Experiments in Neo-computation Based on Emergent Programming

The general objective of this work is to develop a complete programming language in which each instruction is an autonomous agent trying to be in a cooperative state with the other agents of the system, as well as with the environment of the system. By endowing these instruction-agents with self-organizing mechanisms[2], we obtain a system able to continuously adapt to the task required by the programmer (i.e. to program and re-program itself depending on the needs). The work presented here aims at showing the feasibility of such a concept by specifying, and experimenting with, a core of instruction-agents needed for a subset of mathematical calculus. In its most abstract view, Emergent Programming is the automated assembling of instructions of a programming language using mechanisms which are not explicitly informed of the program to be created. We chose to rely on an adaptive multi-agent system using self-organizing mechanisms based on cooperation as it is described in the AMAS theory[1]. An important part of our work on Emergent Programming has been the exploration of the self-organization mechanisms which enable the agents to progress toward the adequate function, depending on the constraints of the environment but without knowing the organization to reach or how to do it.