constitutional Infrastructure of a Developing Nation, 59 Wm. & Mary L. Rev. 1239, 1269 (2018) (“His theory of factional checks and balances is why many consider him the most thoughtful constitution maker.”).Show More He understood the “overall logic of the new order better than anyone else at the time.”15 15.Zuckert, supra note 13, at 55.Show More His understanding of the new order was, as it turns out, deeply shaped by his experience with the old. According to Alison LaCroix, the “centerpiece of Madison’s plan to reconstitute the Republic . . . sprang directly from the institutions and practices of the British Empire, the thralldom of which the American colonies had escaped.”16 16.LaCroix, supra note 5, at 464.Show More Likewise, Michael Zuckert contends that Madison had both “an unparalleled understanding of the political nature of the Constitution,” and possessed “an unexcelled understanding of what judicial review was to be in the new system.”17 17.Zuckert, supra note 13, at 55.Show More Yet underappreciated, until now, is the influence that Privy Council disallowance of his own commonwealth’s legislation had on Madison’s frame of mind and his approach to subordinating the will of state and national electorates to the supreme law of the land.

Herein lies this Note’s third contribution. In short, I seek to enrich the existing scholarship on the origins of judicial review by offering a targeted analysis of the experience in colonial Virginia. Many scholars have argued that the concept of judicial review originated from Madison’s proposals at the Constitutional Convention.18 18.See Steven G. Calabresi, Originalism and James Bradley Thayer, 113 Nw. U. L. Rev. 1419, 1450–51 (2019) (building on James Bradley Thayer’s discussion of Madison’s proposed continuation of the imperial practice of legislative review); see also Sean Gailmard, Imperial Politics, English Law, and the Strategic Foundations of Constitutional Review in America, 113 Am. Pol. Sci. Rev. 778, 788 (2019) (“My argument is that delegates to the Constitutional Convention of 1787 recognized and sought to preserve benefits of Crown review by the Privy Council as an external bound on legislation.”).Show More The general story tracing the link between the Privy Council, the Constitutional Convention, and the federal courts’ ability to disallow repugnant legislation has been told.19 19.Section 25 of the Judiciary Act of 1789 granted federal courts jurisdiction over state courts in matters where “the validity of a statute” is drawn into question “on the ground of their being repugnant to the constitution.” Judiciary Act of Sept. 24, 1789, ch. 20, § 25, 1 Stat. 73, 85. In essence, the federal courts were empowered, much like the Board and the Privy Council, with the duty to enforce the repugnancy principle against state and federal legislation that conflicted, not with the laws of England, but with the text of the Constitution.Show More Against the backdrop of these abstract accounts, this Note restricts the study of Privy Council oversight specifically to colonial Virginia. This narrow focus better facilitates an understanding of how Madison, through his knowledge of actual practice, envisioned the will of subordinate legislatures conforming to the supremacy of the new Federal Constitution.20 20.Indeed, Professor Jordan Cash has observed that “judicial review had long been practiced in Virginia, and the English jurisdictional tradition continued to be influential into the early national period.” Jordan T. Cash, The Court and the Old Dominion: Judicial Review Among the Virginia Jeffersonians, 35 Law & Hist. Rev. 351, 365 (2017). Although less general than most accounts, Professor Cash’s assertion still paints with too broad a brush, as it does not explore British imperial oversight’s influence upon Madison’s proposed constitutional solutions.Show More As this Note uncovers, Madison himself thought deeply about imperial review of colonial legislation—particularly that of colonial Virginia—in the years leading up to the Constitutional Convention. And it was from Madison’s Privy Council-influenced proposals that judicial review ultimately sprung. This Note, therefore, confines itself to the study of Privy Council oversight of colonial Virginia and explores the story of three Virginian colonial acts, and their interaction with the British imperial system, to cast useful light on Madison’s vision of judicial review and constitutional theory more generally.

This Note is divided into three Parts. Part I discusses the history of the Board of Trade and the Privy Council’s enforcement of the repugnancy principle. Surprisingly, that enforcement process, and the innerworkings of both the Privy Council and the Board, has received remarkably little scholarly attention. Part II details the three Virginian Acts in chronological order. Discussing each Act’s historical context and ultimate demise brings to the surface some of the major issues that plagued colonial society. It also calls attention to the process and general cultural perception of legislative review in colonial Virginia. Part III turns to the influence imperial oversight of Virginia’s colonial legislation had on Madison—an influence that inspired Madison’s proposed federal constitutional framework. In short, the influence that both the Privy Council and Board’s scrutiny of Virginia’s colonial legislation had on Madison’s attempt to restrain the democratic will of state and national electorates may help us more clearly understand the imperial, colonial origin of judicial review.

  1. * University of Virginia School of Law, J.D. 2020. I am grateful first and foremost for Professor Cynthia Nicoletti and her insightful input, unwavering patience, and immense generosity. I would like to thank both Christian Talley and Anna Cecile Pepper for helpful comments and also the members of the Virginia Law Review, especially Clay Phillips, for careful editing and feedback. I am solely responsible for all errors. ↑
  2. Disallowance was the term used to proclaim that colonial law was legally inoperative as it diverged from the laws of England. See Dudley Odell McGovney, The British Privy Council’s Power to Restrain the Legislatures of Colonial America: Power to Disallow Statutes: Power to Veto, 94 U. Pa. L. Rev
    .

    59, 81 (1946); see also Robin L. Einhorn, American Taxation, American Slavery 15 (2006) (equating disallowance to a “veto”); Robert J. Steinfeld, The Rejection of Horizontal Judicial Review During America’s Colonial Period, 2 Critical Analysis L. 214, 218 n.19 (2015) (“[D]isallowance operated as a ‘repeal’ of the statute.”). ↑

  3. Mary S. Bilder, The Corporate Origins of Judicial Review, 116 Yale L.J
    .

    502, 538 (2006). ↑

  4. Jonathan R.T. Hughes, Social Control in the Colonial Economy 13 n.12 (1976); see also Leon T. David, Councillors and the Law Officers in the Colonies in America, 12 Am. U. L. Rev
    .

    23, 32 (1963) (“Of some 8,563 acts submitted for approval, it disallowed 469.”); Sharon Hamby O’Connor & Mary Sarah Bilder, Appeals to the Privy Council Before American Independence: An Annotated Digital Catalogue, 104 Law Libr. J

    .

    83, 85 (2012) (“The Council could disallow a law; approximately 8563 were sent for review and 469 (5.5%) disallowed.”). ↑

