par Bersini, Hugues
Référence Self-Organization and Emergence in Life Sciences, Springer Netherlands, page (41-60)
Publication Publié, 2006
Partie d'ouvrage collectif
Résumé : Originally, the field of Artificial Life was born out of the frustration and isolation felt by some "hackers" keen on cellular automata, game of life, genetic algorithms, L-systems and other computer recreations. Fascinated by this surprising cohabitation of programming simple algorithms and the complex working of these same algorithms (this new perception of complex phenomena as emerging from simple algorithms but iterated, distributed and recursive), convinced of the interest of their works for theoreticians of biology but aware of the lack of dialogue with them, they organized a series of workshops whose desired originality was multidisciplinarity and the coming together of researchers sharing the same will to understand the mechanisms and functions characterizing living organisms. These researchers in computer science, mathematics, physics, biology, robotics, philosophy, now meet every year, alternatively in Europe and the USA. What is discussed as inherent to all living organisms, and therefore which represents the bulk of the material dealt with during these workshops, are the mechanisms of self-organization or of the "emerging functionalities" opposing a centralized vision of biology, the need to better balance the coupling of the studied objects with their environment opposing a solipsistic methodology still representative of a certain artificial intelligence, the compulsory passage via the mechanisms of learning and adaptation as the most simple and autonomous way to face the complexity typical of the architecture and dynamics of these systems and, finally, the study of this complexity per se. A same motto brings together all these researchers: "some form of complexity can be faced and domesticated very simply by relying on the computer brute force". The mascots that are most representative of artificial life are: robotic insectoids, the game of life and other cellular automata, genetic algorithms, L-systems and simulations of ecosystems. These first workshops, due to the originality of the process, created a considerable stir. They undoubtedly seemed to reach their primary target, that is to allow better communication between researchers. Today, however, a certain breathlessness is noticeable which goes not without reminding the same dying down that characterized the cybernetic and systemic trends (Alife fathers) of the forties and fifties. The multidisciplinarity although essential to the inspiration does not survive, in principle, the specialization which arises naturally as a consequence of several years of study dedicated to a same subject and which drive researchers to privilege interlocutors sharing their same narrow and deep interest. Gradually new scientific communities appear with a more focused object of study and which, either free themselves of the mother field (like genetic algorithms or cellular automata) or become connected with existing communities (like robotics, study of ecosystems, study of the origin of life, study of insects societies). As we can notice during these workshops, "life" resists whatever unique and narrow definition. This diversity is the de-stabilizing factor which could cause the burst of artificial life. Besides, the risk is important of a forthcoming divorce, which has already taken place in artificial intelligence, between a so called "strong" science which could fuse with an existing scientific tradition (cognitive science for AI and theoretical biology for artificial life) and its so called "weak" counterpart with a more engineering like aftertaste and leading to technological innovations (expert systems, fuzzy logic and knowledge engineering in AI, neural networks, genetic algorithms and autonomous robotics in artificial life). If the artificial life star turns into a supernovae to finally explode and leaves behind, as relics of its glorious past, one and only one scientific pulsar, more focused, firmly grounded, and, above all, perpetrating as well as possible the original enthusiasm, the best candidate I can see could be a more formalized study of the emergent phenomena. My contribution to this characterization of emergent phenomena is currently limited to two of them appearing in a large amount of biological networks: the de-stabilizing effect of frustrated connectivity and the tendency to fragment the whole network into small clusters of units showing similar behavior. Among the networks showing these two emergent properties, the attention will be paid to only two of them: Hopfield Neural Networks (HNN) and Idiotypic Immune Networks (INN). Frustrated connectivity is responsible for perturbing the equilibrium dynamics of the network and provoking "wavering" among alternative equilibrium regimes. When frustrated a homeostatic network exhibits oscillatory behavior while an oscillatory network falls into a new type of chaotic regime which will be designated as frustrated chaos. In HNN, there is a threshold in the degree of connectivity which marks a sharp transition into the dynamics of the network. Below this threshold, i.e. in the case of a strongly diluted connectivity, the network clusters itself into small group of oscillatory units. In IIN also, this clustering phenomenon prevails and follows very regular rules for the dimension and the distribution of the clusters. It is clear that these two properties can be regarded as emergent since in order to appear they require a specific collective configuration of the units, and in order to be detected they require a level of observation which transcends each unit taken separately. In this paper, rather than theoretical analysis, results of computer simulation are given and briefly explained to illustrate these common properties.© 2006 Springer. Printed in the Netherlands.