Incremental Inductive Learning in a Constructivist Agent

International audienceThe constructivist paradigm in Artificial Intelligence has been definitively inaugurated in the earlier 1990's by Drescher's pioneer work. He faces the challenge of design an alternative model for machine learning, founded in the human cognitive developmental process described by Piaget. His effort has inspired many other researchers. In this paper we present an agent learning architecture situated on the constructivist approach. We present details about the architecture, pointing the autonomy of the agent, and defining what is the problem that it needs to solve. We focus mainly on the learning mechanism, designed to incrementally discover deterministic environmental regularities, even in non-deterministic worlds. Finally, we report some experimental results and discuss how this agent architecture can lead to the construction of more abstract concepts