An Architecture for an Adaptive System with Primitive Concept Formation and Consciousness

We have designed an architecture, which we call the Image Driven Architecture, that adapts to an environment using primitive concept formation and consciousness through neural networks. First, to construct the architecture, we devised a simple neural network based on Edelman´s natural selection. Then, using the neural network, we configured a complex multi-module system with three layers of components: micro-features, images and concepts, and combinations of concepts or scenes. The module for selection and integration, which plays the central role of consciousness, chooses efficient and useful concepts for the system through a mutual vote method that we devised. With regard to the Image Driven Architecture, we have clarified the main functions of consciousness, such as the selection of efficient concepts or images, association, displaying dim imagery, and the learning control of the system.