October 18, 2019
Office of the General Counsel
Rules Docket Clerk
Department of Housing and Urban Development
451 Seventh Street SW, Room 10276
Washington, D.C. 20410-0001
Re: Docket No. FR-6111-P-02 - HUD’s Implementation of the Fair Housing Act’s
Disparate Impact Standard
Dear Sir/Madam:
The Center on Race, Inequality, and the Law at New York University School
of Law (“Center”) is a multifaceted institute created to confront laws, policies, and
practices that lead to the oppression and marginalization of people of color. AI Now
Institute at New York University (“AI Now”) is an interdisciplinary research center
dedicated to understanding the social implications of artificial intelligence.
We submit this public comment on behalf of the Center, AI Now and the
undersigned collection of scholars in response to the Department of Housing and
Urban Development’s (“HUD”) proposed rule to amend the interpretation of the Fair
Housing Act’s (“FHA”) disparate impact standard. The Center, AI Now, and the
signatories to this comment letter are particularly concerned with the use of
algorithmic decision-making tools that can exacerbate existing racial disparities and
racial discrimination against communities of color. HUD's proposed rule heightens
the already significant burden that plaintiffs must meet to raise claims of housing
discrimination, and allows defendants accused of housing discrimination to use
algorithmic tools to shield themselves from liability for discriminatory behavior.
The proposed rule is concerning because: (1) it is antithetical to the purpose of
the FHA; (2) it will significantly weaken the effectiveness of the disparate impact
standard by enabling racial discrimination when facilitated by algorithms to escape

 enforcement; and (3) by enabling racial discrimination, it will disadvantage
communities of color even further in the housing market. For these reasons, we
strongly oppose HUD’s proposed rule. We submit this joint public comment and urge
HUD to withdraw its proposed rule in light of the following considerations.
I. The Proposed Rule Is Antithetical to the Purpose of the Fair Housing Act, Which
Is to Eliminate Discriminatory Practices in Housing

1. The Fair Housing Act Was Enacted to Remedy Generations of Racial
Discrimination in Housing
Housing discrimination has firmly established its place in our country’s
narrative. This ugly reality circumscribed the composition of neighborhoods and
communities along racial lines and has perpetuated racial inequality for generations.
The federal government is not merely a passive player in all of this. Through its
policies and practices, the federal government has actively shaped the housing
market into its present-day unequal and discriminatory state. For example, the
Federal Housing Administration’s own Underwriting Manual once dictated that
“incompatible racial groups should not be permitted to live in the same communities,”
denying aspiring homeowners of color the fruits of the post-war housing boom. 1
Barriers like this led to diminished homeownership among Black and Brown
communities, and subsequently contributed to stark wealth disparities that persist
today. Numerous other government practices, such as urban renewal programs and
the strategic construction of highways, also displaced families of color by demolishing
their communities and subsequently limiting their access to homes and constraining
their mobility to relocate outside of disinvested neighborhoods. 2
In 1968, the presidentially established Kerner Commission released a report
declaring that “[o]ur nation is moving toward two societies, one black, one white—
separate and unequal.” 3 It is in this context that the FHA was enacted on April 11,
1968, one week after the assassination of Dr. Martin Luther King, Jr. Operating as a
bulwark against discrimination and a follow-up to the Civil Rights Act of 1964, the
FHA prohibited discrimination concerning the sale, rental, and financing of housing
based on race, religion, national origin, sex, (and as amended) handicap and family

Terry Gross, A 'Forgotten History' of How the U.S. Government Segregated America, NPR (May 3,
2017, 12:47 PM), https://www.npr.org/2017/05/03/526655831/a-forgotten-history-of-how-the-u-sgovernment-segregated-america.
2 See Marc Seitles, The Perpetuation of Residential Racial Segregation in America: Historical
Discrimination, Modern Forms of Exclusion, and Inclusionary Remedies, 14 J. LAND USE & ENVTL. L.
89, 91–106 (1998).
3 NAT’L ADVISORY COMM’N ON CIVIL DISORDER, THE KERNER REPORT 1 (Sean Wilentz ed., Princeton
University Press 2016) (1968).
1

2

 status. 4 HUD, entrusted with enforcing the FHA, was charged with “eliminat[ing]
housing discrimination, promot[ing] economic opportunity, and achiev[ing] diverse,
inclusive communities by leading the nation in the enforcement, administration,
development, and public understanding of federal fair housing policies and laws.” 5

2. The Disparate Impact Standard is a Critical Tool for Fulfilling HUD’s Mission
to Challenge Racial Discrimination
The wide-ranging harms inflicted by housing discrimination cannot be
adequately rectified without the disparate impact standard. The proposed rule
effectively undermines the strength of this safeguard and imposes unnecessary
barriers to disparate impact claims that have the ability to root out an array of
discriminatory housing practices that would otherwise go unchecked. Disparate
impact claims are employed to challenge exclusionary zoning practices that harm
people of color and the destruction of housing under the guise of “urban renewal.” 6
Disparate impact claims are also a tool to challenge the adverse effects of
neighborhood preference policies, arbitrary landlord screening practices, and racially
discriminatory mortgage underwriting and home insurance standards. 7 Disparate
impact liability is triggered “even if the practice was not motivated by a
discriminatory intent.” 8
On June 25, 2015, the Supreme Court upheld and recognized the critical role
that the disparate impact standard plays in Texas Department of Housing and
Community Affairs v. Inclusive Communities Project, Inc: “[U]nlawful practices . . .
that function unfairly to exclude minorities from certain neighborhoods without any
sufficient justification . . . reside at the heartland of disparate-impact liability.” 9 The
current HUD rule governing disparate impact claims is consistent with the Inclusive
Communities decision.

History of Fair Housing, U.S. DEP’T HOUSING & URB. DEV.,
https://www.hud.gov/program_offices/fair_housing_equal_opp/aboutfheo/history (last visited Oct. 14,
2019).
5 OFFICE OF FAIR HOUS. & EQUAL OPPORTUNITY, U.S. DEP’T OF HOUS. & URBAN DEV., ANNUAL REPORT
TO CONGRESS FY 2017, at 3 (2017)
https://www.hud.gov/sites/dfiles/FHEO/images/FHEO_Annual_Report_2017-508c.pdf.
6 See Robert G. Schwemm, Fair Housing Litigation After Inclusive Communities: What’s New and
What’s Not, 115 COLUM. L. REV. SIDEBAR 106 (2015), https://columbialawreview.org/content/fairhousing-litigation-after-inclusive-communities-whats-new-and-whats-not/.
7 See, e.g., Conn. Fair Hous. Ctr. v. CoreLogic Rental Prop. Sols., LLC, 369 F. Supp. 3d 362, 376–79
(D. Conn. 2019), http://www.ctfairhousing.com/PDFs/CoreLogicMTDOrder.pdf (finding a consumerreporting agency potentially liable under the FHA’s disparate impact standard for its use of a tenant
screening product).
8 24 C.F.R. § 100.500 (2019).
9 Tex. Dep't of Hous. & Cmty. Affairs v. Inclusive Cmtys. Project, Inc., 135 S. Ct. 2507, 2521–22
(2015) (citations omitted).
4

