Author:Allen, James A.

TABLE OF CONTENTS INTRODUCTION 221 I. "This is the Procedure": A Brief History of Algorithms, from Byzantine to Boolean 225 II. Algorithmic Redlining: The Threat of Autogenerating Segregaton 230 A. Potential Harms of Discriminatory Credit and Lending Algorithms 235 1. Credit Card Scoring and Algorithms Giving Access to 237 Loans 2. Algorithms Used in Reverse Redlining, Subprime Mortgages, and Payday Loans 239 3. Discriminatory Algorithms in Housing Advertisements and Marketing 241 B. When You Can't Buy, Rent: Affordable Housing and Rental Housing Selection Algorithms 246 1. Improper Preferences in Affordable Housing Algorithms 248 2. Rental Housing Algorithms 252 III. Policies and Solutions: Looking Back and Looking Ahead 253 A. Transparency: Promoting Choice and Accountability Through an Understanding of Automated Decisions 256 B. Auditing Algorithms for Fairness 258 C. Human Oversight and Autonomy: Algorithms as a Method of Retaining Free Will 260 D. Other Reforms and Modernizations 262 1. Improvements to Intellectual Property, Data Protection, and Internet Law 262 2. Adjustments to Internet Law 265 3. Modernizing Public Administration 267 4. Changes by Private Actors 268 CONCLUSION 269 After the lecture, when the judge said, "I'm going to give you boys another chance," I don't know why or what happened, but I heard myself say, "Man, you not givin' us another chance. You givin' us the same chance we had before." - Claude Brown, MANCHILD IN THE PROMISED LAND (1) Harlem, New York INTRODUCTION

Algorithms, or automated decision systems, are being used by public and private entities to optimize efficiency, cut costs, and expand social welfare. (2) Although beneficial in many respects, algorithms deployed by public and private actors also make decisions that go against our better instinct. (3) Algorithms are now being used in the affordable and fair housing arena--to ensure equitable results, these algorithms must be implemented with transparency, auditing, and oversight. (4) This Article explores the potentially inequitable uses of big data and algorithms in the housing arena and suggests how different actors in the United States may be perpetuating a previous era of "redlining."

In its most technical sense, the term "redlining" can be traced back to the 1933 practice of racial discrimination stewarded by the United States government's Home Owners' Loan Corporation (HOLC). (3) At the time, the HOLC was the nation's largest financer of federally-backed government loans, lending to individuals and families to buy homes. (6) However, the HOLC did not disburse these government-backed loans equitably--the agency would literally draw red lines on maps around communities of color, which they identified as too risky to serve. (7) This practice etymologized what became widely known as "redlining," as underwriters across the country followed suit in an attempt to adhere to the federal government's "risk-evaluation" standards. (8)

As the practice of redlining expanded, so did its meaning. The term "redlining" has since become more encompassing and is now commonly considered to describe the general practice of an institution's refusal to provide resources and financial support to areas considered "high-risk." (9) Unfortunately, these "high-risk" areas remain disproportionately those with properties owned by people of color, due in large part to the previous era of technical redlining. (10) Though the housing industry has moved away from maps and red pencils, "redlining" and its historical impact remain relevant today. This Article cautions that overreliance on automated, algorithmic decision-making systems may perpetuate housing segregation through "algorithmic redlining." In line with the modern, broader understanding of redlining, this Article uses the term "algorithmic redlining" to encompass sets of instructions--simple or complex--which carry out procedures that prohibit or limit people of color from procuring housing or housing financing, particularly in non-minority neighborhoods. (11) Algorithmic redlining and the original era of pencil redlining are synchronized in a crucial way: both result in the exclusion of minority and low-income members of society from access to adequate housing. (12)

This Article explores the impact and potential harms of overreliance on automated decision systems, specifically in the housing equity context, and proceeds in three parts. Part I provides a brief history of algorithms, from their origins as a basic set of steps to complex, autonomous procedures. This includes a discussion of the rise of "big data" and an exploration of how machine learning has intensified the proliferation of automated decision systems. Part I also addresses some of the recent concerns about algorithms as expressed by community activists, academics, and lawmakers--as algorithms become more expansive, so too have concerns about their adverse impact on society. A discussion of this history is important in understanding how pencil redlining was, in a way, an algorithm: "area" plus "colored people" equals "do not lend." It is also important to appreciate how redlining resulted in adverse consequences for people of color, consequences that make up the data modern algorithms use to generate decisions.

