Bridging the Proof Gap: Reconstructing Algorithmic Impact Assessment for Anti-Discrimination Litigation

Abstract

Algorithmic Impact Assessment (AIA), as a core tool of algorithmic governance, currently serves mainly administrative regulation and enterprise compliance, and fails to effectively respond to the evidentiary needs of individual victims in anti-discrimination litigation. When a plaintiff claims that an algorithmic decision constitutes discrimination, it is usually necessary to overcome three barriers: blurred subject of responsibility, broken causal relationship and difficulty in obtaining evidence. This paper takes the Mobley v. Workday case as an entry point, analyzes the causes of the above three dilemmas, and proposes four institutional designs for the systematic reconstruction of Algorithmic Impact Assessment. Firstly, human-machine comparison benchmark is introduced to establish the admissibility standard for the algorithmic conclusions in the AIA report; Secondly, the adversarial assessment mechanism shall be established so that group statistical evidence can be used for individual causal proof; Thirdly, a technology-legal sandbox shall be constructed to provide a standardized platform for the generation and review of AIA reports; Fourthly, expand the functional boundaries of the Algorithmic Liability Insurance, link the insurance mechanism with the risk assessment conclusion of the AIA report, and open up the final relief channel from evidence to compensation. The four systems jointly form a closed loop system of standard setting, evidence generation, review accessibility, and practical relief, which promote the shift from prevention without relief to simultaneous prevention and relief, so that the rules of proof in anti-discrimination litigation can be dynamically adapted to the continuous evolution of AI technology.