House Rules: Addressing Algorithmic Discrimination in Housing through State-Level Rulemaking

William Simpson

 

Introduction

As is the case for many federal agencies,[1] the Department of Housing and Urban Development (HUD) is intent on addressing the risk of algorithmic discrimination within its primary statutory domain—housing. But in the wake of Loper Bright,[2] which overturned Chevron[3] deference, and with it the general acquiescence of federal courts to agency interpretations of relevant statutes, HUD is forced to regulate AI and algorithmic decision-making in the housing context through guidance documents and other soft law mechanisms.[4] Such quasi-regulation impairs the efficacy of civil rights law like the Fair Housing Act[5] (FHA) and subjects marginalized groups to continued, and perhaps increasingly insidious,[6] discrimination. With HUD crippled in terms of effectuating meaningful AI regulation, states like Maine—which remains a Chevron state—must step up within their respective jurisdictions to ensure that algorithmic discrimination is mitigated in the housing sector.

 

A Brief Primer on Chevron and Loper Bright

In 1984, the Supreme Court held that where a “statute is silent or ambiguous with respect to a specific issue . . . a [federal] court may not substitute its own construction of [the statute] for a reasonable interpretation made by the administrator of an agency.”[7] In other words, where an agency interpretation of an ambiguous statute is reasonable, a court must defer to the agency. Proponents of Chevron deference have heralded the opinion for its placement of policy decisions in the hands of expert and politically accountable agencies,[8] whereas detractors deemed it a violation of the separation of powers doctrine.[9] In June 2024, the detractors won out.

“Chevron is overruled,” wrote Chief Justice John Roberts.[10] To wit, “courts need not and under the APA may not defer to an agency interpretation of the law simply because a statute is ambiguous.”[11] Roberts rested his opinion on the separation of powers principle,[12] a textualist construction of § 706 of the Administrative Procedure Act,[13] a historical analysis,[14] the insurance of Skidmore deference,[15] and the fact that Chevron was subject to numerous “refinements” over the years.[16]

It goes without saying that this jurisprudential U-turn has profound implications for HUD and the statutes it implements.[17] As a result of Chevron’s demise, “any rulemaking proposed by HUD . . . may be more vulnerable to lawsuits than in years past.”[18] Namely, HUD relies on the FHA to authorize its policies, which “broadly describes . . . prohibited discriminatory conduct,” and which HUD interprets “into enforceable directives to serve Congress’ stated goals.”[19] Without Chevron deference, HUD’s interpretations of the FHA are certain to be questioned, and significant barriers for Americans facing housing discrimination will arise.[20]

 

HUD’s Effort to Combat Algorithmic Discrimination in a Post-Chevron Paradigm

In apparent anticipation of such challenges to its interpretations, HUD has resorted to soft law mechanisms like guidance documents to combat algorithmic discrimination. Importantly, these informal mechanisms do not carry the force of law, and are therefore outside the scope of Chevron deference and unaffected by the Loper Bright decision.[21] Such documents include HUD’s “Guidance on Application of the Fair Housing Act to the Screening of Applicants for Rental Housing,”[22] and “Guidance on Application of the Fair Housing Act to the Advertising of Housing, Credit, and Other Real Estate-Related Transactions through Digital Platforms.”[23] The former pronouncement examines how housing providers and tenant screening services can evaluate rental applicants in a nondiscriminatory way—including by choosing relevant screening criteria, using accurate records, remaining transparent with applicants and allowing them to challenge decisions, and designing screening models for FHA compliance.[24] Of note, the document confirms that the FHA “applies to housing decisions regardless of what technology is used” and that “[b]oth housing providers and tenant screening companies have a responsibility to avoid using these technologies in a discriminatory manner.”[25]

Alternatively, the latter document “addresses the increasingly common use of automated systems, such as algorithmic processes and Artificial Intelligence . . . to facilitate advertisement targeting and delivery” vis-à-vis housing related transactions.[26] Like tenant screening services, algorithmic targeting and delivery of advertisements “risks violating the [FHA] when used for housing-related ads,” and can implicate both advertisers and ad platforms.[27] For example, liability may arise by using algorithmic tools to “segment and select potential audiences by [protected] category,” “deliver ads only to a specified ‘custom’ audience,” or “decide which ads are actually delivered to which consumers, and at what location, time, and price.” [28]  The document recommends that advertisers use ad platforms that proactively mitigate discriminatory practices and that they “monitor outcomes of ad[] campaigns for housing-related ads.”

