The paper “Regulating Algorithmic Harms” by #SylviaLu is a MUST-READ for everyone in #AI governance.

The paper “Regulating Algorithmic Harms” by Sylvia Lu is a MUST-READ for everyone in AI governance. Quotes:

“Given their intangible nature, algorithmic harms are often downplayed as smaller secondary problems. A human who steals our personal identification for identity theft is considered a flagrant violation of privacy, yet an algorithm doing the same thing escapes our attention. Given their elusive nature, algorithmic harms affect us in various imperceptible ways. They compromise our privacy through unauthorized data extraction that reveals our vulnerabilities and intimate details. They distort our autonomy based on highly personalized manipulation and deceptively authentic appearance. They exacerbate inequality through AI-generated content that disproportionately discriminates against minorities. Their addictive and unpredictable operations undermine our health, safety, and security.”(page 5)

“The invisibility of private governance is the third instance of opacity that precludes adequate harm identification and correction. As AI develops into a crucial technology for many businesses, the private governance of AI innovation has grown in significance for stakeholders who wish to mitigate algorithmic harms, including individuals, shareholders, communities, and regulators. The invisibility of such governance hinders harm investigation when external stakeholders lack critical information about the governance regimes for harms arising from AI applications. This governance opacity generally stems from a status quo that has not responded to the intangible nature of algorithmic harms and the commercial secrecy caused by trade secret protection. The absence of legally mandated disclosure of information about private governance of algorithmic operations leads to extensive invisibility in the adequacy of harm mitigation schemes. (…)” (38-39)

“Each of the proposed solutions works to create a synergistic effect for harm correction. Specifically, establishing an individual right against algorithmic harms could potentially act as a deterrent, prompting entities to proactively engage in harm mitigation; implementing harm-centric algorithmic impact assessments and transparency rules would lead to a more holistic interplay among potential wrongdoers, victims, and regulators. (…). A profound understanding of these harms, shaped by these regulatory proposals, can lay the groundwork for more comprehensive reforms. Such reforms could extend beyond procedural efforts and consider more radical changes in areas such as tort law, remedies, and more. (…)” (page 64)