3

 Past practices continue to exert a striking contemporary impact. For example,
neighborhoods once “redlined”—marked by the Home Owners’ Loan Corporation as
hazardous risks to creditors due to the prevalence of residents of color—continue to
experience “both economic disadvantage and majority-minority presence.” 10 And
HUD’s own programs continue to subject Black renters to segregated conditions and
fewer housing options. 11
Despite HUD’s acknowledgement that housing “[d]iscrimination isn’t always
obvious,” 12 with this proposed rule, the agency is taking extraordinary steps to
dismantle the very protections and tools that counteract the often covert policies and
practices that disparately impact on communities of color. In light of both HUD’s
history of culpability and its mandate to affirmatively further fair housing, it is
incumbent upon HUD to ensure that people of all races can participate equally in the
housing market.
II. The Proposed Rule Will Significantly Weaken the Effectiveness of the Disparate
Impact Standard by Enabling Racial Discrimination Facilitated by Algorithms to
Escape Enforcement

1. The Use of Algorithmic Tools Can Perpetuate Racial Discrimination
Algorithmic tools operate within the historical and social context of the data
they use. This means that the destructive racial and social conditions of our society
are reflected within housing data and, in turn, the algorithmic tools that are trained
upon it. Therefore, while some argue that algorithms are favorable because they are
trained on extensive, empirically accurate, and objective data, such data actually
reflects existing and historical social inequities as well as discriminatory institutional
values, systems, and practices. One such example can be seen in the context of
predictive policing algorithms, for which the training data frequently reflects
historical police practices and policies rather than the actual prevalence and
BRUCE MITCHELL & JUAN FRANCO, NAT’L CMTY. REINVESTMENT COAL., HOLC “REDLINING” MAPS:
THE PERSISTENT STRUCTURE OF SEGREGATION AND ECONOMIC INEQUALITY 19 (2018),
https://ncrc.org/wp-content/uploads/dlm_uploads/2018/02/NCRC-Research-HOLC-10.pdf.
11 Elizabeth Julian & Michael M. Daniel, HUD-Assisted Low-Income Housing: Is It Working and for
Whom?, POVERTY & RACE, July–Aug. 2009, at 3, 6–7,
https://www.prrac.org/newsletters/julaug2009.pdf (finding that the “unavoidable conclusion one
comes to after reviewing the somewhat tedious data in the HUD report is that poor Black renters, as
a result of accepting HUD rental assistance, will be subjected to worse conditions or more segregated
conditions, or both”).
12 Examples of Housing Discrimination, U.S. DEP’T HOUSING & URB. DEV.,
https://www.hud.gov/program_offices/fair_housing_equal_opp/examples_housing_discrimination.
HUD’s first example of housing discrimination on its Office of Fair Housing and Equal Opportunity
website is the story of “John.” John’s landlord rejects his application on the pretense of a negative
recommendation from John’s past landlord. Following an investigation into the present landlord’s
practices, HUD finds a pattern of discrimination based on race and color.
10

4

 frequency of crime. 13 Yet because racial discrimination in housing has been so
historically prevalent, even a predictively accurate algorithm that is based on data
which accurately reflects historical conditions will reproduce racial bias and
discriminatory outcomes.
Algorithmic tools are also subject to the biases of individuals and institutions
that create and design them. 14 Individual and institutional bias can be introduced at
many different stages, including framing the problem that the algorithm is designed
to solve, choosing what metrics to optimize for, collecting and preparing the data,
developing the model that guides the performance of the tool, and deciding how to
present that information to practitioners. 15
The pervasive nature of racial bias has led researchers across a variety of
sectors including computer science, criminal justice, education, health, and
employment to conclude that algorithmic tools carry the vast potential to
disproportionately affect the lives of communities of color by perpetuating and
concealing inequality and racial discrimination. 16
Any introduction of guidelines, protocols or procedures relying on algorithmic
tools must therefore acknowledge that the undue influence of race will play a key role
in the design, implementation, and effect of algorithmic tools. Comprehensive and
established guidelines must be established to mitigate these effects, ensuring that
algorithmic tools are adequately accounted for and that communities of color are
protected.
See, e.g., Rashida Richardson et al., Dirty Data, Bad Predictions: How Civil Rights Violations
Impact Police Data, Predictive Policing Systems and Justice, 94 N.Y.U. L. REV. 193, 199–203, 218–19

13

(2019), https://www.nyulawreview.org/wp-content/uploads/2019/04/NYULawReview-94-RichardsonSchultz-Crawford.pdf (discussing the bias inherent in police data, leading to significant challenges in
predictive policing); Sarah Brayne, Big Data Surveillance: The Case of Policing, 82 AM. SOC. REV.
977 (2017), https://journals.sagepub.com/doi/10.1177/0003122417725865 (discussing the rise of big
data in surveillance and policing using a research study conducted within the Los Angeles Police
Department); Kristian Lum & William Isaac, To Predict and Serve?, SIGNIFICANCE MAG., Oct. 2016,
at 5, 14, https://rss.onlinelibrary.wiley.com/doi/epdf/10.1111/j.1740-9713.2016.00960.x (finding that
the big data analytics used in policing amplify prior surveillance practices that create greater social
inequities and consequences).
14 Danielle Keats Citron & Frank Pasquale, The Scored Society: Due Process for Automated
Predictions, 89 WASH. L. REV. 1, 4 (2014).
15 Karen Hao, This is How AI Bias Really Happens—and Why It’s So Hard To Fix, MIT TECH. REV.
(Feb. 4, 2019), https://www.technologyreview.com/s/612876/this-is-how-ai-bias-really-happensandwhy-its-so-hard-to-fix/; see also Timnit Gebru et al., Datasheets for Datasets (Apr. 14, 2019)
(working paper), https://arxiv.org/abs/1803.09010 (highlighting the lack of standards in creating and
organizing data sets); Ben Green, The Just City: Machine Learning’s Social and Political
Foundations, in THE SMART ENOUGH CITY (2019),
https://smartenoughcity.mitpress.mit.edu/pub/vmjl8djz (describing the variety of design choices that
shape the impacts of machine learning models).
16 See, e.g., VIRGINIA EUBANKS, AUTOMATING INEQUALITY: HOW HIGH-TECH TOOLS PROFILE, POLICE,
AND PUNISH THE POOR (2018); SAFIA U. NOBLE, ALGORITHMS OF OPPRESSION (2018).

