Evolutionary learning

The Novamente Cognition Engine (NCE) and its open-source derivative, OpenCog, is a serious attempt at commercializing a complete theory of general artificial intelligence. NCE is an agent-based architecture that combines inference in a probabilistic logic with the generality of evolvable procedures that are implemented in a Turing-complete language. This architecture therefore unites probabilistic deductive reasoning with an open-ended evolutionary learning architecture for discovering new categories, representations and long term self-improvement.

The Novamente Cognition Engine is a commercially motivated effort, finding initial application in the creation of intelligent virtual agents for interactive online worlds and computer games such as AGIsim and Second Life. These agents are designed as simple pets that, despite their limited intelligence, are able to learn novel skills from a coach. By building up from simple skills, Novamente hopes to build systems with true general purpose artificial intelligence.

The Novamente Cognition engine stems from earlier work in discovery systems that use evolutionary algorithms to search over reward and solution spaces and, thereby, perform both learning and meta-learning. Well known examples include Lenat’s Eurisko and Haase’s CYRANO. These systems have been used for finding novel and unexpected solutions in many problem domains, but have not found widespread application as a method of general-purpose problem solving.