  5. Oliver Morton Dickerson, American Colonial Government 1696–1765, at 234 (1912). ↑
  6. See Mary Sarah Bilder, The Transatlantic Constitution: Colonial Legal Culture and the Empire (2004); Alison L. LaCroix, The Authority for Federalism: Madison’s Negative and the Origins of Federal Ideology, 28 Law & Hist. Rev. 451, 466–69 (2010). But see Philip Hamburger, A Tale of Two Paradigms: Judicial Review and Judicial Duty, 78 Geo. Wash. L. Rev. 1162, 1174–75 n.38 (2010). His research shows that “judges had for centuries done their duty by holding government acts unlawful and void. They had done this as to sovereign acts of the king and even as to legislation, other than acts of Parliament. As a result, early American judges did not need to establish precedents for a power of judicial review.” Id. Although Professor Hamburger offers a compelling alternative account, he overlooks the fact that even though crown officials “consistently recognized the assemblies’ authority to pass laws, they always insisted that those bodies were subordinate institutions.” Jack P. Greene, Law and Origins of the American Revolution in The Cambridge History of Law in America 447, 449 (Michael Grossberg & Christopher Tomlins eds., 2008). The insubordination of colonial assemblies beneath the British imperial apparatus thus also provides a historical antecedent from which Americans, like James Madison, could derive intellectual inspiration for American judicial review. ↑
  7. Joseph H. Smith, Administrative Control of the Courts of the American Plantations, 61 Colum. L. Rev. 1210, 1253 (1961). ↑
  8. Astonishingly, Oliver Morton Dickerson’s American Colonial Government, which was published in 1912, remains the authoritative source on the mechanics of imperial disallowance. ↑
  9. See Mary Sarah Bilder, Colonial Constitutionalism and Constitutional Law, in Transformations in American Legal History: Essays in Honor of Professor Morton J. Horwitz 28, 43 (Daniel W. Hamilton & Alfred L. Brophy eds., 2009). ↑
  10. Id. ↑
  11. Id. ↑
  12. William E. Nelson, The Law of Colonial Maryland: Virginia Without Its Grandeur, 54 Am. J. Legal Hist. 168, 198–99 (2014). ↑
  13. Mary Carroll Johansen, The Relationship Between the Board of Trade and Plantations and the Colonial Government of Virginia, 1696–1775, at 38 (1992) (unpublished M.A. thesis, The College of William & Mary) (on file with The College of William & Mary Libraries). ↑
  14. Michael P. Zuckert, Judicial Review and the Incomplete Constitution: A Madisonian Perspective on the Supreme Court and the Idea of Constitutionalism, in The Supreme Court and the Idea of Constitutionalism 53, 55 (Steven Kautz et al. eds., 2009). ↑
  15. Charles Evans Hughes, James Madison, 18 A.B.A. J. 854, 854 (1932) (referring to Madison as the “Father of the Constitution”); see also Daniel J. Hulsebosch, Being Seen Like a State: How Americans (and Britons) Built the Constitutional Infrastructure of a Developing Nation, 59 Wm. & Mary L. Rev. 1239, 1269 (2018) (“His theory of factional checks and balances is why many consider him the most thoughtful constitution maker.”). ↑
  16. Zuckert, supra note 13, at 55. ↑
  17. LaCroix, supra note 5, at 464. ↑
  18. Zuckert, supra note 13, at 55. ↑
  19. See Steven G. Calabresi, Originalism and James Bradley Thayer, 113 Nw. U. L. Rev. 1419, 1450–51 (2019) (building on James Bradley Thayer’s discussion of Madison’s proposed continuation of the imperial practice of legislative review); see also Sean Gailmard, Imperial Politics, English Law, and the Strategic Foundations of Constitutional Review in America, 113 Am. Pol. Sci. Rev. 778, 788 (2019) (“My argument is that delegates to the Constitutional Convention of 1787 recognized and sought to preserve benefits of Crown review by the Privy Council as an external bound on legislation.”). ↑
  20. Section 25 of the Judiciary Act of 1789 granted federal courts jurisdiction over state courts in matters where “the validity of a statute” is drawn into question “on the ground of their being repugnant to the constitution.” Judiciary Act of Sept. 24, 1789, ch. 20, § 25, 1 Stat. 73, 85. In essence, the federal courts were empowered, much like the Board and the Privy Council, with the duty to enforce the repugnancy principle against state and federal legislation that conflicted, not with the laws of England, but with the text of the Constitution. ↑
  21. Indeed, Professor Jordan Cash has observed that “judicial review had long been practiced in Virginia, and the English jurisdictional tradition continued to be influential into the early national period.” Jordan T. Cash, The Court and the Old Dominion: Judicial Review Among the Virginia Jeffersonians, 35 Law & Hist. Rev. 351, 365 (2017). Although less general than most accounts, Professor Cash’s assertion still paints with too broad a brush, as it does not explore British imperial oversight’s influence upon Madison’s proposed constitutional solutions. ↑

Myopic Consumer Law

People make mistakes with debt, partly because the chance to buy now and pay later tempts them to do things that are not in their long-term interest. Lenders sell credit products that exploit this vulnerability. In this Article, I argue that critiques of these products that draw insights from behavioral law and economics have a blind spot: they ignore what the borrowed funds are used for. By evaluating financing transactions in isolation from the underlying purchase, the cost-benefit analysis of consumer financial regulation is truncated and misleading. I show that the same psychological bias that allows someone to be sold an exploitative loan also makes it possible that the exploitative loan benefits them by causing them to purchase a product or service that they should, but would not otherwise, buy. I demonstrate the importance of this effect in a study of tax refund anticipation loans. I find that regulation curtailing these loans increased the use of an alternative credit product and reduced the use of paid tax preparers and the take-up of the earned income tax credit.

Introduction

Behavioral law and economics has had significant influence on the regulation of consumer credit.1 1.See, e.g., Ryan Bubb & Richard H. Pildes, How Behavioral Economics Trims Its Sails and Why, 127 Harv. L. Rev. 1593, 1644–47 (2014).Show More This is both important and justified. It is important because consumer finance is central to the functioning of a modern economy; it is what President Obama called the “lifeblood” during the height of the financial crisis in 2009.2 2.Address Before a Joint Session of the Congress, 1 Pub. Papers 145, 147 (Feb. 24, 2009).Show More At the level of individual households, consumer credit is important because the timing of income and expenses are rarely contemporaneous. And yet, credit transactions are fraught. Credit both reflects and perpetuates wide differences in individuals’ economic opportunities and their vulnerability to financial adversity. Credit is more expensive for the poor, and this fact creates a patina of exploitation and abuse over debt transactions that has resulted in extensive state and federal regulation.

The influence of behavioral economics on consumer credit regulation is justified because two features of consumer credit raise doubts about consumers’ ability to make borrowing choices that are in their best interests. The first feature is complexity. Consumer debt often has a complex fee structure, opaque repayment terms, and default consequences that are hard to evaluate.3 3.For a discussion of the importance of complexity and faulty borrower comprehension in consumer credit markets, see Lauren E. Willis, Decisionmaking and the Limits of Disclosure: The Problem of Predatory Lending: Price, 65 Md. L. Rev. 707, 766–98 (2006) [hereinafter Willis, Decisionmaking and the Limits of Disclosure]. Unfortunately, interventions to increase consumer financial literacy do not appear to help remedy these problems. Lauren E. Willis, Against Financial-Literacy Education, 94 Iowa L. Rev. 197, 201 (2008); Lauren E. Willis, The Financial Education Fallacy, 101 Am. Econ.Rev. 429, 429 (2011). Because financial education and disclosure have proven to be largely ineffective, Professor Willis has provocatively argued for an alternative known as “performance-based consumer law.” Lauren E. Willis, Performance-Based Consumer Law, 82 U. Chi. L. Rev. 1309, 1311 (2015).Show More The second feature is the tradeoff between current and future purchasing power that is at the heart of every credit transaction. It is the essence of debt that the borrower exchanges her promise to pay amounts in the future for the ability to consume more now. This intertemporal tradeoff is one that individuals often struggle to make properly, and the challenge is especially great for individuals who focus excessively on the short term and who are therefore inclined to borrow impulsively and on terms that they subsequently regret.4 4.See Ian M. McDonald, The Global Financial Crisis and Behavioural Economics, 28 Econ. Papers 249, 251 (2009).Show More Both complexity and intertemporal choice are areas where behavioral law and economics scholarship is able to traffic in deep intuitions and draw on strong empirical evidence to make recommendations about how to regulate imperfectly rational consumers.5 5.I am unaware of any data about the intuitive appeal of complexity and impatience as explanations for why people struggle to evaluate credit contracts. Nevertheless, I trust that most readers, particularly those with home mortgages, will be inclined to agree that understanding all the terms of a secured loan, even when one is trained in law or economics, demands a great deal of time and effort. It is unsurprising then that some do not even make the effort. Judge Posner famously declined to read the “boilerplate” on his own home mortgage. David Lat, Do Lawyers Actually Read Boilerplate Contracts?, Above the Law (June 22, 2010, 2:42 PM), http://abovethelaw.com/2010/06/do-lawyers-actaully-read-boilerplate-contracts-judge-richard-posner-doesnt-do-you/ [https://perma.cc/R574-VCQS]. I also expect that most of us identify with the present-biased individual, who procrastinates when it comes to unpleasant tasks and acts impulsively when it comes to food or leisure. For a review of the literature, see Lee Anne Fennell, Willpower and Legal Policy, 5 Ann. Rev. L. & Soc. Sci. 91 (2009).Show More

In this Article, I focus on arguments about consumer finance regulation that draw on research about “present bias,” which is a sort of myopia that causes people to focus on the present and neglect the future. I argue that consumer law scholarship that draws on these insights has itself been myopic. People borrow money in order to buy things, and scholarship has generally neglected to consider what borrowed funds are used for.6 6.Some researchers do think it is broadly relevant what consumers do with the loan proceeds, but none evaluate the bundled loan and purchase together from the perspective of a biased consumer. See, e.g., Shmuel I. Beecher, Yuval Feldman & Orly Lobel, Poor Consumer(s) Law: The Case of High-Cost Credit and Payday Loans, in Legal Applications of Marketing Theory (Jacob Gersen & Joel Steckel eds.) (forthcoming 2020) (manuscript at 10), http://ssrn.com/abstract = 3235810 [https://perma.cc/2ZRB-QQ44].Show More I show that focusing on the terms of a loan, isolated from the good or service that is purchased with the proceeds, leads to misleading conclusions about the benefits to the borrower. Integrating the costs and benefits of the underlying purchase with the terms of the credit transaction can upend standard conclusions about the effects of present bias and relocate efforts to improve consumer welfare from the regulation of financial products to the circumstances that create demand for high-cost credit in the first place. I demonstrate the significance of this theoretical claim by reporting results from a study of tax refund anticipation loans (RALs), which shows how RALs increase the use of paid tax preparers and the take-up of the earned income tax credit (EITC) by low-income households. Because of the size of the EITC, these loans may make present-biased taxpayers better off, even if the loans are designed to exploit their bias.

