Systèmes d'Information et Management


The software evolution is constrained by the symbolic nature of computers. Nervertheless, the brain is not a symbolic machine, it is able to adapt itself to the environment modifications. The brain is continually building up, it modifies its own structure to adapt itself to the perceived world and to the life events. We examine the possibilities of a research way which is usually considered as a dead end: how to provide computers with the ability to modify themselves. This article proposes a neuronal multi-agent system which relies on recent works in neurosciences and genetic. It borrows from the nervous system some adaptation facilities which are not currently included in usual neural networks. Neural reactive agents are provided with operators which are used not only to tune-up some parameters but to change radically the communication mode according to the frequency and the intensity of their stimulation inputs.