5

 2. Racial Discrimination by Algorithms Already Contributes to Housing
Inequality
In housing, algorithmic tools used for mortgage underwriting (such as credit
scoring 17 and disqualification), 18 tenant screening practices, 19 and social media
advertising 20 have all been found to discriminate against communities of color. 21 One
such example can be seen in HUD’s 2016 guidance statement that an algorithm that
considered criminal records as a way to exclude rental applicants disproportionately
harmed Black and Latinx populations. 22
Accordingly, while we recognize that it is appropriate to assess individuals for
housing applications and that algorithms can play a role in this assessment,
safeguards must be established to prevent the exacerbation of racial harms and social
exclusion experienced by communities of color. Yet the proposed rule fails to
acknowledge the potential misuse of, or negative outcomes produced by, these
See David Murakami Wood, Spatial Profiling, Sorting and Prediction, in UNDERSTANDING SPATIAL
MEDIA 225, 226–32 (Rob Kitchin et al. eds., 2015).
18 Susan Saegert et al., Mortgage Foreclosure and Health Disparities: Serial Displacement as Asset
Extraction in African American Populations, 88 J. URB. HEALTH 390 (2011),
https://link.springer.com/article/10.1007/s11524-011-9584-3 (finding that mortgage foreclosure is a
way of serial displacement highlighting the current crisis in the context of historically repeated
extraction of capital). The paper finds that for Black households, this reflects structural inequality in
health and housing. The scale of displacement makes a large difference for this community.
19 See, e.g., Conn. Fair Hous. Ctr. v. CoreLogic Rental Prop. Sols., LLC, 369 F. Supp. 3d 362, 373 (D.
Conn. 2019), http://www.ctfairhousing.com/PDFs/CoreLogicMTDOrder.pdf (finding that the lack of
an individualized review of a Latinx disabled man’s suitability for tenancy prevented him from
obtaining housing, relying heavily upon the automated algorithm that determined him “disqualified”
for housing).
20 Julia Angwin & Terry Parris Jr., Facebook Lets Advertisers Exclude Users by Race, PROPUBLICA
(Oct. 28, 2016, 1:00 PM), https://www.propublica.org/article/facebook-lets-advertisers-exclude-usersby-race.
21 While this comment centers on race, other algorithms used in the housing context have been found
to discriminate based on other social characteristics. See e.g., Marie C. Baca, Housing Companies
17

Used Facebook’s Ad System to Discriminate Against Older People, According to New Human Rights
Complaints, WASH. POST (Sept. 18, 2019),

https://www.washingtonpost.com/technology/2019/09/18/housing-companies-used-facebooks-adsystem-discriminate-against-older-people-according-new-human-rights-charges/ (discussing
discrimination based on age); HUD Files Housing Discrimination Complaint Against Facebook,
Press Release, HUD No. 18-085 (Aug. 17, 2018),
https://www.hud.gov/press/press_releases_media_advisories/HUD_No_18_085 (discussing HUD’s
complaint against Facebook regarding ads that discriminated on the basis of characteristics that
include gender, disability, religion, and zip codes); Muhammed Ali et al, Discrimination Through
Optimization: How Facebook's Ad Delivery Can Lead to Skewed Outcomes, 2019 PROC. ACM ON
HUM.-COMPUTER INTERACTION (forthcoming), https://arxiv.org/abs/1904.02095 (finding that
algorithms used by Facebook resulted in housing ads that align with race and gender stereotypes).
22 OFFICE OF GEN. COUNSEL, U.S. DEP’T OF HOUS. & URBAN DEV., GUIDANCE ON APPLICATION OF FAIR
HOUSING ACT STANDARDS TO THE USE OF CRIMINAL RECORDS BY PROVIDERS OF HOUSING AND REAL
ESTATE-RELATED TRANSACTIONS 2 (2016),
https://www.hud.gov/sites/documents/HUD_OGCGUIDAPPFHASTANDCR.PDF.

6

 algorithmic tools or provide substantial safeguards or protections against these
harms. In fact, the proposed rule gives unprecedented deference to mortgage lenders,
landlords, banks, insurance companies, and others in the housing industry, placing
potential tenants and homebuyers of color at a severe disadvantage.
3. The Proposed Rule Will Encourage, Facilitate, and Exacerbate Racial

Discrimination by Algorithms in Housing

The proposed rule encourages and facilitates discrimination by providing those
in the housing industry with new ways to discriminate and conceal discriminatory
motives with the protection of the law, and it thus dramatically reduces their
responsibility for discriminatory outcomes. 23 In addition, because of the
discriminatory nature of housing algorithms, the rule will enable discrimination even
in cases where housing industry actors had no intention to discriminate. When a
plaintiff alleges that the cause of discrimination is attributable to an algorithmic tool,
the proposed rule creates three defenses that absolve the defendant from
responsibility. The three defenses against claims of discrimination by algorithm
provides such parties with a feasible pathway to both intentionally and
unintentionally discriminate with little accountability. As detailed below, the
defenses themselves are therefore inherently flawed, making it more difficult for
plaintiffs to successfully challenge racial discrimination and exacerbating racial
inequality in housing. 24

The First Defense
The first defense allows parties in the housing industry to rely on algorithmic
tools provided the inputs “do not rely in any material part on factors that are
substitutes or close proxies for protected classes under the FHA.” 25 However, there
are significant challenges relating to the use of proxies in algorithmic tools.
The emphasis upon substitutes or close proxies suggests that there is a
hierarchy of relevant proxies for protected classes. In fact, the scope and hierarchy of
close and relevant proxies is impossible to define in an algorithmic tool given the
Emily Badger, Who’s to Blame When Algorithms Discriminate? , N.Y. TIMES, (Aug. 20, 2019),
https://www.nytimes.com/2019/08/20/upshot/housing-discrimination-algorithms-hud.html; Andrew
D. Selbst, A New HUD Rule Would Effectively Encourage Discrimination by Algorithm, SLATE (Aug.
19, 2019), https://slate.com/technology/2019/08/hud-disparate-impact-discrimination-algorithm.html;
Olatunde Johnson & Michelle Aronowitz, The Trump Administration’s Assault on Fair Housing,
TAKE CARE BLOG (Aug. 19, 2019), https://takecareblog.com/blog/the-trump-administration-s-assaulton-fair-housing.
24 See, e.g., Alistair Croll, Big Data Is Our Generation’s Civil Rights Issue, and We Don’t Know It,
SOLVE FOR INTERESTING (Jul. 31, 2012), https://perma.cc/BS8S-6T7S; see also Solon Barocas &
Andrew D. Selbst, Big Data’s Disparate Impact, 104 CAL. L. REV. 671 (2016),
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2477899.
25 HUD’s Implementation of the Fair Housing Act’s Disparate Impact Standard, 84 Fed. Reg. 42,854,
42,862 (proposed Aug. 19, 2019).
23

