CORE Talk: Understanding Adversarial Influence Campaigns in the Age of Digital Surveillance and GAI

Understanding Adversarial Influence Campaigns in the Age of Digital Surveillance and Generative AI with a Multiplex, Agent-Based Model

Abstract: 
In our work, we aim to complement empirical research on the measurement of influence campaigns on social media platforms with a new simulation framework that allows us (and other interested researchers) to experiment with the changes in policies, strategies, and technologies that have previously been out of view. We accomplish this goal through the creation and release of an agent-based model framework utilizing multiplex network representations. Our work is inspired by the idea of creating digital twins of real world systems to better understand and predict their functions and interactions as well as approaches to uncertainty quantification that leverage simulations to propagate honest estimates of variations across complex systems. Specifically, we create computational objects that represent different advertising-tech surveillance and generative AI tools, distinct types of adversaries with (or without) access to those tools, and diverse constellations of experts and citizens. Each of the these components is modular and thus the framework allows different affordances and features to be turned on or off, tuned, or mixed and matched within and across simulations. Uniquely, this framework allows researchers to run paired experiments with and without adversaries and/or technologies, in order to understand and probe the mechanisms that lead to democratic problem-solving or dis-function through explicit comparisons between a baseline control and treatment condition. We provide a summary of the multiplex agent-based framework here as well as preliminary evidence of the potential combined dangers of social media surveillance and generative AI tools when available to external adversaries. 

Join via Zoom: https://aarhusuniversity.zoom.us/j/61750343089