Title: Building Trustworthy Autonomous Systems under Uncertainty: a Probabilistic Approach to Ethical Decision-Making
Abstract: Autonomous systems, such as driverless cars and drones, are increasingly deployed across various fields, promising significant socioeconomic benefits. However, these systems − especially fully autonomous ones − raise concerns regarding safety, ethics, and compliance with legal and social norms.
Trustworthy autonomous systems must demonstrate reliability, safety, and ethical behavior in complex, uncertain environments. Since it is impossible to program responses for all potential scenarios, these systems need to handle uncertainty and make ethically informed decisions within their reasoning processes. A critical component in this respect is the formal verification of their decision-making mechanisms, which requires a verifiable architecture and a sufficiently expressive formal language.
This thesis integrates two research areas: Computationally Grounded, Weighted Doxastic Logic (COGWED) and SimpleBDI, which models an autonomous system’s beliefs and reasoning cycles. The result is theWeighted Doxastic SimpleBDI (WeDo- BDI) logic, which combines SimpleBDI’s reasoning cycle with probabilistic belief representation. This allows the system to compute belief degrees based on sensor data and probabilistic transitions. The WeDo-BDI semantics also includes a preference relation that evaluates plans based on ethical principles and their probability of causing violations.
An example scenario involving an unmanned aerial vehicle demonstrates how WeDo-BDI operates under uncertainty and for ethical decision-making.