7

 social construction and changing nature of many protected classes. 26 As noted by
computer and information scientists, “how we measure race often changes how we
conceive of it, and changing conceptions of race may force us to alter what we
measure.” 27 In the context of race, proxies that were once considered far removed
from race may now be construed as racial in nature. As a practical matter, the cost of
proving that a given factor is a proxy may make prosecuting many legitimate
discrimination cases economically infeasible.
In a society as racially stratified as ours, many things seemingly unrelated to
race can act as proxies. Machine learning algorithms “learn” from historical data to
identify patterns of related attributes or activities that can collectively serve as
proxies for protected groups. 28 For example, the well-known history of racially
segregated housing would suggest that neighborhoods and zip codes could reasonably
be considered as close proxies for race. 29 Yet the power of machine learning lies not
just in finding salient connections, but in “produc[ing] novel insights that probably
couldn't have been revealed in any other way.” 30 This means that many factors that
seem unrelated to race—such as musical taste, the number of “likes” on a social
network such as Facebook, and a network of friends 31—can act as proxies for race.
The challenge is that there is no clear way of knowing how “close” an attribute is in
order for it to be considered “a close proxy or substitute” for a protected group. 32 Even
attributes that appear themselves to be unrelated to protected categories can, when
combined with other attributes, collectively be close proxies for protected categories.
Whether an attribute serves as a proxy for a protected category depends in part on
factors such as what dataset is being used and what other attributes are included. It
is therefore impossible to identify whether a specific attribute is a close proxy. Almost
all algorithmic patterns are infused with data that could be construed as substitutes
or proxies for race. But the proposed rule’s limited focus on “close proxies” ignores this
essential fact.
See e.g., Ian Haney Lopez, The Social Construction of Race, 29 HARV. C.R.-C.L. L. REV. 1 (1994),
https://scholarship.law.berkeley.edu/cgi/viewcontent.cgi?article=2815&context=facpubs (explaining
that the law plays a significant role in carving out notions of race often reflecting, or working in
tandem with societal norms and culture; by relying on jurisprudence, Lopez notes that throughout
history, the notion of race as it pertains to different groups has changed); Angela Onwuachi-Willig,
Race and Racial Identity Are Social Constructs, N.Y. TIMES (Sept. 6, 2016),
https://www.nytimes.com/roomfordebate/2015/06/16/how-fluid-is-racial-identity/race-and-racialidentity-are-social-constructs (explaining that while the concept of race is socially constructed and
fluid, the social, political and economic meaning attached to racial groups has remained the same).
27 Solon Barocas et al., Introduction, in FAIRNESS IN IN MACHINE LEARNING (2019),
https://fairmlbook.org/pdf/introduction.pdf.
28 Pedro Domingos, A Few Useful Things to Know About Machine Learning, COMM. ACM, Oct. 2012,
at 78, 78–80, https://homes.cs.washington.edu/~pedrod/papers/cacm12.pdf.
29 The Justice map demonstrates that zip code and census data combined does track race. JUSTICE
MAP, http://www.justicemap.org/ (last visited Oct. 16, 2019).
30 VIKTOR MAYER-SCHON
̈ BERGER & KENNETH CUKIER, BIG DATA: A REVOLUTION THAT WILL
TRANSFORM HOW WE LIVE, WORK, AND THINK (2013).
31 See, e.g., Barocas, supra note 24, at 712.
32 Id. at 720.
26

8

 The first defense also allows those in the housing industry to rely on an
algorithmic tool provided that the model “is predictive of credit risk or other similar
valid objective.” 33 But even a tool that is predictive can still be discriminatory. Indeed,
given the deeply racialized nature of housing in the United States, 34 many predictions
that accurately forecast risk will do so in part or whole by using attributes that
identify or are correlated with the race of the subject. In the context of housing, it
would be almost impossible for a model to make accurate predictions without making
decisions based on race.
Moreover, while algorithms rely upon a range of patterns in order to make
predictions, what constitutes reasonable predictive accuracy varies greatly according
to context and involves tradeoffs. 35 A comparative example illustrates this point:
algorithmic tools used to classify radiographs can have a predictive accuracy measure
of 0.98, 36 while recidivism prediction models tend to have a predictive accuracy
measure of approximately 0.7. 37 Given the context-specific nature of “predictive
accuracy,” it is difficult to determine by what standard or means a model could be
deemed sufficiently predictive of credit risk. This ambiguity would again make it
exceedingly difficult for plaintiffs to succeed against this defense. It would likewise
make it easy for those engaging in housing discrimination to escape liability.

The Second Defense
The second defense permits those in the housing industry to rely on algorithms
that may have a discriminatory impact, provided they were developed by “a
recognized third-party that determines industry standards.” 38 This defense raises
serious concerns given the lack of transparency surrounding the use of third-party
vendors and the lack of uniformity in industry standard setting.
HUD’s Implementation of the Fair Housing Act’s Disparate Impact Standard, 84 Fed. Reg. 42,854,
42,862 (proposed Aug. 19, 2019).
34 See supra Section 1.
35 Sam Corbett-Davies et al., Algorithmic Decision Making and the Cost of Fairness, 23 PROC. ACM
SIGKDD INT’L CONF. KNOWLEDGE DISCOVERY & DATA MINING 797 (2017),
https://dl.acm.org/citation.cfm?id=3098095/ (finding that accuracy and fairness of a machine learning
model can be in tension).
36 Jared A. Dunnmon et al., Assessment of Convolutional Neural Networks for Automated
Classification of Chest Radiographs, 2018 RADIOLOGY 1, 1 (2018),
https://jdunnmon.github.io/dunnmon_radiology_2018.pdf.
37 See SARAH L. DESMARAIS & JAY P. SINGH, RISK ASSESSMENT INSTRUMENTS VALIDATED AND
IMPLEMENTED IN CORRECTIONAL SETTINGS IN THE UNITED STATES (2013),
https://csgjusticecenter.org/wp-content/uploads/2014/07/Risk-Assessment-Instruments-Validatedand-Implemented-in-Correctional-Settings-in-the-United-States.pdf; see also NORTHPOINTE,
PRACTITIONERS GUIDE TO COMPAS (2012),
http://www.northpointeinc.com/files/technical_documents/FieldGuide2_081412.pdf.
38 HUD’s Implementation of the Fair Housing Act’s Disparate Impact Standard, 84 Fed. Reg. at
42,862.
33

9

 The onus-shifting approach to third-party vendors is troubling. It provides
those in the housing industry with little incentive to evaluate the potential
discriminatory effects of the algorithmic tools that they are using. Some may argue
that this defense follows a similar approach to §230 of the Communications Decency
Act, which provides legal protection for content produced by online providers who use
third-party intermediaries to host or republish speech online. Yet even the broad
protection offered under §230 exhorts online providers to monitor and eliminate the
worst of that content and does not absolve online providers from complete
responsibility. 39 The proposed HUD defense fails to incorporate this type of provision
and enables those in the housing industry to shift their responsibilities onto thirdparty vendors.
This defense is particularly troubling given the opaque nature of algorithmic
tools and the challenges to transparency and accountability that accompany them.
For instance, the Center and AI Now, in 2018 and 2019 Litigating Algorithms
Reports, highlighted legal challenges to algorithmic tools used in education, public
benefits, and criminal justice, where third party vendors used broad, and ultimately
illegitimate, trade secrecy or confidentiality claims to obstruct efforts to examine the
algorithm. 40 Due to trade secrecy laws and non-disclosure agreements, third-party
developers are generally free to design algorithmic tools within a “black-box.” 41
Details pertaining to how algorithmic tools are designed and created are therefore
inaccessible to those who may wish to challenge the use of an algorithmic tool. 42 This
makes any effort to hold third parties accountable extremely challenging.
Researchers and scholars have already noted the tension between trade secrecy and
transparency, and have called for mechanisms to be put in place to make these
processes more transparent and accountable. 43
Joshua A. Geltzer, The President and Congress Are Thinking of Changing This Important Internet
Law, SLATE (Feb. 25, 2019), https://slate.com/technology/2019/02/cda-section-230-trump39