When considering the benefits of credit transactions for present-biased consumers, why do the motivating purchases matter? The answer is that many goods and services are characterized by significant upfront costs but benefits that are only realized in the future. As I show in Part I, present-biased consumers tend to undervalue products with this temporal pattern of costs and benefits.7 7.See discussion infra Section I.A.Show More Durable goods, such as homes, cars, and appliances, are like this. Purchasing durable goods involves a significant cash outlay at the time of purchase in exchange for a stream of consumption benefits that are realized over time. In fact, all sorts of choices present this same temporal pattern of immediate costs and future benefits. For example, the benefits of education are mostly realized long after the classroom experience. Applying for social welfare benefits can require an upfront investment of time and effort in exchange for benefits that are received in the future. The EITC, which is the largest federal cash transfer to low-income households,8 8.Chris Edwards & Veronique de Rugy, Earned Income Tax Credit: Small Benefits, Large Costs, Cato Inst. (Oct. 14, 2015), https://www.cato.org/publications/tax-budget-bulletin/earned-income-tax-credit-small-benefits-large-costs [https://perma.cc/5L9L-RHX9].Show More is only available to individuals who file a tax return and complete the burdensome earned income credit (EIC) schedule.9 9.On the difficulties of filing for the EITC, see Michelle Lyon Drumbl, Beyond Polemics: Poverty, Taxes, and Noncompliance, 14 eJournal Tax Res. 253, 275–77 (2016); Francine J. Lipman, The Working Poor Are Paying for Government Benefits: Fixing the Hole in the Anti-Poverty Purse, 2003 Wis. L. Rev. 461, 464; George K. Yin et al., Improving the Delivery of Benefits to the Working Poor: Proposals to Reform the Earned Income Tax Credit Program, 11 Am. J. Tax Pol’y 225, 254–56 (1994). In her latest annual report to Congress, however, the National Taxpayer Advocate noted that the IRS has been working to improve EITC outreach and education. Internal Revenue Serv., Nat’l Taxpayer Advoc.,Ann. Rep. to Congress 144 (2017).Show More The key point is that when the deferred costs and immediate benefits of certain exploitative credit products are added to the immediate costs and deferred benefits of durable goods and services, the bundled transaction may be one that is appealing to a present-biased individual and makes them better off. The exploitative loan tempts the present-biased individual to do something that is in her interest but that she would not otherwise do.10 10.I say that a loan is exploitative if only biased borrowers want to borrow on its terms. This definition does not imply anything about the profitability of these loans to the lender or about the division of the gains from trade. For a philosophical treatment of exploitation, see Alan Wertheimer, Exploitation 7–8(1996).Show More

The results from this analysis sound a note of caution about decontextualizing the choices that consumers make. At the most general level, this Article shows that if consumer law is to help imperfectly rational consumers, it is not enough to show that certain goods or services would only be purchased by consumers acting on a bias that operates against their own interests. It must also consider what other choices these consumers are likely to make that depend on that product and how the exploitative product fits into the overall way that they have arranged their lives. The personal affairs of present-biased individuals are likely to be characterized by a variety of biased decisions that may be interconnected in important ways. Although the entire constellation of choices made by present-biased individuals will leave them worse off than if they made the same choices rationally, this does not imply that compelling them to make any one of these choices rationally will leave them better off.11 11.Law and economics scholars will recognize this as an application of the general theory of the second best to intra-personal choice. R.G. Lipsey & R.K. Lancaster, TheGeneral Theory of Second Best, 24 Rev. Econ. Stud. 11, 11–12 (1956).Show More

The second contribution of this Article is to consumer finance regulation specifically. Regulating the substantive terms of consumer credit requires distinguishing between different kinds of loan products and the uses to which the loan proceeds are put. Specifically, secured debt that must be used to purchase goods and services with deferred benefits has different effects on present-biased consumers than general unsecured debt that can be used to change the timing of consumption generally.12 12.Seediscussion infra Section I.A.Show More When we integrate the loan’s terms with the pattern of costs and benefits from the purchase that necessitated the loan, we see that the bundled transaction may in fact be beneficial for present-biased consumers.13 13.Seediscussion infra Section I.A.Show More If the bundled transaction is beneficial, then prohibiting credit terms that are designed to tempt present-biased individuals might hurt those that the ban is meant to help.

Third, and at the level of most direct application, the results of my empirical study have very specific implications for the regulation of RALs and refund anticipation checks (RACs). The results sound a warning to regulators about the effects of eliminating these products. RALs disappeared almost entirely following a regulatory change in 2011,14 14.See discussion infra Section II.D.Show More a change that was celebrated by consumer advocates.15 15.Chi Chi Wu & Jean Ann Fox, Nat’l Consumer Law Ctr. & Consumer Fed’n of Am., The Party’s Over for Quickie Tax Loans: But Traps Remain for Unwary Taxpayers 2 (2012), https://www.nclc.org/images/pdf/pr-reports/report-ral-2012.pdf [https://perma.cc/J9QX-QM­XK] (“While an occasional fringe lender may make a tax-time loan, the sale of RALs as a widespread industry-wide practice is over. RALs will no longer drain the tax refunds of millions of mostly low-income taxpayers.”).Show More The near elimination of RALs reduced the use of paid tax preparers, lowered take-up of the EITC, and increased demand for RACs.16 16.See discussion infra Part II.Show More RACs are popular, and RALs have begun to make a comeback, but both credit products are the focus of opposition from advocates and concern by regulators.17 17.Tax RALs are resurgent, albeit in smaller amounts than before. For a sense of the magnitude of this resurgence, there were 35,000 refund loans made in 2014 and approximately one million loans made in 2016. Kevin Wack, Tax Refund Loans Get a Second Life, Am. Banker (June 15, 2016, 2:49 PM), https://www.americanbanker.com/news/tax-refund-loans-get-a-second-life [https://perma.cc/ZG58-WG4M].Show More Thus, understanding the role they play in affecting tax compliance and the take-up of valuable social benefits is important and timely.

To be clear, present bias is not the only reason to be suspicious of credit transactions, and the purpose of my analysis is not to provide an all-things-considered appraisal of high-cost credit products. Complexity, unrealistic optimism about repayment prospects, and other psychological biases may cause people to choose financial products that are not in their best interests.18 18.Overly optimistic borrowers may borrow too much or too little. See Richard M. Hynes, Overoptimism and Overborrowing, 2004 BYU L. Rev. 127, 131.Show More I agree with scholars who emphasize the problem of complexity and the potential role for regulation in this area.19 19.See, e.g., Saurabh Bhargava & George Loewenstein, Behavioral Economics and Public Policy 102: Beyond Nudging, 105 Am. Econ. Rev. 396, 396 (2015) (arguing that behavioral economics should leverage gaps in the traditional economic approach that assume fully rational and informed individuals to deliver policy solutions).Show More But when regulation is motivated by concerns about borrowers’ psychological biases, it must consider not just how those biases generate demand for the product being regulated but also how that product is likely to fit into the life of someone who exhibits that bias more generally.

Part I explains the present bias framework for thinking about credit transactions and describes how present bias has been used to explain demand for three economically important, high-cost credit products. I show how integrating the underlying purchase transaction into the analysis of these credit products can change our conclusions about whether these products are beneficial. In Parts II–V, I report and discuss the results of an original study of the effects of regulating RALs. The results illustrate the theoretical effects I describe in Part I, provide evidence that is relevant for regulating this financial product, and raise hard questions about the intermediating role of the private sector between individuals and the U.S. Treasury. In Part VI, I describe a framework for thinking about the regulation of consumer credit products, paying special attention to RALs.

