Meta-axioms and Complex Preferences in Evolving Logical Agents

Abstract. In this paper, we introduce some improvements to a framework that we have previously proposed, based upon a multi-layered general agent model, where at each layer temporal-logic-like axioms and meta-axioms can be defined and dynamically checked. Their violation determines suitable repair actions to be undertaken by means of appropriate agent’s self-modifications. This in the perspective of a society of agents, where individuals should be able to learn and enlarge their patterns of behavior by observing and generalizing their observations, and also by “imitating” other agents, or by “being told” by them. Here, we extend the overall setting on the one hand by augmenting the temporal-logic-like rules with an ordered conjunction for checking/enforcing an order on events and actions, and on the other hand by introducing the possibility of defining and exploiting complex preferences based upon a (simplified) form of modal reasoning.