Economic Theory Seminar Series - Ruitian Lang (ANU)

Title: Scaling Autonomous Decisions: The Critical Role of Cybersecurity

Abstract: The rapid advancement and scalability of machine learning have made it the engine of automated decision-making. As the operational cost of such systems ceases to be the primary limiting factor for scale, a new constraint emerges: the cybersecurity threat. We present a theoretical model demonstrating that for a reasonably accurate system, this threat significantly impacts its optimal size. Our core finding shows that as a system's accuracy improves, the risk of a black-box adversarial attack becomes the sole relevant factor for determining its scale.