Part II begins with a brief discussion of the importance of access to housing equity, and then offers a research agenda to explore how algorithmic redlining has the potential to exacerbate existing segregation. This Part examines the use of algorithms in three particular areas of the housing market and explores the segregation that can result when algorithms themselves are discriminatory. The first, and perhaps most important, area is that of housing finance--because of biases built into automated credit evaluation and mortgage lending decisions, algorithms act as initial barriers for a subset of people attempting to access credit and housing finance. (13) This initial barrier to financing comes in the form of scoring systems that rate the riskiness of credit and mortgage applicants, as well as algorithms used to target minorities and low-income individuals for usurious loans. (14) By reviewing initial reports of systems that rely on these kinds of algorithms, it is possible to see that they result in disproportionately negative lending practices towards people of color. The second area of potential biases is found in the algorithms that direct advertising or marketing towards different races. While present research concerning online advertising in the housing ownership and rental market is scant, bias in other areas of online discriminatory marketing is reason for concern. (16) Finally, the third area of potential bias is in unfair algorithms that may be used in the housing selection process. This is exhibited by recent calls of housing advocates for reform in offline affordable housing lottery algorithms (17) and in the algorithms used to evaluate private market rental applicants. (18)

Part III reviews previous reforms and discusses how these once beneficial policies are outdated in the modern, online housing economy. This Part borrows from previous policies to suggest various ways in which the adverse impacts of algorithmic redlining can be curbed: transparency, oversight, and greater human autonomy in automated decision-making. These suggestions draw on previous legislation in the United States and contemporary legislation in the European Union, while also advocating for reforms previously put forth by academics and stakeholders, who call for a modernization of laws impacting the areas mentioned above. While some of these reforms may seem improbable, particularly given America's political climate, this Part discusses why these laws are really commonsense and necessary to protect the modern consumer.


    In their simplest form, "algorithms" are processes that rely on a specific set of steps to consistently produce the same result. (19) When an entity with knowledge of how a given procedure functions conveys this information in a step-by-step formulation, it can reproduce the steps or particular method of analysis for the given criteria as an algorithm. (20) Once conveyed in this way, anyone can apply data to the algorithm, and the results remain predictable. (21)

    In a sense, algorithms have always been an engine that drives societal advancement. Some of the earliest algorithms were used to advance community and economic development by conveying how to build infrastructure or how to determine the rate of carried interest on a potential investment. (22) For example, early societies used algorithmic methods to manage the crucial practice of water management: if a person knew the length, width, and height of a certain area, with the appropriate algorithm, that person could use that data to build a cistern. (23) If a person knew they had capital commitments from potential investors for their cistern-building business, with the appropriate algorithm, they could take the investor's pledged commitments and calculate their potential carried interest. (24) These basic examples are not too different from those used in recent history or even today.

    While ancient algorithms are similar in their objective to today's algorithms, the latter are obviously far more complex. (25) Modern algorithms are tremendously intricate and take many forms. Most relevant to the discussion here are "machine learning" algorithms, (26) which are informed and powered by "big data." (27) In a broad sense, big data is the process of aggregating massive amounts of information from various online platforms and data capturing entities for the purpose of identifying potential patterns; machine learning algorithms are a form of artificial intelligence, which operate to process big data, learn from it, and then perform tasks and analytics. (28) These are such...

To continue reading