congress.html.
40 RASHIDA RICHARDSON ET AL., AI NOW INST. & CENT. ON RACE, INEQUALITY, & THE LAW, LITIGATING
ALGORITHMS 2019 US REPORT: NEW CHALLENGES TO GOVERNMENT USE OF ALGORITHMIC DECISION
SYSTEMS (2019), https://ainowinstitute.org/litigatingalgorithms-2019-us.pdf; MEREDITH WHITTAKER
ET AL., AI NOW INST., AI NOW REPORT 2018, at 39–40 (2018),
https://ainowinstitute.org/AI_Now_2018_Report.pdf.
41 See, e.g., FRANK PASQUALE, THE BLACK BOX SOCIETY: THE SECRET ALGORITHMS THAT CONTROL
MONEY AND INFORMATION 12–15 (2015).
42 Robert Brauneis & Ellen P. Goodman, Algorithmic Transparency for the Smart City, 20 Yale J.L.
& Tech. 103 (2018), https://yjolt.org/sites/default/files/20_yale_j._l._tech._103.pdf; Rebecca Wexler,
Life, Liberty, and Trade Secrets: Intellectual Property in the Criminal Justice System, 70 STAN. L.
REV. 1343 (2018), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2920883 (highlighting the
lack of meaningful transparency regarding government use of algorithmic systems and how the
aggressive use of trade secrecy and confidentiality serves as an obstacle).
43 See, e.g., Joshua A. Kroll et al., Accountable Algorithms, 165 U. PA. L. REV. 633, 636 (2017)
(observing that “accountability mechanisms and legal standards that govern decision processes have
not kept pace with technology” and arguing that “[c]itizens, and society as a whole, have an interest
in making these processes more accountable).

10

 Yet HUD’s proposed defense steps away from these recommendations and
encourages an approach that increases third-party opacity. As we have seen in other
contexts, such as policing, the lack of third-party transparency can lead to third-party
vendors having an undue influence on direct service providers. 44 In the housing
context, this could have drastic consequences for communities that are already
marginalized and subject to racial discrimination.
The reference to industry standards raises further cause for concern. There is
no criteria that determines industry standards for standard-setting boards to follow
when regulating algorithmic models. Even within the research community, there are
ongoing debates about what standards or metrics should govern such models. This
challenge, along with the highly politicized nature of industry standard setting, has
been observed in the context of newly created Wifi connectivity standards, 45 privacy
certifications, 46 and DRM-protected video. 47 The result is that third-party developers
are free to design and implement algorithmic tools with no requirements to test the
data sets or the algorithms that are used, no regulation, no oversight, and no clear
standards by which to test the models against.
HUD has provided no guidance regarding the industry standards required of
third-party developers. Yet even if they had, there is no current method or standard
in place that would ensure that algorithmic tools are “discrimination proof.” 48 As
noted in the examples above, industry standard creation is an inherently political
process that will likely undermine the societal goal of reducing discrimination in
housing. Even the most advanced systems still pose a risk of amplifying and
perpetuating structural or systemic bias. Unlike the proposed rule suggests, it is
See Elizabeth E. Joh, The Undue Influence of Surveillance Technology Companies on Policing, 92
N.Y.U. L. REV. ONLINE 101 (2017), https://www.nyulawreview.org/wp-content/uploads/2018/08/JohFINAL_0.pdf; William Alden, There's a Fight Brewing Between the NYPD and Silicon Valley's
Palantir, BUZZFEED NEWS (Jun. 28, 2017),
https://www.buzzfeednews.com/article/williamalden/theres-a-fight-brewing-between-the-nypd-andsilicon-valley (demonstrating the power vendors can have).
45 Joshua Sisco & Leah Nylen, DOJ Probes Role of Special Interest Group in New WiFi Standard,
MARKET INSIGHT (Jan. 26, 2018), https://mlexmarketinsight.com/insights-center/editorspicks/antitrust/north-america/doj-probes-role-of-special-interest-group-in-new-wifi-standard.
46 TRUSTe Settles FTC Charges It Deceived Consumers Through Its Privacy Seal Program, FED.
TRADE COMMISSION (Nov. 17, 2014), https://www.ftc.gov/news-events/press-releases/2014/11/trustesettles-ftc-charges-it-deceived-consumers-through-its.
47 Jacob Kastrenakes, A DRM Standard Has Been Approved for the Web, and Security Researchers
are Worried, VERGE (Jul. 8, 2017), https://www.theverge.com/2017/7/8/15942238/web-drm-standardeme-approved-controversy.
48 Sandra G. Mayson, Bias In, Bias Out, 128 YALE L.J. 2218, 2230–51 (2019),
https://www.yalelawjournal.org/pdf/Mayson_p5g2tz2m.pdf (describing how attempts to standardize
racial equality are hampered by the lack of consensus on definitions and adequate technical
approaches to balance the tradeoffs of varying notions of equality in the criminal justice context); Jon
Kleinberg et al., Inherent Trade-Offs in the Fair Determination of Risk Scores, 8 PROC. INNOVATIONS
THEORETICAL COMPUTER SCI. (2017), https://arxiv.org/pdf/1609.05807.pdf (describing how different
notions of “fairness” in algorithms are inherently in conflict with one another).
44

11

 unclear what industry is best suited to create standards that could mitigate
discrimination.

The Third Defense
The third defense permits those in the housing industry to rely upon a neutral
third-party expert to analyze the algorithm and determine that the inputs are not
substitutes for protected characteristics. The defense seems to suggest that a thirdparty expert could determine whether the inputs for the algorithmic tool were
intended to be discriminatory. As described above, however, it is unclear and likely
impossible to determine how this could be proven. Research on data privacy has
shown that “one can never know what information may later be extracted from any
particular collection of big data”; 49 by the same token, it is impossible to determine
that any piece of information could not (either now or in the future) be connected to
race and used to create discriminatory models.
While biased tools can sometimes be identified by following the discriminatory
effects of an algorithmic model, it is less clear what the absence of bias looks like in
the design process and what could constitute a “fair” algorithmic tool. 50 Indeed, there
are myriad definitions of what it means for a machine learning model to be “fair,” 51
and those definitions are often in tension with one another. 52 While a third-party
expert may claim that an algorithmic model is fair, there are no approved standards
that state what constitutes “fair” and whether the determination of fairness aligns
with societal and constitutional notions of equality. In fact, researchers have shown
that common notions of “fairness” are in tension with broader notions of justice and
equality. 53 The term “fairness” may also differ for third-party experts when compared
to the impact that it has on communities. Finally, there is no way to determine the
credibility of experts. No standards currently exist to assess which experts could
confirm that tools are “fair” and will produce racially equitable outcomes.
HUD claims that these defenses “are not intended to provide a special
exemption for parties who use algorithmic models, but merely to recognize that
additional guidance is necessary in response to the complexity of disparate impact
PRESIDENT’S COUNCIL OF ADVISORS ON SCI. & TECH., BIG DATA AND PRIVACY: A TECHNOLOGICAL
PERSPECTIVE ix (2014), https://bigdatawg.nist.gov/pdf/pcast_big_data_and_privacy_-_may_2014.pdf.
50 Hao, supra note 15.
51 Arvind Narayanan, Tutorial: 21 Fairness Definitions and Their Politics, YOUTUBE (Mar. 1, 2018),
https://www.youtube.com/watch?v=jIXIuYdnyyk.
52 Kleinberg, supra note 48.
53 Anna Lauren Hoffmann, Where Fairness Fails: Data, Algorithms, and the Limits of
Antidiscrimination Discourse, 7 J. INFO. COMM. & SOC. 22 (2019),
https://www.tandfonline.com/doi/full/10.1080/1369118X.2019.1573912; Ben Green & Lily Hu, The
Myth in the Methodology: Towards a Recontextualization of Fairness in Machine Learning, 35 INT’L
CONF. ON MACHINE LEARNING (2018), https://www.benzevgreen.com/wp-content/uploads/2019/02/18icmldebates.pdf.
49