Indeed, “[w]hile the guidance represents an important step forward in safeguarding housing rights, it isn’t currently more than a suggestion to housing providers.”[29] Hence the dilemma facing regulators in this post-Chevron paradigm: issue a formal rule that will provide the intended protection but is prone to litigation, or deliver informal pronouncements that remain largely immune to challenge but fail to offer enforceable requirements against harmful practices.[30] As this administrative predicament persists, it is state governments, including Maine, that must fill the resulting void.

Maine Should Prohibit Algorithmic Discrimination in Housing Through Formal Rulemaking

Like a handful of other states, “Maine . . . has a long history of giving what [its] Supreme Judicial Court describe[s] as ‘deference to an agency’s reasonable interpretation of an ambiguous statute that it administers.’”[31] Such deference results in due part from the fact that “[i]n Maine, a statute must spell out in advance what kind of rules an agency can adopt under it, and through what process.”[32] Fortunately, this deference has permitted the Maine Human Rights Commission (MHRC), which is responsible for implementing  the Maine Human Rights Act (MHRA),[33] to promulgate various regulations which protect against discrimination in the housing sector.[34]

These regulations include prohibitions against “engag[ing] in any conduct relating to the provision of housing which otherwise makes unavailable or denies dwellings to persons because of protected class status,”[35] and “caus[ing] to be made, printed or published, any notice, statement or advertisement with respect to the sale or rental of a dwelling that indicates any preference, limitation or discrimination because of protected class status.”[36] But in the age of algorithmic decision-making and machine learning, and where the state legislature has failed to address the issue, these prohibitions don’t go far enough.[37]

Accordingly, the MRHC should propose formal rules that speak specifically to algorithmic discrimination in the context of screening applicants and advertising for housing, as well as confirm that the MHRA applies not only to housing providers, but to tenant screening services and advertisers as well. Moreover, such rules should include transparency requirements as well as demand regular audits of algorithmic tools, among other safeguards. Short of formal rulemaking, the MHRC should provide guidance, as HUD has, or public hearings on how relevant parties can avoid algorithmic discrimination, and how the public can further educate itself on the risk of algorithmic harms as well as on housing rights in Maine. Only then can we hope that renters and buyers in the state will be effectively protected.

Such regulatory action is a novel proposal—in fact, as of publication, no other state agency has proposed rules pertaining to the mitigation of algorithmic bias in the housing context. Even so, agencies like California’s Civil Rights Department have “proposed regulations to protect against discrimination in employment resulting from the use of artificial intelligence, algorithms, and other automated decision-making systems.”[38] Moreover, an emerging slate of comprehensive AI laws, most notably Colorado’s AI Act,[39] regulate “‘high-risk artificial intelligence systems’—systems that decide whether a consumer receives educational or employment opportunities, financial, healthcare, legal or essential government services, housing, or insurance.”[40] Novelty notwithstanding, in the absence of federal regulations and state law, state-level administrative rules to address algorithmic discrimination in housing are crucial.

 

Conclusion

The end of Chevron deference will have ongoing consequences for federal agencies and their regulatory efforts, especially in the realm of fast-evolving technology like algorithmic decision-making tools. As discussed here, such is the case for HUD and its efforts to limit algorithmic discrimination in the housing sector. In such circumstances, states like Maine, which continue to afford heightened deference to their agencies, should lean on their regulators to fill in legislative gaps where emerging issues demand agile and responsive remediation. Here in Maine, as elsewhere, the risk of algorithmic discrimination in violation of civil rights law is a glaring problem. State-level administrative rules just might be the glaring solution.