  1. * Class of 1948 Professor of Scholarly Research in Law, University of Virginia School of Law. Thanks to Jennifer Arlen, Oren Bar-Gill, Gustavo Bobonis, Tom Brennan, Ryan Bubb, Mihir Desai, Brian Galle, Yehonatan Givati, Jacob Goldin, Daniel Hemel, Louis Kaplow, Lewis Kornhauser, Kory Kroft, Day Manoli, Ruth Mason, Patricia McCoy, Alex Raskolnikov, Kyle Rozema, Emily Satterthwaite, David Schizer, Kathryn Spier, Rory Van Loo, David Walker, George Yin, workshop participants at the American Law and Economics Association Annual Meeting, the Columbia Law School-Hebrew University Tax Conference, the University of Toronto, Boston University, Cardozo Law School, New York University, and Harvard Law School. Thanks to the Brookings Institution and AggData LLC for providing data. I am especially indebted to Kent Olson of the UVA Law Library for exceptional research assistance. ↑
  2. See, e.g., Ryan Bubb & Richard H. Pildes, How Behavioral Economics Trims Its Sails and Why, 127 Harv. L. Rev. 1593, 1644–47 (2014). ↑
  3. Address Before a Joint Session of the Congress, 1 Pub. Papers 145, 147 (Feb. 24, 2009). ↑
  4. For a discussion of the importance of complexity and faulty borrower comprehension in consumer credit markets, see Lauren E. Willis, Decisionmaking and the Limits of Disclosure: The Problem of Predatory Lending: Price, 65 Md. L. Rev. 707, 766–98 (2006) [hereinafter Willis, Decisionmaking and the Limits of Disclosure]. Unfortunately, interventions to increase consumer financial literacy do not appear to help remedy these problems. Lauren E. Willis, Against Financial-Literacy Education, 94 Iowa L. Rev. 197, 201 (2008); Lauren E. Willis, The Financial Education Fallacy, 101 Am. Econ.

    Rev. 429, 429 (2011). Because financial education and disclosure have proven to be largely ineffective, Professor Willis has provocatively argued for an alternative known as “performance-based consumer law.” Lauren E. Willis, Performance-Based Consumer Law, 82 U. Chi. L. Rev. 1309, 1311 (2015). ↑

  5. See Ian M. McDonald, The Global Financial Crisis and Behavioural Economics, 28 Econ. Papers 249, 251 (2009). ↑
  6. I am unaware of any data about the intuitive appeal of complexity and impatience as explanations for why people struggle to evaluate credit contracts. Nevertheless, I trust that most readers, particularly those with home mortgages, will be inclined to agree that understanding all the terms of a secured loan, even when one is trained in law or economics, demands a great deal of time and effort. It is unsurprising then that some do not even make the effort. Judge Posner famously declined to read the “boilerplate” on his own home mortgage. David Lat, Do Lawyers Actually Read Boilerplate Contracts?, Above the Law (June 22, 2010, 2:42 PM), http://abovethelaw.com/2010/06/do-lawyers-actaully-read-boilerplate-contracts-judge-richard-posner-doesnt-do-you/ [https://perma.cc/R574-VCQS]. I also expect that most of us identify with the present-biased individual, who procrastinates when it comes to unpleasant tasks and acts impulsively when it comes to food or leisure. For a review of the literature, see Lee Anne Fennell, Willpower and Legal Policy, 5 Ann. Rev. L. & Soc. Sci. 91 (2009). ↑
  7. Some researchers do think it is broadly relevant what consumers do with the loan proceeds, but none evaluate the bundled loan and purchase together from the perspective of a biased consumer. See, e.g., Shmuel I. Beecher, Yuval Feldman & Orly Lobel, Poor Consumer(s) Law: The Case of High-Cost Credit and Payday Loans, in Legal Applications of Marketing Theory (Jacob Gersen & Joel Steckel eds.) (forthcoming 2020) (manuscript at 10), http://ssrn.com/abstract = 3235810 [https://perma.cc/2ZRB-QQ44]. ↑
  8. See discussion infra Section I.A. ↑
  9. Chris Edwards & Veronique de Rugy, Earned Income Tax Credit: Small Benefits, Large Costs, Cato Inst. (Oct. 14, 2015), https://www.cato.org/publications/tax-budget-bulletin/earned-income-tax-credit-small-benefits-large-costs [https://perma.cc/5L9L-RHX9]. ↑
  10. On the difficulties of filing for the EITC, see Michelle Lyon Drumbl, Beyond Polemics: Poverty, Taxes, and Noncompliance, 14 eJournal Tax Res. 253, 275–77 (2016); Francine J. Lipman, The Working Poor Are Paying for Government Benefits: Fixing the Hole in the Anti-Poverty Purse, 2003 Wis. L. Rev. 461, 464; George K. Yin et al., Improving the Delivery of Benefits to the Working Poor: Proposals to Reform the Earned Income Tax Credit Program, 11 Am. J. Tax Pol’y 225, 254–56 (1994). In her latest annual report to Congress, however, the National Taxpayer Advocate noted that the IRS has been working to improve EITC outreach and education. Internal Revenue Serv., Nat’l Taxpayer Advoc.,

    Ann. Rep. to Congress 144 (2017). ↑

  11. I say that a loan is exploitative if only biased borrowers want to borrow on its terms. This definition does not imply anything about the profitability of these loans to the lender or about the division of the gains from trade. For a philosophical treatment of exploitation, see Alan Wertheimer, Exploitation 7–8

    (1996). ↑

  12. Law and economics scholars will recognize this as an application of the general theory of the second best to intra-personal choice. R.G. Lipsey & R.K. Lancaster, The General Theory of Second Best, 24 Rev. Econ. Stud. 11, 11–12 (1956). ↑
  13. See discussion infra Section I.A. ↑
  14. See discussion infra Section I.A. ↑
  15. See discussion infra Section II.D. ↑
  16. Chi Chi Wu & Jean Ann Fox, Nat’l Consumer Law Ctr. & Consumer Fed’n of Am., The Party’s Over for Quickie Tax Loans: But Traps Remain for Unwary Taxpayers 2 (2012), https://www.nclc.org/images/pdf/pr-reports/report-ral-2012.pdf [https://perma.cc/J9QX-QM­XK] (“While an occasional fringe lender may make a tax-time loan, the sale of RALs as a widespread industry-wide practice is over. RALs will no longer drain the tax refunds of millions of mostly low-income taxpayers.”). ↑
  17. See discussion infra Part II. ↑
  18. Tax RALs are resurgent, albeit in smaller amounts than before. For a sense of the magnitude of this resurgence, there were 35,000 refund loans made in 2014 and approximately one million loans made in 2016. Kevin Wack, Tax Refund Loans Get a Second Life, Am. Banker (June 15, 2016, 2:49 PM), https://www.americanbanker.com/news/tax-refund-loans-get-a-second-life [https://perma.cc/ZG58-WG4M]. ↑
  19. Overly optimistic borrowers may borrow too much or too little. See Richard M. Hynes, Overoptimism and Overborrowing, 2004 BYU L. Rev. 127, 131. ↑
  20. See, e.g., Saurabh Bhargava & George Loewenstein, Behavioral Economics and Public Policy 102: Beyond Nudging, 105 Am. Econ. Rev. 396, 396 (2015) (arguing that behavioral economics should leverage gaps in the traditional economic approach that assume fully rational and informed individuals to deliver policy solutions). ↑

A Right to a Human Decision

Recent advances in computational technologies have spurred anxiety about a shift of power from human to machine decision makers. From welfare and employment to bail and other risk assessments, state actors increasingly lean on machine-learning tools to directly allocate goods and coercion among individuals. Machine-learning tools are perceived to be eclipsing, even extinguishing, human agency in ways that compromise important individual interests. An emerging legal response to such worries is to assert a novel right to a human decision. European law embraced the idea in the General Data Protection Regulation. American law, especially in the criminal justice domain, is moving in the same direction. But no jurisdiction has defined with precision what that right entails, furnished a clear justification for its creation, or defined its appropriate domain.

This Article investigates the legal possibilities and normative appeal of a right to a human decision. I begin by sketching its conditions of technological plausibility. This requires the specification of both a feasible domain of machine decisions and the margins along which machine decisions are distinct from human ones. With this technological accounting in hand, I analyze the normative stakes of a right to a human decision. I consider four potential normative justifications: (a) a concern with population-wide accuracy; (b) a grounding in individual subjects’ interests in participation and reason giving; (c) arguments about the insufficiently reasoned or individuated quality of state action; and (d) objections grounded in negative externalities. None of these yields a general justification for a right to a human decision. Instead of being derived from normative first principles, limits to machine decision making are appropriately found in the technical constraints on predictive instruments. Within that domain, concerns about due process, privacy, and discrimination in machine decisions are typically best addressed through a justiciable “right to a well-calibrated machine decision.”

