AI Ethics – Beyond Piecemeal Evaluations: Comprehensive Threat Surface Analysis for Advanced AI Systems

Abstract:

Current AI risk assessment methods often fall short in addressing the full spectrum of potential threats posed by advanced AI systems. This talk will explore the need for a paradigm shift towards comprehensive threat surface analysis, drawing on techniques like Probabilistic Risk Assessment (PRA). We'll discuss how to model complex, cascading risk pathways across technological, social, and environmental dimensions, with a focus on operational AI hazards in sociotechnological settings.

Bio:

Anna Katariina Wisakanto is an AI safety researcher and strategist at the Center for AI Risk Management & Alignment (CARMA), where she heads the Comprehensive Risk Assessment research programme addressing risks from advanced AI systems. Her current work involves developing novel analytical methods to model pathways connecting AI capabilities to potential global-scale harms, while also exploring the intersection of AI, complex adaptive systems, and cognitive science to understand AI's impact on human cognitive and moral autonomy.