Applying categories with prototypes to incomplete observations in artificial cognitive agents

Abstract

Studies show that artificial cognitive agents can be equipped with mechanisms for learning of basic language categories. However, after a period of initial learning performed under perfect circumstances, agents get to be deployed into dynamic real world environments leading to possibly partial observations of agent's surroundings. This paper presents a general strategy for applying categories with prototypes on incomplete observational data. It is assumed that the task is carried out by an artificial agent which has autonomously developed its private ontological knowledge base using complete observations. The agent expresses its internal uncertainty about an assignment of a category to an observed object by relying on epistemic modal operators of possibility, belief, and knowledge. An underlying theory builds upon accomplishments of a theory of grounding of feasible epistemic statements in artificial cognitive agents.