Introduction

Every tectonic technological change—from the first grain domesticated to the first smartphone set abuzz1 1.For recent treatments of these technological causes of social transformations, see generally James C. Scott, Against the Grain: A Deep History of the Earliest States (2017), and Ravi Agrawal, India Connected: How the Smartphone is Transforming the World’s Largest Democracy (2018).Show More—begets a new society. Among the ensuing birth pangs are novel anxieties about how power is distributed—how it is to be gained, and how it will be lost. A spate of sudden advances in the computational technology known as machine learning has stimulated the most recent rush of inky public anxiety. These new technologies apply complex algorithms,2 2.An algorithm is simply a “well-defined set of steps for accomplishing a certain goal.” Joshua A. Kroll et al., Accountable Algorithms, 165 U. Pa. L. Rev. 633, 640 n.14 (2017); see also Thomas H. Cormen et al., Introduction to Algorithms 5 (3d ed. 2009) (defining an algorithm as “any well-defined computational procedure that takes some value, or set of values, as input and produces some value, or set of values, as output” (emphasis omitted)). The task of computing, at its atomic level, comprises the execution of serial algorithms. Martin Erwig, Once Upon an Algorithm: How Stories Explain Computing 1–4 (2017).Show More called machine-learning instruments, to vast pools of public and government data so as to execute tasks previously beyond mere human ability.3 3.Machine learning is a general purpose technology that, in broad terms, encompasses “algorithms and systems that improve their knowledge or performance with experience.” Peter Flach, Machine Learning: The Art and Science of Algorithms that Make Sense of Data 3 (2012); see also Ethem Alpaydin, Introduction to Machine Learning 2–3 (3d ed. 2014) (defining machine learning in similar terms). For the uses of machine learning, see Susan Athey, Beyond Prediction: Using Big Data for Policy Problems, 355 Science 483, 483 (2017) (noting the use of machine learning to solve prediction problems). I discuss the technological scope of the project, and define relevant terms, infra at text accompanying note 111. I will use the terms “algorithmic tools” and “machine learning” interchangeably, even though the class of algorithms is technically much larger.Show More Corporate and state actors increasingly lean on these tools to make “decisions that affect people’s lives and livelihoods—from loan approvals, to recruiting, legal sentencing, and college admissions.”4 4.Kartik Hosanagar & Vivian Jair, We Need Transparency in Algorithms, But Too Much Can Backfire, Harv. Bus. Rev. (July 23, 2018), https://hbr.org/2018/07/we-need-transparency-in-algorithms-but-too-much-can-backfire [https://perma.cc/7KQ9-QMF3]; accord Cary Coglianese & David Lehr, Regulating by Robot: Administrative Decision Making in the Machine-Learning Era, 105 Geo. L.J. 1147, 1149 (2017).Show More

As a result, many people feel a loss of control over key life decisions.5 5.Shoshana Zuboff, Big Other: Surveillance Capitalism and the Prospects of an Information Civilization, 30 J. Info. Tech. 75, 75 (2015) (describing a “new form of information capitalism [that] aims to predict and modify human behavior as a means to produce revenue and market control”).Show More Machines, they fear, resolve questions of critical importance on grounds that are beyond individuals’ ken or control.6 6.See, e.g., Rachel Courtland, The Bias Detectives, 558 Nature 357, 357 (2018) (documenting concerns among the public that algorithmic risk scores for detecting child abuse fail to account for an “effort . . . to turn [a] life around”).Show More Many individuals experience a loss of elementary human agency and a corresponding vulnerability to an inhuman and inhumane machine logic. For some, “the very idea of an algorithmic system making an important decision on the basis of past data seem[s] unfair.”7 7.Reuben Binns et al., ‘It’s Reducing a Human Being to a Percentage’; Perceptions of Justice in Algorithmic Decisions, 2018 CHI Conf. on Hum. Factors Computing Systems 9 (emphasis omitted).Show More Machines, it is said, want fatally for “empathy.”8 8.Virginia Eubanks, Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor 168 (2017).Show More For others, machine decisions seem dangerously inscrutable, non-transparent, and so hazardously unpredictable.9 9.Will Knight, The Dark Secret at the Heart of AI, MIT Tech. Rev. (Apr. 11, 2017), https://www.technologyreview.com/s/604087/the-dark-secret-at-the-heart-of-ai/ [https://perma.cc/L94L-LYTJ] (“The computers that run those services have programmed themselves, and they have done it in ways we cannot understand. Even the engineers who build these apps cannot fully explain their behavior.”).Show More Worse, governments and companies wield these tools freely to taxonomize their populations, predict individual behavior, and even manipulate behavior and preferences in ways that give them a new advantage over the human subjects of algorithmic classification.10 10.For consideration of these issues, see Mariano-Florentino Cuéllar & Aziz Z. Huq, Economies of Surveillance, 133 Harv. L. Rev. 1280 (2020), and Mariano-Florentino Cuéllar & Aziz Z. Huq, Privacy’s Political Economy and the State of Machine Learning: An Essay in Honor of Stephen J. Schulhofer, N.Y.U. Ann. Surv. Am. L. (forthcoming 2020).Show More Even the basic terms of political choice seem compromised.11 11.See, e.g., Daniel Kreiss & Shannon C. McGregor, Technology Firms Shape Political Communication: The Work of Microsoft, Facebook, Twitter, and Google with Campaigns During the 2016 U.S. Presidential Cycle, 35 Pol. Comm. 155, 156–57 (2018) (describing the role of technology firms in shaping campaigns).Show More At the same time that machine learning is poised to recalibrate the ordinary forms of interaction between citizen and government (or big tech), advances in robotics as well as machine learning appear to be about to displace huge tranches of both blue-collar and white-collar labor markets.12 12.For what has become the standard view, see Larry Elliott, Robots Will Take Our Jobs. We’d Better Plan Now, Before It’s Too Late, Guardian (Feb. 1, 2018, 1:00 AM), https://www.theguardian.com/commentisfree/2018/feb/01/robots-take-our-jobs-amazon-go-seattle [https://perma.cc/2CFP-3JJV]. For a more nuanced account, see Martin Ford, Rise of the Robots: Technology and the Threat of a Jobless Future 282–83 (2015).Show More A fearful future looms, one characterized by massive economic dislocation, wherein people have lost control of many central life choices, and basic consumer and political preferences are no longer really one’s own.

This Article is about one nascent and still inchoate legal response to these fears: the possibility that an individual being assigned a benefit or a coercive intervention has a right to a human decision rather than a decision reached by a purely automated process (a “machine decision”). European law has embraced the idea. American law, especially in the criminal justice domain, is flirting with it.13 13.See infra text accompanying notes 70–73.Show More My aim in this Article is to test this burgeoning proposal, to investigate its relationship with technological possibilities, and to ascertain whether it is a cogent response to growing distributional, political, and epistemic anxieties. My focus is not on the form of such a right—statutory, constitutional, or treaty-based—or how it is implemented—say, in terms of liability or property rule protection—but more simply on what might ab initio justify its creation.

To motivate this inquiry, consider some of the anxieties unfurling already in public debate: A nursing union, for instance, launched a campaign urging patients to demand human medical judgments rather than technological assessment.14 14.‘When It Matters Most, Insist on a Registered Nurse,’ Nat’l Nurses United, https://www.­nationalnursesunited.org/insist-registered-nurse [https://perma.cc/MB66-XTXW] (last visited Jan. 19, 2020).Show More And a majority of patients surveyed in a 2018 Accenture survey preferred treatment by a doctor in person to virtual care.15 15.Accenture Consulting, 2018 Consumer Survey on Digital Health: US Results 9 (2018), https://www.accenture.com/_acnmedia/PDF-71/Accenture-Health-2018-Consumer-Survey-Digital-Health.pdf#zoom=50 [https://perma.cc/TU5F-9J82].Show More When California proposed replacing money bail with a “risk-based pretrial assessment” tool, a state court judge warned that “[t]echnology cannot replace the depth of judicial knowledge, experience, and expertise in law enforcement that prosecutors and defendants’ attorneys possess.”16 16.Quentin L. Kopp, Replacing Judges with Computers Is Risky, Harv. L. Rev. Blog (Feb. 20, 2018), https://blog.harvardlawreview.org/replacing-judges-with-computers-is-risky/ [https://perma.cc/WS5S-ARVF]. On the current state of affairs, see California Set to Greatly Expand Controversial Pretrial Risk Assessments, Filter (Aug. 7, 2019), https://filtermag.org/­california-slated-to-greatly-expand-controversial-pretrial-risk-assessments/ [https://perma.cc­/2FNX-U3C9].Show More In 2018, the City of Flint, Michigan, discontinued the use of a highly effective machine-learning tool designed to identify defective water pipes, reverting under community pressure to human decision making with a far lower hit rate for detecting defective pipes.17 17.Alexis C. Madrigal, How a Feel-Good AI Story Went Wrong in Flint, Atlantic (Jan. 3, 2019), https://www.theatlantic.com/technology/archive/2019/01/how-machine-learning-fou­nd-flints-lead-pipes/578692/ [https://perma.cc/V8VA-F22W].Show More Finally, and perhaps most powerfully, consider the worry congealed in an anecdote told by data scientist Cathy O’Neil: An Arkansas woman named Catherine Taylor is denied federal housing assistance because she fails an automated, “webcrawling[,] data-gathering” background check.18 18.Cathy O’Neil, Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy 152–53 (2016).Show More It is only when “one conscientious human being” takes the trouble to look into the quality of this machine result that it is discovered that Taylor has been red-flagged in error.19 19.Id. at 153.Show More O’Neil’s plainly troubling anecdote powerfully captures the fear that machines will be unfair, incomprehensive, or incompatible with the flexing of elementary human agency: it provides a sharp spur to the inquiry that follows.

