Top Legal Risks With Generative Ai

• Publication year: 2024
• Citation: Vol. 4 No. 2

[Page 103]

Graham Reynolds, Robin Sagstetter, and Damon W.D. Wright ^*

Abstract: AI is dramatically transforming our world in positive and negative ways. Among the positives, AI makes it amazingly fast to make decisions or create interesting content. Among the negatives, AI makes it amazingly easy to violate, even unknowingly, a variety of laws. In the near future, we can expect to see a robust body of "AI law" through federal and state legislative action, as well as Federal Trade Commission and other agency rulemaking. But the absence of "AI law" today does not mean there is no law. Rather, businesses are being sued every day for alleged misuse of AI in violation of well-established existing law. As discussed in this article, the courts are seeing a host of AI cases involving a variety of issues, including algorithmic bias, copyright infringement, right of publicity, and web scraping.

Algorithmic Bias

The term algorithmic bias is broadly understood to mean the unfair treatment of individuals or groups by automated decision-making tools such as artificial intelligence (AI). In the same way human bias can trigger liability, algorithmic bias can do the same—making it vital that AI developers and deploying businesses develop and monitor their AI tools to avoid unfair or illegal discriminatory outcomes.

The complexities were recently highlighted in Federal Trade Commission v. Rite Aid Corporation et al., ¹ where the Federal Trade Commission (FTC) accused Rite Aid of using facial recognition technology in ways that potentially harmed consumers due to algorithmic bias. The FTC's complaint outlined several areas where Rite Aid purportedly fell short when using facial recognition software to combat in-store robberies, including inadequate consideration

[Page 104]

of the risks associated with false-positive identifications and failure to test the system's accuracy in a thorough manner. These false-positives were particularly detrimental to Black, Asian, Latino, and female consumers, who the FTC alleged were more likely to be wrongly identified by the technology.

To rectify these issues, the FTC settlement requires that Rite Aid implement a comprehensive oversight program, mandating annual risk assessments to identify and mitigate potential harms, especially those disproportionately impacting consumers based on race, ethnicity, gender, and other characteristics. Businesses that deploy AI tools, especially in the context of facial recognition, will need to familiarize themselves with this decision. It will not be enough for a business to highlight the oversight features implemented before the use of AI. If the AI generates outcomes that are detrimental to a protected class, a business can find itself in litigation crosshairs.

Currently, there is legislation on the horizon with nine states introducing measures aimed at curbing algorithmic discrimination. These proposed bills vary but generally share common features to the Rite Aid decision, such as mandatory impact assessments to identify discriminatory risks, disclosure requirements to government agencies, and mechanisms for individuals to opt out of AI-driven decisions. Notably, most legislative efforts emphasize government enforcement, rather than a private right of action. Nevertheless, we can expect to see an increase in private plaintiffs reciprocating the FTC's claims under civil rights and unfair and deceptive trade practices laws.

Automated Employment Decision Making

The integration of AI in employment hiring, promotion, compensation, and termination decisions has likewise led to a closer examination of their implications on fairness and nondiscrimination in the workplace.

A significant case here is Equal Employment Opportunity Commission v. iTutorGroup, Inc., ² which addressed the challenges of using AI in hiring process. The Equal Employment Opportunity Commission (EEOC)...