12

 cases challenging these models.” 54 However, in practice, the proposed rule provides
just such an exemption, by imposing a burden on plaintiffs to interrogate opaque
systems and to prove what algorithmic fairness means, in the absence of any standard
to do so.

An Onerous and Unreasonable Burden upon Plaintiffs
The broad defenses that the proposed rule would grant defendants are
complemented by the onerous burden placed on plaintiffs. Under the proposed rule’s
five point prima facie evidentiary test, plaintiffs would need to essentially establish
the elements of their claim at the outset of the case, prior to any discovery process.
This heavy burden, which will disqualify many meritorious claims, is
particularly ill-suited to cases involving algorithmic tools. As discussed, the internal
workings of algorithmic tools are frequently hidden within a “black box” that prevents
outsider analysis under ordinary circumstances. 55 For example, in Houston
Federation of Teachers, Local 2415 v. Houston Independent School District, the court
found that open discovery was essential to understanding the inner workings of an
algorithm established to improve teaching quality by providing a standardized
assessment of teachers. 56 The plaintiffs were successful on due process grounds, with
the court noting that without open discovery, it was almost impossible for teachers to
access, understand, or act on their own evaluations. 57 Without access to discovery, it
is highly implausible that a victim of a racially discriminatory algorithmic tool will
possess the information to capably contest the validity of an algorithmic tool’s inputs
or to otherwise discuss the tool with the specificity that the proposed rule would
require.
III. By Enabling Racial Discrimination, the Proposed Rule Will Disadvantage
Communities of Color Even Further in the Housing Market

1. The Housing Market Is Already Rife with Racial Discrimination
The challenges that algorithmic tools pose and the broad defenses that the
proposed regulation offers to defendants will make it substantially more difficult for
plaintiffs with meritorious claims to obtain relief. This is especially troubling because
even with existing protections, the realm of housing is plagued with racial
discrimination and inequality.

HUD’s Implementation of the Fair Housing Act’s Disparate Impact Standard, 84 Fed. Reg. 42,854,
42,859 (proposed Aug. 19, 2019).
55 See Pasquale, supra note 41 at 12–15.
56 See Houston Fed’n of Teachers, Local 2415 v. Houston Indep. Sch. Dist., 51 F. Supp. 3d 1168, 1179
(S.D. Tex. 2017); see also RASHIDA RICHARDSON ET AL., supra note 40.
57 Houston Fed’n of Teachers, 51 F. Supp. 3d at 1176–80.
54

13

 This discrimination and the racial inequality that results manifests in a
variety of forms. A mortgage loan is often a requirement for buying a house. Yet Black
and Latinx borrowers face unique obstacles in acquiring such loans. A recent lawsuit
against Wells Fargo, for example, revealed that Black borrowers who qualified for
regular loans were 2.9 times more likely than similarly situated White borrowers to
be directed to a subprime loan and Latinx borrowers 1.8 times more likely, thus
encumbering Black and Latinx borrowers with higher interest rates and
unconscionable terms. Further data indicated that in Chicago, relative to similarly
situated White customers, customers who borrowed $300,000 through an
independent broker paid an average of $2,937 more in broker fees if Black and $2,187
if Latinx. 58
Wells Fargo is not alone in these types of practices, 59 nor are banks the only
actors at fault. Courts have recognized that there is a “direct connection of availability
of property insurance and ability to purchase a house.” 60 Yet property insurers have
overcharged or denied coverage to residents of Black neighborhoods, further
restricting free and fair access to the housing market. 61
Buyers and renters of color face additional discrimination in the housing
market even when they can acquire adequate funds. A 2012 HUD-sponsored study
found that homeseekers of color are shown and told about fewer homes than White
homeseekers. 62 And homeseekers of color who get beyond this stage must then
Charlie Savage, Wells Fargo Will Settle Mortgage Bias Charges, N.Y. TIMES (July 12, 2012),
https://www.nytimes.com/2012/07/13/business/wells-fargo-to-settle-mortgage-discriminationcharges.html; see also Robert Barlett et al., Consumer-Lending Discrimination in the FinTech Era 4
(Nat’l Bureau of Econ. Research, Working Paper No. 25943, 2019),
https://www.nber.org/papers/w25943.pdf (finding that that “accepted Latinx and African-American
borrowers pay 7.9 and 3.6 basis points more in interest for home purchase and refinance mortgages
respectively because of discrimination”).
59 See, e.g., Richard Rothstein, A Comment on Bank of America/Countrywide’s Discriminatory
Mortgage Lending and Its Implications for Racial Segregation, ECON. POL. INS. (Jan. 23, 2012),
https://www.epi.org/publication/bp335-boa-countrywide-discriminatory-lending; see also EMMANUEL
MARTINEZ & AARON GLANTZ, CENTER FOR INVESTIGATIVE REPORTING, HOW REVEAL IDENTIFIED
LENDING DISPARITIES IN FEDERAL MORTGAGE DATA (2018), https://s3-us-west2.amazonaws.com/revealnews.org/uploads/lending_disparities_whitepaper_180214.pdf (finding based
on 2015 and 2016 records that “in certain areas of the country, people of color were more likely to be
denied a conventional mortgage than white applicants, even after controlling for a wide array of
economic and social factors”).
60 Nationwide Mut. Ins. Co. v. Cisneros, 52 F.3d 1351, 1359 (6th Cir. 1995).
61 See Stephen Koff, HUD Tries to Crack Down on Discrimination by Insurers, While Insurers Deny
They're Discriminating, CLEVELAND.COM (Oct. 6, 2016),
https://www.cleveland.com/metro/2016/10/hud_to_crack_down_on_discrimin.html (reporting on
HUD’s recognition of this issue); see also John H. Gilmore, Insurance Redlining & the Fair Housing
Act: The Lost Opportunity of Mackey v. Nationwide Insurance Companies, 34 CATH. U. L. REV. 563,
575–78 (1985) (discussing how this practice has functioned).
62 MARGERY AUSTIN TURNER ET AL., URBAN INST., HOUSING DISCRIMINATION AGAINST RACIAL AND
ETHNIC MINORITIES 2012, at xi (2013), https://www.huduser.gov/portal/Publications/pdf/HUD514_HDS2012.pdf.
58

