From Programming Agents to Educating Agents - A Jason-Based Framework for Integrating Learning in the Development of Cognitive Agents

摘要

Recent advances and successes of machine learning techniques are paving the way to what is referred as Software 2.0 era and cognitive computing, in which traditional programming and software development is meant to be replaced by such techniques for many applications. If we consider agent-oriented programming, we believe that such developments trigger new interesting scenarios blending cognitive architecture such as the BDI one and techniques like Reinforcement Learning (RL) even more deeply compared to what has been proposed so far in the literature. In that perspective, we aim at exploring the integration of cognitive agent-oriented programming based on BDI with learning techniques so as to systematically exploit them in the agent development stage. The approach should support the design of BDI agents in which some plans can be explicitly programmed and others instead can be learned by the agent during the development/engineering stage. In that view, the development of an agent is metaphorically similar to an education process, in which first an agent is created with a set of basic programmed plans and then grow up in order to learn plans to achieve the goals for which the agent is meant to be designed. This paper presents and discusses this medium-term view, introducing a first model for a BDI agent programming framework integrating RL, a first implementation based on Jason programming language/platform and sketching a roadmap for this research line.