ATheoretical and Applicative Framework for Evolutive Computers

With the generalisation of massively parallel systems, computer engineering lifts a new scientific challenge: that of formulating an evolution theory, able to rule the self-organisation of artificial collectivities and individuals, inducing a rise in their performance, ability, reliability, and robustness; in other words, a theory which would ground their viability and intelligence. Towards this goal, genetic algorithms provide a framework both conceptually new and powerful on the computational side. However, this computer engineering prospect for evolution stays insufficient in many respects. The work we present aims at expressing a theory of evolution which counters the weaknesses of genetic algorithms. Essentially, we seek to heighten their operationnality by refining the darwinian model upon which they are based, to enable a fast replay of the history of life over silicon components. In practice, we suggest a reinstatement of Lamarckian laws, leading to a view reconciling constructivism and finalism, and restoring Monod's idea of a project of life. So doing, we propose to move towards the autonomisation of the concept of machine, considering it as an integral part of the living mechanics from which it was born.