On the Impact of Gaslighting on Partially Observed Stochastic Control

Abstract: Recent years have witnessed a significant increase in cyber crimes and system failures caused by misinformation. Many of these instances can be classified as gaslighting, which involves manipulating the perceptions of others through the use of information. In this paper, we propose a dynamic game-theoretic framework built on a partially observed stochastic control system to study gaslighting. The decision-maker (DM) in the game only accesses partial observations, and she determines the controls by constructing information states that capture her perceptions of the system. The gaslighter in the game influences the system indirectly by designing the observations to manipulate the DM's perceptions and decisions. We analyze the impact of the gaslighter's efforts using robustness analysis of the information states and optimal value to deviations in the observations. A stealthiness constraint is introduced to restrict the power of the gaslighter and to help him stay undetected. We consider approximate feedback Stackelberg equilibrium as the solution concept and estimate the cost of gaslighting.