Projective Simulation and the Taxonomy of Agency

摘要

In this paper we focus on behaviourism and materialism as theory-driven approaches to the classification of AI and agency in general. We present them and we analyse a specific utility-based agent, the PS model presented first in [2], which has as its key feature the capability to perform projections. We then show that this feature is not accounted for solely by materialistic or behaviouristic stance but represents rather a functional link between the two approaches. This is at the same time central for agency. This analysis allows us to present a feature-driven (or reversed) taxonomy of the concept of agency: we sketch its main characteristics and we show that it allows a comparison of different agents which is richer than the solely be-haviouristic and materialistic approaches. The reason for that lies in the fact that we have reversed the approach to agency from a theory-driven stance to a process-driven one.