14

 contend with racial discrimination on the part of sellers and renters. 63 In addition to
explicit racial discrimination, sellers discriminate through the use of nominally raceneutral screening factors that apply in a highly racialized manner. 64 HUD has
recognized this problem and the importance of adequate redress. 65
For Black homeowners, the ultimate acquisition of a house does not signal an
end to racial discrimination in housing. A recent Brookings Institution study found
that homes in neighborhoods where the residents are 50 percent Black are valued at
about half the price of homes in neighborhoods without any Black residents.
Considering whether this could be explained by non-racial factors, the study
concluded that “we are left with the fact that a square foot of residential real-estate
is worth 23 percent less in neighborhoods where half the population is black compared
to neighborhood with few or no black residents, even after adjusting for housing
quality and neighborhood quality.” This devaluation of Black homes has resulted in
a cumulative loss of $156 billion. 66 Additionally, homeowners of color are
disproportionately subjected to improper enforcement of local laws and regulations
that can result in illegal seizures and foreclosures, undermining the opportunities for
building wealth that homeownership offers. 67
See, e.g., B. Rose Kelly, Hispanics Face Racial Discrimination in New York City’s Rental Housing
Market, PRINCETON U. (Oct. 24, 2018, 4:34 PM),

63

https://www.princeton.edu/news/2018/10/24/hispanics-face-racial-discrimination-new-york-citysrental-housing-market.
64 See, e.g., Esme Caramello & Nora Mahlberg, Combating Tenant Blacklisting Based on Housing
Court Records: A Survey of Approaches, SARGENT SHRIVER NAT’L CENTER ON POVERTY L. (Sept.
2017), https://www.povertylaw.org/clearinghouse/article/blacklisting (discussing one “fair housing
challenge to a landlord’s blanket policy of rejecting or adversely treating any tenant with a record of
a housing court case,” despite the fact that “in the county of the lawsuit, African Americans are
almost four times more likely than whites to have been sued in an eviction case, and African
American women are sued more than five times as often as households headed by white men”);
Jennifer Safstrom & Rachel Goodman, Lawsuit Challenges Discriminatory Housing Policy in
Chesterfield County, Virginia, ACLU (June 4, 2019, 12:00 PM), https://www.aclu.org/blog/racialjustice/race-and-economic-justice/lawsuit-challenges-discriminatory-housing-policy (discussing a
challenge to a landlord’s blanket ban on tenants with felony convictions, even though “individuals
who are Black represented 46% of those convicted of a felony between 2007 and 2017 [in the county],
despite only accounting for 22% of the population”).
65 See OFFICE OF GEN. COUNSEL, supra note 22 (noting that since “African Americans and Hispanics
are arrested, convicted and incarcerated at rates disproportionate to their share of the general
population. . . . [C]riminal history-based restrictions on housing opportunities violate the Act if,
without justification, their burden falls more often on renters or other housing market participants of
one race or national origin over another”).
66 ANDRE M. PERRY, JONATHAN ROTHWELL & DAVID HARSHBARGER, THE DEVALUATION OF ASSETS IN
BLACK NEIGHBORHOODS: THE CASE OF RESIDENTIAL PROPERTY 11–15 (2018),
https://www.brookings.edu/wp-content/uploads/2018/11/2018.11_Brookings-Metro_DevaluationAssets-Black-Neighborhoods_final.pdf.
67 See, e.g., Mary Frost, Brooklyn Officials Demand Full-Scale Investigation of Home Theft in Black
& Brown Neighborhoods, BROOKLYN DAILY EAGLE (Nov. 26, 2018),
https://brooklyneagle.com/articles/2018/11/26/brooklyn-officials-demand-full-scale-investigation-ofhome-theft-in-black-brown-neighborhoods/; Stephen Witt, City Caught Trying To Grab Senior
Citizen’s Brownstone, KINGS COUNTY POL. (Sept. 17, 2018),

15

 2. Racial Discrimination in Housing Has Substantial Collateral Effects
Racial discrimination in housing is not only a profound wrong in itself, but one
that substantially affects other areas of life. Wealth accumulation, involvement with
the criminal legal system, education, and health are all shaped by housing
discrimination. For most American families, a house is one of their most valuable
assets. However, homeownership rates and typical home equity vary greatly along
racial lines, with Black and Latinx individuals sharply disadvantaged. 68 The
encumbrance of homeseekers of color with excessive mortgage interest rates, property
devaluation, and other discriminatory measures thus contributes significantly to the
growing racial wealth gap in the United States. 69
This spillover effect of housing discrimination can also be seen in the criminal
legal system. In Boston, for example, one study found that, controlling for crime and
other non-racial factors, the number of police-civilian interactions in a neighborhood
was driven by the concentration of Black residents. 70 Similarly, statistical data
submitted in New York’s stop-and-frisk cases showed that the numbers of stops per
crime were highest in areas with the highest concentration of Black residents. 71 The
racially concentrated state of America’s neighborhoods—an inevitable result of
racially discriminatory housing practices—can only encourage this brand of raciallytargeted law enforcement.
Racial segregation in housing also engenders racial segregation in schooling.
Since most school districts rely on residence-based school assignment and a large
portion of school district funding is derived from local property taxes, housing
discrimination is also a primary driver of the extreme racial segregation and resource
https://www.kingscountypolitics.com/1217-dean-street/; Stephen Witt & Kelly Mena, City Takes
Property From Working Class Latinos, KINGS COUNTY POL. (Oct. 9, 2018),
https://www.kingscountypolitics.com/city-takes-property-from-working-class-latinos/.
68 LAURA SULLIVAN ET AL., INSTITUTE FOR ASSETS & SOCIAL POLICY, BRANDEIS UNIVERSITY & DEMOS,
THE RACIAL WEALTH GAP: WHY POLICY MATTERS 9 (2015),
https://www.demos.org/sites/default/files/publications/RacialWealthGap_1.pdf (finding that “73
percent of whites as compared to 47 percent of Latinos and 45 percent of Blacks” own homes, and
that typical home equity is “$86,800 for white homeowners at the median as compared to $50,000 for
Black homeowners and $48,000 for Latino homeowners).
69 See id. at 13 (finding that “[e]qualizing wealth returns to homeownership raised wealth among
Black and Latino families while white wealth was held constant, significantly reducing the racial
wealth gap” and that “[e]qualizing the returns to homeownership reduces the wealth gap between
white and Black families by” 16 percent and between White and Latinx families by 41 percent)
70 ACLU, A REPORT ON BOSTON POLICE DEPARTMENT STREET ENCOUNTERS FROM 2007–2010, at 1
(2014), https://www.aclum.org/sites/default/files/wp-content/uploads/2015/06/reports-black-brownand-targeted.pdf.
71 Second Supplemental Report of Jeffrey Fagan, Ph.D. at 13, Floyd v. City of New York, 959 F.
Supp. 2d 540 (S.D.N.Y. 2013),
https://ccrjustice.org/sites/default/files/assets/files/FaganSecondSupplementalReport.pdf.

