The overarching question imparting urgency to this exploration is: Can U.S.-Russian contention in cyberspace cause the two nuclear superpowers to stumble into war? In considering this question we were constantly reminded of recent comments by a prominent U.S. arms control expert: At least as dangerous as the risk of an actual cyberattack, he observed, is cyber operations’ “blurring of the line between peace and war.” Or, as Nye wrote, “in the cyber realm, the difference between a weapon and a non-weapon may come down to a single line of code, or simply the intent of a computer program’s user.”
When algorithms fall short of articulated expectations, people get hurt.
In order to hold those who build AI systems accountable for the consequences of their actions, we can operationalize a system for auditing algorithmic systems. Algorithm audits have for years been part of the conversation in the context of online platforms, but are just beginning to emerge as a mode of external oversight and evaluation regarding the deployment of "automated decision systems" (ADS), and making their way into critical policy proposals as a primary mechanism for algorithmic accountability. However, as we learnt from many other industries, not all audits are made equal. In this talk, Raji will discuss the ongoing challenges involved in executing algorithm audits effectively and highlight the technical, legal, and institutional design interventions necessary for audits to serve as a meaningful approach to accountability.