 

 

 

References 

[1] See Consumer Fin. Prot. Bureau et al., Joint Statement on Enforcement of Civil Rights, Fair Competition, Consumer Protection, and Equal Opportunity Laws in Automated Systems (2024), https://www.justice.gov/opa/pr/five-new-federal-agencies-join-justice-department-pledge-enforce-civil-rights-laws.

[2] Loper Bright Enter. v. Raimondo, 144 S. Ct. 2244 (2024).

[3] Chevron U.S.A. Inc. v. Nat. Res. Def. Council, Inc., 467 U.S. 837 (1984), overruled by Loper Bright Enter. v. Raimondo, 144 S. Ct. 2244 (2024).

[4] See Dep’t of Hous. & Urb. Dev., Off. of Fair Hous. & Equal Opportunity, Guidance on Application of the Fair Housing Act to the Screening of Applicants for Rental Housing (2024) [hereinafter Guidance on Tenant Screening], https://www.hud.gov/sites/dfiles/FHEO/documents/FHEO_Guidance_on_Screening_of_ Applicants_for_Rental_Housing.pdf; see also Dep’t of Hous. & Urb. Dev., Off. of Fair Hous. & Equal Opportunity, Guidance on Application of the Fair Housing Act to the Advertising of Housing, Credit, and Other Real Estate-Related Transactions through Digital Platforms (2024) [hereinafter Guidance on Advertising], https://www.hud.gov/sites/dfiles/FHEO/documents/FHEO_Guidance_on_Advertising_through_Digital_Platforms.pdf.

[5] 42 U.S.C. § 3601 et. seq.

[6] See e.g., Guidance on Tenant Screening at 1 (“[t]hese technologies can . . . lead to a less transparent process by obscuring the precise reasons for a denial from the housing provider and applicant.”).

[7] Chevron, 467 U.S. at 843-44.

[8] Loper Bright, 144 S. Ct. at 2294 (Kagan, J., dissenting) (“Congress has conferred on that expert, experienced, and politically accountable agency the authority to administer . . . the statute giving rise to the ambiguity or gap.”). Moreover, “Chevron has allowed consistent application of regulations in district courts across the country and across [varied] agencies . . . .” Jeevna Sheth & Devon Ombres, Loper Bright and Relentless: Ending Judicial Deference to Cement Judicial Activism in the Courts, Center for American Progress (Jan. 10, 2024), https://www.americanprogress.org/article/loper-bright-and-relentless-ending-judicial-deference-to-cement-judicial-activism-in-the-courts/.

[9] See Buffington v. McDonough, 143 S. Ct. 14 (2022) (Gorsuch, J., dissenting from denial of certiorari). In his dissent, Justice Gorsuch declared Chevron deference to be an “abdication” of judicial duty. Id. at 16.

[10] Loper Bright, 144 S. St. at 2273.

[11] Id.

[12] See id. at 2257.

[13] See id. at 2261.

[14] See id. at 2262.

[15] See id. at 2267 (“expertise has always been one of the factors which may give an Executive Branch interpretation particular ‘power to persuade, if lacking power to control’” (quoting Skidmore v. Swift & Co., 323 U.S. 134, 140 (1944))).

[16] See id. at 2268-70.

[17] See e.g., Ending Chevron Deference: Impacts on Civil Rights, Center for American Progress (Jan. 10, 2024), https://www.americanprogress.org/wp-content/uploads/sites/2/2024/01/Civil-Rights-1.pdf (“Chevron deference has allowed HUD to curtail widespread segregation and discrimination in housing, ameliorating the devastating effects of redlining policies that included refusing to insure mortgages in Black neighborhoods.”).

[18] Jason E. Bring et al., What the Supreme Court’s Loper Decision Means for the Affordable Housing Industry, Arnall Golden Gregory (July 22, 2024), https://www.agg.com/news-insights/publications/what-the-supreme-courts-loper-decision-means-for-the-affordable-housing-industry/.

[19] Jeevna Sheth & Devon Ombres, Loper Bright and Relentless: Ending Judicial Deference to Cement Judicial Activism in the Courts, Center for American Progress (Jan. 10, 2024), https://www.americanprogress.org/article/loper-bright-and-relentless-ending-judicial-deference-to-cement-judicial-activism-in-the-courts/.