16

 disparities in public schools. 72 For example, a recent Washington Post investigation
found that children throughout the country “remain locked in deeply segregated
districts,” and that Black students—also more likely to grow up in impoverished
neighborhoods 73—are especially likely to be enrolled in segregated districts. 74
Housing discrimination even endangers the health of its victims. Even at
higher income levels, families of color are still subjected to racial discrimination and
thus more likely than similarly situated White families to live in high-poverty
neighborhoods. Such neighborhoods present greater risks to children, including
exposure to lead and vermin, fewer safe spaces for play, and more limited access to
nutritious food. 75 As HUD has written, “for people residing in neighborhoods of
concentrated poverty, a number of neighborhood level indicators are linked to
important outcomes . . . . [E]ducation, psychological distress, and various health
problems, among many other issues, are affected by neighborhood characteristics.” 76
And when Black homeowners face foreclosure, as mortgage discrimination makes
more likely, it brings with it a host of dangers both to the health of individuals and
to the social cohesiveness of the community. 77
See EDBUILD, $23 BILLION 2–5 (2019), https://edbuild.org/content/23-billion/full-report.pdf
(analyzing data on this relationship).
73 Sociologist Patrick Sharkey has found that “young African Americans (from 13 to 28 years old) are
now ten times as likely to live in poor neighborhoods . . . as young whites.” Richard Rothstein, The
72

Racial Achievement Gap, Segregated Schools, and Segregated Neighborhoods – A Constitutional
Insult, ECON. POL’Y INST. (Nov. 12, 2014), https://www.epi.org/publication/the-racial-achievement-

gap-segregated-schools-and-segregated-neighborhoods-a-constitutional-insult/.
74 Laura Meckler & Kate Rabinowitz, The Changing Face of School Integration, WASH. POST (Sept.
12, 2019), https://www.washingtonpost.com/education/2019/09/12/more-students-are-going-schoolwith-children-different-races-schools-big-cities-remain-deeply-segregated/?arc404=true; see also
GROVER J. WHITEHURST ET AL, BROOKINGS INST., BALANCING ACT: SCHOOLS, NEIGHBORHOOD AND
RACIAL IMBALANCE 14–19 (2017), https://www.brookings.edu/wpcontent/uploads/2017/11/es_20171120_schoolsegregation.pdf (analyzing the relationship between
racial imbalances in schooling and racially segregated housing); Erica Frankenberg, The Role of
Residential Segregation in Contemporary School Segregation, 45 EDUC. & URB. SOC’Y 548, 549–56,
https://journals.sagepub.com/doi/pdf/10.1177/0013124513486288 (discussing data on the relationship
between housing and school segregation); Adam Harris, The Whiter, Richer School District Right
Next Door, ATLANTIC (Aug 1, 2019),
https://www.theatlantic.com/education/archive/2019/08/segregated-school-districts-trapped-theirborders/595318 (discussing how the composition of segregated schools reflects the surrounding
neighborhoods).
75 Brian Smedley & Rachel A. Davis, The Effects of Housing Discrimination on Health Can
Reverberate for Decades, HILL (Aug. 27, 2019, 8:00 AM), https://thehill.com/opinion/civilrights/458899-the-effects-of-housing-discrimination-on-health-can-reverberate-for.
76 Understanding Neighborhood Effects of Concentrated Poverty, EVIDENCE MATTERS, Winter 2011,
https://www.huduser.gov/portal/periodicals/em/winter11/highlight2.html.
77 See Susan Saegert, Desiree Fields & Kimberly Libman, Mortgage Foreclosure and Health
Disparities, Serial Displacement as Asset Extraction in African American Populations, 88 J. URB.
HEALTH 390 (2011), https://link.springer.com/content/pdf/10.1007%2Fs11524-011-9584-3.pdf
(discussing how “[t]he health impacts of serial displacement for individuals, social groups, and
neighborhoods . . . contribut[e] to poorer health for individuals while maintaining and exacerbating

17

 More protection is needed for buyers and renters of color—not less. This
includes protection from algorithmic tools, which have contributed to racial
discrimination even without the broad defenses that the proposed rule would create. 78
By weakening existing protections in general and with regard to algorithmic tools
specifically, the proposed rule will only exacerbate racial discrimination in housing,
with devastating effects for communities of color.
Conclusion
Racial discrimination in housing has a long, disturbing history. It was a
predominant practice and policy of the Jim Crow era and a major focus of the civil
rights movement, exemplified by Dr. Martin Luther King, Jr.’s 1966 campaign in
Chicago for fair housing. In enacting the FHA, the federal government signaled its
will to combat this manifestation of racism. Although racial discrimination continues
to be all too prevalent in the realm of housing, the FHA has served as an important
legal tool in challenging it.
If HUD adopts this rule, it will dramatically undermine the FHA, enable
racially discriminatory housing practices, and further disadvantage communities of
color more generally. At a time when racial discrimination increasingly manifests
through the use of algorithmic tools alongside explicit, intentional racially
discriminatory action, the animating spirit of the FHA demands that HUD reject a
proposed rule that will erect impossibly high hurdles for countless victims of racial
discrimination. We therefore urge HUD to withdraw the proposed rule.
Sincerely,

Deborah N. Archer

Associate Professor of Clinical Law, New York University, School of Law; and Faculty
Co-Director, Center on Race, Inequality, and the Law
Ruha Benjamin

Associate Professor of African Americans Studies, Princeton University
Meredith Broussard

Associate Professor, New York University
existing health disparities between African Americans and whites, socially and spatially,” as well as
how foreclosures cause neighborhoods to “gain hazards including vacant properties and their
negative effects on crime, public safety, and health”).
78 See supra Section II.2.

18

 Wendy Chun

Professor and Simon Fraser University's Canada 150 Research Chair in New Media
in the School of Communication
Kate Crawford

Distinguished Research Professor, New York University; and Co-founder of AI Now
Institute
Steve Demarest

Research Scholar, Center on Race, Inequality, and the Law
Ben Green

Research Fellow, AI Now Institute
Sarah L. Hamilton-Jiang

Research Scholar, Center on Race, Inequality, and the Law
Frank Pasquale

Piper & Marbury Professor of Law, University of Maryland
Rashida Richardson

Director of Policy Research, AI Now Institute
Jason M. Schultz

Professor of Clinical Law, New York University School of Law
Robert Seamans

Associate Professor of Management and Organizations, New York University Stern
School of Business
Vincent M. Southerland

Executive Director, Center on Race, Inequality, and the Law
Meredith Whittaker

Co-founder of AI Now Institute

19