Exploiting Value System Structure for Value-Aligned Decision-Making

Nowadays, Artificial Intelligence faces a significant challenge in implementing ethical behaviour in autonomous agents. The development of technologies that consider human values in the decision-making process of such agents is crucial. While this feature should be included in the agent’s design objectives, the primary goal is to build value-aware agents that act appropriately in changing environments. This requires the creation of mechanisms that support the agent’s own value system. This paper presents a model of an agent that makes decisions based on its value system and a value promotion scheme for available actions. The model uses a fuzzy measure to represent the value system, and aggregation operators to determine the action that best aligns with the agent’s preferences. Schwartz’s basic value theory and Choquet’s discrete integral are used as an example of model implementation. Finally, the model is illustrated by determining the value alignment of different decisions in a simplified tax-paying context.