Artificial morality based on particle filter: For moral dilemmas in intelligent transportation systems

Abstract

Advanced driver assistance systems seem to be the core of many intelligent transportation systems aiming to increase safety. Although, the ever-present question of how safe will it be to allow an artificial intelligence based system to perform safety-relevant functions remains. We believe this problem falls in the field of affective computing, also called emotion AI. This paper offers an exemplary solution by means of particle filters, creating and developing the concept of Artificial Faith to apply it into methods of Artificial Morality. Subsequently, these new concepts are applied to an AI localization system estimating position by Global Navigation Satellite Systems. The tested AI uses spatial road networks data for the identification of which lane the vehicle travels, exploiting prior networks information. However, incorporating complex information constraints the system to highly non-Gaussian posterior densities, extremely difficult to represent with accuracy. The particle filter approaches with and without digital map-matching present no restrictions for non-linearity of models as well as noise distribution, therefore allowing both the speed and the heading measurement errors to be modelled accurately. Therefore, it is proven that the tested AI system can increase its quality by improving its accuracy, capturing multi-modal distributions, while having a natural way of incorporating information such as digital maps, when available. These particle filter based location estimators work as part of the tested AI localization system, providing the newly developed concepts of soft and hard artificial morality and solving a simple artificial moral dilemma for artificial intelligent accuracy-based quality function, reducing its referential position uncertainty.