[20] Id.

[21] See United States v. Mead Corp., 533 U.S. 218, 226-27 (2001) (holding that Chevron only applies “when it appears that Congress delegated authority to the agency generally to make rules carrying the force of law, and that the agency interpretation claiming deference was promulgated in the exercise of that authority.”).

[22] Guidance on Tenant Screening.

[23] Guidance on Advertising.

[24] Guidance on Tenant Screening at 11, 13, 14.

[25] Id. at 1. It’s notable, albeit inapposite to the discussion at hand, that the United States District Court of Massachusetts arrived at a similar conclusion in denying defendant’s motion to dismiss plaintiff’s claims under the FHA, where defendant provided algorithmic tenant screening services to housing providers, effectively controlling the decision to approve or reject a rental application by determining how its proprietary algorithm calculated applicant scores. See Louis v. Saferent Sols., Civil Action 22-CV-10800-AK (D. Mass. Jul. 26, 2023).

[26] Guidance on Advertising at 1.

[27] Id. at 1-2.

[28] Id. at 4, 7, 8, 11.

[29] Unpacking HUD’s New Guidance on Algorithmic Tenant Screening, TechEquity (May 29, 2024), https://techequity.us/2024/05/29/unpacking-huds-new-guidance-on-algorithmic-tenant-screening/.

[30] Take for instance HUD’s proposed rule “Reducing Barriers to HUD-Assisted Housing,” which would likely reduce the discriminatory criteria available for algorithmic screening tools. See Reducing Barriers to HUD-Assisted Housing, 89 Fed. Reg. 25332 (published Apr. 10, 2024). Before Loper Bright, “a court would have provided deference to HUD’s interpretation of the [FHA] in implementing this proposed rule, but now a court might view HUD’s new screening requirements as an overreach of HUD’s authority under the [FHA] . . . .” Bring, supra note 18.

[31] Kaitlin Caruso & Anthony Moffa, Opinion: The Chevron doctrine is dead, long live the Chevron doctrine, Portland Press Herald (July 3, 2024), https://www.pressherald.com/2024/07/03/opinion-the-chevron-doctrine-is-dead-long-live-the-chevron-doctrine/. For the Law Court’s adherence to Chevron deference, see e.g., Scamman v. Shaw’s Supermarkets, Inc., 2017 ME 41, ¶ 18, 157 A.3d 223, as corrected (Mar. 23, 2017).

[32] Caruso, supra note 31.

[33] 5 M.R.S. § 4551, et seq.

[34] See 94-348 C.M.R. ch. 8, § 1, et seq.

[35] Id. at § 4(A)(3).

[36] Id. at § 4(A)(4).

[37] See MHRC Ann. Rep. 6 (2023), https://www.maine.gov/mhrc/sites/maine.gov.mhrc/files/inline-files/MHRC%20Annual%20Report%202023%20FY.pdf (“[i]n FY 2023, the [MHRC] received exponentially more housing complaints than it had ever received before.”). While the increase in complaints was not necessarily linked to algorithmic discrimination, HUD confirms that algorithmic discrimination has “been magnified in recent years by the increasing reliance by housing providers on tenant screening companies,” which in turn rely on “advanced technologies, such as machine learning and other forms of artificial intelligence . . . .” The inference here is that increasing algorithmic discrimination and increasing discrimination complaints are at least nominally related.

[38] Civil Rights Council Releases Proposed Regulations to Protect Against Employment Discrimination in Automated Decision-Making Systems, State of California Civil Rights Department (May 17, 2024), https://calcivilrights.ca.gov/2024/05/17/civil-rights-council-releases-proposed-regulations-to-protect-against-employment-discrimination-in-automated-decision-making-systems/ (emphasis added).

[39] S.B. 205, 74th Gen. Assemb., 2nd Reg. Sess. (Colo. 2024).

[40] John Quinn, The AI Landscape is Shifting Amid State-Level Regulation, Forbes (July 29, 2024), https://www.forbes.com/councils/forbesbusinesscouncil/2024/07/29/the-ai-landscape-is-shifting-amid-state-level-regulation/ (emphasis added).