Alongside 40 other civil rights and technology advocacy organizations, Upturn called on the Federal Trade Commission to develop specific, concrete civil rights protections in the Commission’s ongoing Commercial Surveillance and Data Security Rulemaking. Specifically, the letter urges the FTC to promulgate a specific rule prohibiting discrimination as an unfair practice under the FTC Act, as well as a specific rule requiring reasonable and appropriate measures to detect and address algorithmic discrimination in sensitive domains.
Dear Chair Khan, Commissioner Slaughter, and Commissioner Bedoya,
We, the undersigned groups, write to emphasize the importance of developing specific, concrete civil rights protections in the Federal Trade Commission’s (FTC’s) ongoing Commercial Surveillance and Data Security Rulemaking (ANPR). We applaud the Commission’s broad efforts to rein in unfair and deceptive commercial surveillance practices. In order to protect civil rights in a data-driven economy, the forthcoming proposed rules must clearly prohibit discrimination as an unfair trade practice.
As documented throughout the many comments submitted in response to the FTC’s ANPR, the effects of discrimination still define the lived experiences for many people in the United States. Despite decades of effort to root out and redress discrimination, people who are marginalized because of race, ethnicity, national origin, religion, sex, including pregnancy, sexual orientation, gender identity, disability status, or income continue to experience discrimination when accessing basic goods and services, seeking economic opportunities, and pursuing safe and healthy lives.
In particular, the comments document how the widespread use of automated, data-driven technologies by companies to shape key decisions about people’s lives can exacerbate structural discrimination. Today, many of the technologies that actively contribute to discrimination in credit, employment, education, housing, and healthcare rely on a range of basic statistical models and more complex machine learning and other artificial intelligence techniques. But even as commercial practices evolve, the underlying material harms to consumers as a result of discrimination persist.
As a recent joint statement from the FTC, CFPB, EEOC, and DOJ makes clear, automated systems can contribute to unlawful discrimination. The statement clarifies that federal agencies have existing authorities to “ensure that these rapidly evolving automated systems are developed and used in a manner consistent with federal laws.”
We agree that these existing authorities must be used to their fullest extent to combat discrimination. But it is also true that existing civil rights laws and regulations have not kept pace with technological changes. As a result, “we must consider what other legal protections currently exist outside of direct civil rights statutes.” In order to ensure that these technologies do not harm people of color and other protected classes, specific FTC intervention is necessary.
To firmly protect civil rights in the forthcoming rulemaking process, the FTC must:
1. Issue a specific rule prohibiting discrimination as an unfair practice under the FTC Act.
Discriminatory practices often easily satisfy the statutory unfairness test. We strongly agree that “[w]hen a business substantially injures a person because of who they are, and that injury is not reasonably avoidable or outweighed by a countervailing benefit, that business has acted unlawfully.” The FTC has applied this simple, straight-forward framework in Passport Auto, and should enshrine this reasoning into a clear rule. Public policies, statutory interpretation, market dynamics, and longstanding FTC practices support this approach.
Such a rule would provide a systemic regulatory response to many forms of discrimination that data minimization efforts alone cannot provide. For example, even robust, laudatory efforts at data minimization do not reach, or explicitly exempt, public records, criminal and eviction records, employee records, or other de-identified records. But these are precisely the types of records that frequently drive discriminatory outcomes when they are used to train and validate algorithmic systems.
2. Issue a specific rule requiring reasonable and appropriate measures to detect and address algorithmic discrimination in sensitive domains.
Given the urgent harms that algorithmic discrimination poses to consumers, the FTC should require companies to take affirmative and proactive measures to identify and redress discrimination in their use of automated systems. Failure to do so should constitute an unfair practice. This approach bears resemblance to the FTC’s long-standing data security work, which has led companies to employ reasonable and appropriate measures to protect consumers’ data. In the context of discrimination, such measures could include routine and ongoing quantitative and qualitative assessment of model performance across demographics; evaluation of multiple models to determine if there exists less discriminatory, but similarly effective, alternative models; and evaluation of training data for representativeness and bias. Demographic testing and evaluation for discrimination throughout the model development pipeline is necessary.
The FTC should ensure that these specific anti-discrimination rules proceed as their own Notice of Proposed Rulemaking (NPRM), separate from other potential data security and privacy rulemakings. There are at least two reasons to do so. First, each proposed rule would be simpler and provide more regulatory certainty. For example, the relevant definitions required for regulations on discrimination are quite contextual and distinct from those for data security and privacy. As former FTC Commissioner and current CFPB Director Rohit Chopra notes, “[m]arkets work best when rules are simple, easy to understand, and easy to enforce.” A large, multi-issue rule would be needlessly complex. Second, the rulemaking process would be significantly easier for FTC staff to manage, and for interested stakeholders to engage in. Separate rulemaking processes that focus on different issues would lead to more targeted engagement and ultimately better considered outcomes.
Thank you for your attention to these matters. For any questions or further discussion, please contact Logan Koepke (firstname.lastname@example.org), David Brody (email@example.com), and Frank Torres (firstname.lastname@example.org).
Algorithmic Justice League
American Civil Liberties Union
Asian Americans Advancing Justice - AAJC
Center for Democracy & Technology
Center for Digital Democracy
Center on Privacy & Technology at Georgetown Law
Communications Workers of America
Data & Society Research Institute
Demand Progress Education Fund
Electronic Frontier Foundation
Electronic Privacy Information Center
Fight for the Future
Government Information Watch
Japanese American Citizens League
Joint Center for Political and Economic Studies
Lawyers’ Committee for Civil Rights Under Law
NAACP Legal Defense and Educational Center, Inc.
National Fair Housing Alliance
National Health Law Program
National Hispanic Media Coalition
National Housing Law Project
National Urban League
National Women's Law Center
Open Technology Institute
Restore The Fourth
The Leadership Conference on Civil and Human Rights
United Church of Christ Media Justice Ministry