The most important formulation of a right to a human decision to date is found in European law. In April 2016, the European Parliament enacted a new regime of data protection in the form of a General Data Protection Regulation (GDPR).20 20.Regulation 2016/679, of the European Parliament and of the Council of 27 April 2016 on the Protection of Natural Persons with Regard to the Processing of Personal Data and on the Free Movement of Such Data, and Repealing Directive 95/46/EC (General Data Protection Regulation), 2016 O.J. (L 119) (EU) [hereinafter GDPR]; see also Christina Tikkinen-Piri, Anna Rohunen & Jouni Markkula, EU General Data Protection Regulation: Changes and Implications for Personal Data Collecting Companies, 34 Computer L. & Security Rev. 134, 134–35 (2018) (documenting the enactment process of the GDPR).Show More Unlike the legal regime it superseded,21 21.See Directive 95/46, of the European Parliament and of the Council of 24 October 1995 on the Protection of Individuals with Regard to the Processing of Personal Data and on the Free Movement of Such Data, art. 1, 1995 O.J. (L 281) (EC) [hereinafter Directive 95/46].Show More the GDPR as implemented in May 2018 is legally mandatory even in the absence of implementing legislation by member states of the European Union (EU).22 22.Bryce Goodman & Seth Flaxman, European Union Regulations on Algorithmic Decision Making and a “Right to Explanation,” AI Mag., Fall 2017, at 51–52 (explaining the difference between a non-binding directive and a legally binding regulation under European law).Show More Hence, it can be directly enforced in court through hefty financial penalties.23 23.Id. at 52.Show More Article 22 of the GDPR endows a natural individual with “the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.”24 24.GDPR, supra note 20, arts. 4(1), 22(1) (inter alia, defining “data subject”).Show More That right covers private and some (but not all) state entities.25 25.See id. art. 4(7)–(8) (defining “controller” and “processor” as key scope terms). The Regulation, however, does not apply to criminal and security investigations. Id. art. 2(2)(d).Show More On its face, it fashions an opt-out of quite general scope from automated decision making.26 26.As I explain below, this is not the only provision of the GDPR that can be interpreted to create a right to a human decision. See infra text accompanying notes 53–58.Show More

The GDPR also has extraterritorial effect.27 27.GDPR, supra note 20, art. 3.Show More It reaches platforms, such as Google and Facebook, that offer services within the EU.28 28.There is sharp divergence in the scholarship over the GDPR’s extraterritorial scope, which ranges from the measured, see Griffin Drake, Note, Navigating the Atlantic: Understanding EU Data Privacy Compliance Amidst a Sea of Uncertainty, 91 S. Cal. L. Rev. 163, 166 (2017) (documenting new legal risks to American companies pursuant to the GDPR), to the alarmist, see Mira Burri, The Governance of Data and Data Flows in Trade Agreements: The Pitfalls of Legal Adaptation, 51 U.C. Davis L. Rev. 65, 92 (2017) (“The GDPR is, in many senses, excessively burdensome and with sizeable extraterritorial effects.”).Show More And American law is also making tentative moves toward a similar right to a human decision. In 2016, for example, the Wisconsin Supreme Court held that an algorithmically generated risk score “may not be considered as the determinative factor in deciding whether the offender can be supervised safely and effectively in the community” as a matter of due process.29 29.State v. Loomis, 881 N.W.2d 749, 760 (Wis. 2016).Show More That decision precludes full automation of bail determinations. There must be a human judge in the loop. The Wisconsin court’s holding is unlikely to prove unique. State deployment of machine learning has, more generally, elicited sharp complaints sounding in procedural justice and fairness terms.30 30.See, e.g., Julia Angwin, Jeff Larson, Surya Mattu & Lauren Kirchner, Machine Bias: There’s Software Used Across the Country to Predict Future Criminals. And It’s Biased Against Blacks, ProPublica 2 (May 23, 2016), https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing [https://perma.cc/Q9ZU-VY6J] (criticizing machine-learning instruments in the criminal justice context).Show More Further, the Sixth Amendment’s right to a jury trial has to date principally been deployed to resist judicial factfinding.31 31.See, e.g., Apprendi v. New Jersey, 530 U.S. 466, 477 (2000) (explaining that the Fifth and Sixth Amendments “indisputably entitle a criminal defendant to a jury determination that [he] is guilty of every element of the crime with which he is charged, beyond a reasonable doubt” (alteration in original) (internal quotation marks omitted) (quoting United States v. Gaudin, 515 U.S. 506, 510 (1995))).Show More But there is no conceptual reason why the Sixth Amendment could not be invoked to preclude at least some forms of algorithmically generated inputs to criminal sentencing. Indeed, it would seem to follow a fortiori that a right precluding a jury’s substitution with a judge would also block its displacement by a mere machine.

In this Article, I start by situating a right to a human decision in its contemporary technological milieu. I can thereby specify the feasible domain of machine decisions. I suggest this comprises decisions taken at high volume in which sufficient historical data exists to generate effective predictions. Importantly, this excludes many matters presently resolved through civil or criminal trials but sweeps in welfare determinations, hiring decisions, and predictive judgments in the criminal justice contexts of bail and sentencing. Second, I examine the margins along which machine decisions are distinct from human ones. My focus is on a group of related technologies known as machine learning. This is the form of artificial intelligence diffusing most rapidly today.32 32.See infra text accompanying note 88 (defining machine learning). I am not alone in this focus. Legal scholars are paying increasing attention to new algorithmic technologies. For leading examples, see Kate Crawford & Jason Schultz, Big Data and Due Process: Toward a Framework to Redress Predictive Privacy Harms, 55 B.C. L. Rev. 93, 109 (2014) (arguing for “procedural data due process [to] regulate the fairness of Big Data’s analytical processes with regard to how they use personal data (or metadata . . . )”); Andrew Guthrie Ferguson, Big Data and Predictive Reasonable Suspicion, 163 U. Pa. L. Rev. 327, 383–84 (2015) (discussing the possible use of algorithmic prediction in determining “reasonable suspicion” in criminal law); Kroll et al., supra note 2, at 636–37; Michael L. Rich, Machine Learning, Automated Suspicion Algorithms, and the Fourth Amendment, 164 U. Pa. L. Rev. 871, 929 (2016) (developing a “framework” for integrating machine-learning technologies into Fourth Amendment analysis).Show More A right to a human decision cannot be defined or evaluated without some sense of the technical differences between human decision making and decisions reached by these machine-learning technologies. Indeed, careful analysis of how machine learning is designed and implemented reveals that the distinctions between human and machine decisions are less crisp than might first appear. Claims about a right to human decision, I suggest, are better understood to turn on the timing, and not the sheer fact, of such involvement.

With this technical foundation in hand, I evaluate the right to a human decision in relation to four normative ends it might plausibly be understood to further. A first possibility turns on overall accuracy worries. My second line of analysis takes up the interests of an individual exposed to a machine decision. The most pertinent of these interests hinge upon an individual’s participation in decision making and her opportunity to offer reasons. A third analytic salient tracks ways that a machine instrument might be intrinsically objectionable because it uses a deficient decisional protocol. I focus here on worries about the absence of individualized consideration and a machine’s failure to offer reasoned judgments. Finally, I consider dynamic, system-level effects (i.e., negative spillovers), in particular in relation to social power. None of these arguments ultimately provides sure ground for a legal right to a human decision.

