On the relevance of value system structure for automated value-aligned decision-making

Technological advances, especially in the field of artificial intelligence, raise subjects related to the behaviour of autonomous systems. The ability to judge how far some behaviour is ethically correct is inherent to humans, and one of the great challenges is to replicate such ability in artificial systems. Engineering, aware that human values must be considered in the decision-making process, is currently focused on developing mechanisms to embed these values into autonomous systems. This article proposes a decision model that enables autonomous intelligent agents to select actions that are aligned with their value system. It considers not only the relevance that different values have for an agent, and how much an action promotes a certain value, but also models in a rigorous manner the interaction of different values within a value system. An instantiation of the model is put forward, based on Schwartz's basic values theory, non-additive fuzzy measures, and the discrete Choquet integral. Then the performance of the approach is illustrated by determining the value alignment of different decisions using the well-known iterated prisoner's dilemma.