Rather, I suggest that the limits of machine decision making be plotted based on its technical constraints. Machines should not be used when there is no tractable parameter amenable to prediction. For example, if there is no good parameter that tracks job performance, then machine evaluation of those employees should be abandoned. Nor should they be used when decision making entails ethical or otherwise morally charged judgments. Most important, I suggest that machine decisions should be subject to a right to a well-calibrated machine decision that folds in due process, privacy, and equality values.33 33.A forthcoming companion piece develops a more detailed account of how this right would be vindicated in practice through a mix of litigation and regulation. See Aziz Z. Huq, Constitutional Rights in the Machine Learning State, 105 Cornell L. Rev. (forthcoming 2020).Show More This is a better response than a right to a human decision to the many instruments now implemented by the government that are highly flawed.34 34.For a catalog, see Meredith Whittaker et al., AI Now Inst., AI Now Report 2018, at 18–22 (2018), https://ainowinstitute.org/AI_Now_2018_Report.pdf [https://perma.cc/2BCG-M4­54].Show More

My analysis here focuses on state action that imposes benefits or coercion on individuals—and not on either private action or a broader array of state action—for three reasons. First, salient U.S. legal frameworks, unlike the GDPR’s coverage, are largely (although not exclusively) trained on state action. Accordingly, a focus on state action makes sense in terms of explaining and evaluating the current U.S. regulatory landscape. Second, the range of private uses of algorithmic tools is vast and heterogenous. Algorithms are now deployed in private activities ranging from Google’s PageRank instrument,35 35.See, e.g., David Segal, The Dirty Little Secrets of Search: Why One Retailer Kept Popping Up as No. 1, N.Y. Times, Feb. 13, 2011, at BU1.Show More to “fintech” applied to generate new revenue streams,36 36.See Falguni Desai, The Age of Artificial Intelligence in Fintech, Forbes (June 30, 2016, 10:42 PM), http://www.forbes.com/sites/falgunidesai/2016/06/30/the-age-of-artificial-intelli­gence-in-fintech [https://perma.cc/DG8N-8NVS] (describing how fintech firms use artificial intelligence to improve investment strategies and analyze consumer financial activity).Show More to medical instruments used to calculate stroke risk,37 37.See, e.g., Benjamin Letham, Cynthia Rudin, Tyler H. McCormick & David Madigan, Interpretable Classifiers Using Rules and Bayesian Analysis: Building a Better Stroke Prediction Model, 9 Annals Applied Stat. 1350, 1350 (2015).Show More to engineers’ identification of new stable inorganic compounds.38 38.See, e.g., Paul Raccuglia et al., Machine-Learning-Assisted Materials Discovery Using Failed Experiments, 533 Nature 73, 73 (2016) (identifying new vanadium compounds).Show More Algorithmic tools are also embedded within new applications, such as voice recognition software, translation software, and visual recognition systems.39 39.Yann LeCun et al., Deep Learning, 521 Nature 436, 438–41 (2015).Show More In contrast, the state is to date an unimaginative user of machine learning, with a relatively constrained domain of deployments.40 40.See infra text accompanying notes 117–21 (describing state uses of machine learning).Show More This makes for a more straightforward analysis. Third, where the state does use algorithmic tools, it often results directly or indirectly in deprivations of liberty, freedom of movement, bodily integrity, or basic income. These normatively freighted machine decisions present arguably the most compelling circumstances for adopting a right to a human decision and so are a useful focus of normative inquiry.

The Article proceeds in three steps. Part I catalogs ways in which law has crafted, or could craft, a right to a human decision. This taxonomical enterprise demonstrates that such a right is far from fanciful. Part II defines the class of computational tools to be considered, explores the manner in which such instruments can be used, and teases out how they are (or are not) distinct from human decisions. Doing so helps illuminate the plausible forms of a right to a human decision. Part III then turns to the potential normative foundations of such a right. It provides a careful taxonomy of those grounds. It then shows why they all fall short. Finally, a brief conclusion inverts the Article’s analytic lens to gesture at the possibility that a right to a well-calibrated machine decision can be imagined, and even defended, on more persuasive terms than a right to a human decision.

  1. * Frank and Bernice J. Greenberg Professor of Law, University of Chicago Law School. Thanks to Faith Laken for terrific research aid. Thanks to Tony Casey, David Driesen, Lauryn Gouldin, Daniel Hemel, Darryl Li, Anup Malani, Richard McAdams, Eric Posner, Julie Roin, Lior Strahilevitz, Rebecca Wexler, and Annette Zimmermann for thoughtful conversation. Workshop participants at the University of Chicago, Stanford Law School, the University of Houston, William and Mary Law School, and Syracuse University School of Law also provided thoughtful feedback. I am grateful to Christiana Zgourides, Erin Brown, and the other law review editors for their careful work on this Article. All errors are mine, not the machine’s. ↑
  2. For recent treatments of these technological causes of social transformations, see generally James C. Scott, Against the Grain: A Deep History of the Earliest States (2017), and Ravi Agrawal, India Connected: How the Smartphone is Transforming the World’s Largest Democracy (2018). ↑
  3. An algorithm is simply a “well-defined set of steps for accomplishing a certain goal.” Joshua A. Kroll et al., Accountable Algorithms, 165 U. Pa. L. Rev. 633, 640 n.14 (2017); see also Thomas H. Cormen et al., Introduction to Algorithms 5 (3d ed. 2009) (defining an algorithm as “any well-defined computational procedure that takes some value, or set of values, as input and produces some value, or set of values, as output” (emphasis omitted)). The task of computing, at its atomic level, comprises the execution of serial algorithms. Martin Erwig, Once Upon an Algorithm: How Stories Explain Computing 1–4 (2017). ↑
  4. Machine learning is a general purpose technology that, in broad terms, encompasses “algorithms and systems that improve their knowledge or performance with experience.” Peter Flach, Machine Learning: The Art and Science of Algorithms that Make Sense of Data 3 (2012); see also Ethem Alpaydin, Introduction to Machine Learning 2–3 (3d ed. 2014) (defining machine learning in similar terms). For the uses of machine learning, see Susan Athey, Beyond Prediction: Using Big Data for Policy Problems, 355 Science 483, 483 (2017) (noting the use of machine learning to solve prediction problems). I discuss the technological scope of the project, and define relevant terms, infra at text accompanying note 111. I will use the terms “algorithmic tools” and “machine learning” interchangeably, even though the class of algorithms is technically much larger. ↑
  5. Kartik Hosanagar & Vivian Jair, We Need Transparency in Algorithms, But Too Much Can Backfire, Harv. Bus. Rev. (July 23, 2018), https://hbr.org/2018/07/we-need-transparency-in-algorithms-but-too-much-can-backfire [https://perma.cc/7KQ9-QMF3]; accord Cary Coglianese & David Lehr, Regulating by Robot: Administrative Decision Making in the Machine-Learning Era, 105 Geo. L.J. 1147, 1149 (2017). ↑
  6. Shoshana Zuboff, Big Other: Surveillance Capitalism and the Prospects of an Information Civilization, 30 J. Info. Tech. 75, 75 (2015) (describing a “new form of information capitalism [that] aims to predict and modify human behavior as a means to produce revenue and market control”). ↑
  7. See, e.g., Rachel Courtland, The Bias Detectives, 558 Nature 357, 357 (2018) (documenting concerns among the public that algorithmic risk scores for detecting child abuse fail to account for an “effort . . . to turn [a] life around”). ↑
  8. Reuben Binns et al., ‘It’s Reducing a Human Being to a Percentage’; Perceptions of Justice in Algorithmic Decisions, 2018 CHI Conf. on Hum. Factors Computing Systems 9 (emphasis omitted). ↑
  9. Virginia Eubanks, Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor 168 (2017). ↑
  10. Will Knight, The Dark Secret at the Heart of AI, MIT Tech. Rev. (Apr. 11, 2017), https://www.technologyreview.com/s/604087/the-dark-secret-at-the-heart-of-ai/ [https://perma.cc/L94L-LYTJ] (“The computers that run those services have programmed themselves, and they have done it in ways we cannot understand. Even the engineers who build these apps cannot fully explain their behavior.”). ↑
  11. For consideration of these issues, see Mariano-Florentino Cuéllar & Aziz Z. Huq, Economies of Surveillance, 133 Harv. L. Rev. 1280 (2020), and Mariano-Florentino Cuéllar & Aziz Z. Huq, Privacy’s Political Economy and the State of Machine Learning: An Essay in Honor of Stephen J. Schulhofer, N.Y.U. Ann. Surv. Am. L. (forthcoming 2020). ↑
  12. See, e.g., Daniel Kreiss & Shannon C. McGregor, Technology Firms Shape Political Communication: The Work of Microsoft, Facebook, Twitter, and Google with Campaigns During the 2016 U.S. Presidential Cycle, 35 Pol. Comm. 155, 156–57 (2018) (describing the role of technology firms in shaping campaigns). ↑
  13. For what has become the standard view, see Larry Elliott, Robots Will Take Our Jobs. We’d Better Plan Now, Before It’s Too Late, Guardian (Feb. 1, 2018, 1:00 AM), https://www.theguardian.com/commentisfree/2018/feb/01/robots-take-our-jobs-amazon-go-seattle [https://perma.cc/2CFP-3JJV]. For a more nuanced account, see Martin Ford, Rise of the Robots: Technology and the Threat of a Jobless Future 282–83 (2015). ↑
  14. See infra text accompanying notes 70–73. ↑
  15. ‘When It Matters Most, Insist on a Registered Nurse,’ Nat’l Nurses United, https://www.­nationalnursesunited.org/insist-registered-nurse [https://perma.cc/MB66-XTXW] (last visited Jan. 19, 2020). ↑
  16. Accenture Consulting, 2018 Consumer Survey on Digital Health: US Results 9 (2018), https://www.accenture.com/_acnmedia/PDF-71/Accenture-Health-2018-Consumer-Survey-Digital-Health.pdf#zoom=50 [https://perma.cc/TU5F-9J82]. ↑
  17. Quentin L. Kopp, Replacing Judges with Computers Is Risky, Harv. L. Rev. Blog
    (Feb. 20, 2018), https://blog.harvardlawreview.org/replacing-judges-with-computers-is-risky/ [https://perma.cc/WS5S-ARVF]. On the current state of affairs, see California Set to Greatly Expand Controversial Pretrial Risk Assessments, Filter (Aug. 7, 2019), https://filtermag.org/­california-slated-to-greatly-expand-controversial-pretrial-risk-assessments/ [https://perma.cc­/2FNX-U3C9]. ↑
  18. Alexis C. Madrigal, How a Feel-Good AI Story Went Wrong in Flint, Atlantic (Jan. 3, 2019), https://www.theatlantic.com/technology/archive/2019/01/how-machine-learning-fou­nd-flints-lead-pipes/578692/ [https://perma.cc/V8VA-F22W]. ↑
  19. Cathy O’Neil, Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy 152–53 (2016). ↑
  20. Id. at 153. ↑
  21. Regulation 2016/679, of the European Parliament and of the Council of 27 April 2016 on the Protection of Natural Persons with Regard to the Processing of Personal Data and on the Free Movement of Such Data, and Repealing Directive 95/46/EC (General Data Protection Regulation), 2016 O.J. (L 119) (EU) [hereinafter GDPR]; see also Christina Tikkinen-Piri, Anna Rohunen & Jouni Markkula, EU General Data Protection Regulation: Changes and Implications for Personal Data Collecting Companies, 34 Computer L. & Security Rev. 134, 134–35 (2018) (documenting the enactment process of the GDPR). ↑
  22. See Directive 95/46, of the European Parliament and of the Council of 24 October 1995 on the Protection of Individuals with Regard to the Processing of Personal Data and on the Free Movement of Such Data, art. 1, 1995 O.J. (L 281) (EC) [hereinafter Directive 95/46]. ↑
  23. Bryce Goodman & Seth Flaxman, European Union Regulations on Algorithmic Decision Making and a “Right to Explanation,” AI Mag., Fall 2017, at 51–52 (explaining the difference between a non-binding directive and a legally binding regulation under European law). ↑
  24. Id. at 52. ↑
  25. GDPR, supra note 20, arts. 4(1), 22(1) (inter alia, defining “data subject”). ↑
  26. See id. art. 4(7)–(8) (defining “controller” and “processor” as key scope terms). The Regulation, however, does not apply to criminal and security investigations. Id. art. 2(2)(d). ↑
  27. As I explain below, this is not the only provision of the GDPR that can be interpreted to create a right to a human decision. See infra text accompanying notes 53–58. ↑
  28. GDPR, supra note 20, art. 3. ↑
  29. There is sharp divergence in the scholarship over the GDPR’s extraterritorial scope, which ranges from the measured, see Griffin Drake, Note, Navigating the Atlantic: Understanding EU Data Privacy Compliance Amidst a Sea of Uncertainty, 91 S. Cal. L. Rev. 163, 166 (2017) (documenting new legal risks to American companies pursuant to the GDPR), to the alarmist, see Mira Burri, The Governance of Data and Data Flows in Trade Agreements: The Pitfalls of Legal Adaptation, 51 U.C. Davis L. Rev. 65, 92 (2017) (“The GDPR is, in many senses, excessively burdensome and with sizeable extraterritorial effects.”). ↑
  30. State v. Loomis, 881 N.W.2d 749, 760 (Wis. 2016). ↑
  31. See, e.g., Julia Angwin, Jeff Larson, Surya Mattu & Lauren Kirchner, Machine Bias: There’s Software Used Across the Country to Predict Future Criminals. And It’s Biased Against Blacks, ProPublica 2 (May 23, 2016), https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing [https://perma.cc/Q9ZU-VY6J] (criticizing machine-learning instruments in the criminal justice context). ↑
  32. See, e.g., Apprendi v. New Jersey, 530 U.S. 466, 477 (2000) (explaining that the Fifth and Sixth Amendments “indisputably entitle a criminal defendant to a jury determination that [he] is guilty of every element of the crime with which he is charged, beyond a reasonable doubt” (alteration in original) (internal quotation marks omitted) (quoting United States v. Gaudin, 515 U.S. 506, 510 (1995))). ↑
  33. See infra text accompanying note 88 (defining machine learning). I am not alone in this focus. Legal scholars are paying increasing attention to new algorithmic technologies. For leading examples, see Kate Crawford & Jason Schultz, Big Data and Due Process: Toward a Framework to Redress Predictive Privacy Harms, 55 B.C. L. Rev. 93, 109 (2014) (arguing for “procedural data due process [to] regulate the fairness of Big Data’s analytical processes with regard to how they use personal data (or metadata . . . )”); Andrew Guthrie Ferguson, Big Data and Predictive Reasonable Suspicion, 163 U. Pa. L. Rev. 327, 383–84 (2015) (discussing the possible use of algorithmic prediction in determining “reasonable suspicion” in criminal law); Kroll et al., supra note 2, at 636–37; Michael L. Rich, Machine Learning, Automated Suspicion Algorithms, and the Fourth Amendment, 164 U. Pa. L. Rev. 871, 929 (2016) (developing a “framework” for integrating machine-learning technologies into Fourth Amendment analysis). ↑
  34. A forthcoming companion piece develops a more detailed account of how this right would be vindicated in practice through a mix of litigation and regulation. See Aziz Z. Huq, Constitutional Rights in the Machine Learning State, 105 Cornell L. Rev. (forthcoming 2020). ↑
  35. For a catalog, see Meredith Whittaker et al., AI Now Inst., AI Now Report 2018, at 18–22 (2018), https://ainowinstitute.org/AI_Now_2018_Report.pdf [https://perma.cc/2BCG-M4­54]. ↑
  36. See, e.g., David Segal, The Dirty Little Secrets of Search: Why One Retailer Kept Popping Up as No. 1, N.Y. Times, Feb. 13, 2011, at BU1. ↑
  37. See Falguni Desai, The Age of Artificial Intelligence in Fintech, Forbes (June 30, 2016, 10:42 PM), http://www.forbes.com/sites/falgunidesai/2016/06/30/the-age-of-artificial-intelli­gence-in-fintech [https://perma.cc/DG8N-8NVS] (describing how fintech firms use artificial intelligence to improve investment strategies and analyze consumer financial activity). ↑
  38. See, e.g., Benjamin Letham, Cynthia Rudin, Tyler H. McCormick & David Madigan, Interpretable Classifiers Using Rules and Bayesian Analysis: Building a Better Stroke Prediction Model, 9 Annals Applied Stat. 1350, 1350 (2015). ↑
  39. See, e.g., Paul Raccuglia et al., Machine-Learning-Assisted Materials Discovery Using Failed Experiments, 533 Nature 73, 73 (2016) (identifying new vanadium compounds). ↑
  40. Yann LeCun et al., Deep Learning, 521 Nature 436, 438–41 (2015). ↑
  41. See infra text accompanying notes 117–21 (describing state uses of machine learning). ↑
©Virginia Law Review Association. All rights reserved.