Traditional engineered ICT systems are qualitatively different from natural complex systems (CS). The former are made of unique, heterogeneous components assembled in complicated but precise ways, whereas the latter mostly rely on the repetition of agents following identical rules under stochastic dynamics. Thus, while natural CS often generate random patterns (spots, stripes, waves, trails, clusters, hubs, etc.), these patterns generally do not exhibit a true *architecture* like human-made ICT systems possess. There are, however, major exceptions. (a) ICT-like CS: On the one hand, biology strikingly demonstrates the possibility of combining pure self-organization and elaborate architectures, such as: the self-assembly of cells into body plans and appendages, the synchronization of neuronal signals into cognitive states of the brain, or the stigmergic collaboration of social insects toward giant constructions. (b) CS-like ICT: Conversely, large-scale ICT systems already exhibit complex emergent effects, albeit still mostly uncontrolled and unwanted. Segmentation and distribution of large computing systems over a multitude of smaller and simpler components (integrated parts, software agents, network hosts), have become an inescapable reality in many domains of computer science & engineering, AI and robotics. Thus, while some natural CS seemingly exhibit all the attributes of ICT systems, ICT systems are becoming natural objects of study for CS science. Such cross-boundary cases are examples of *self-organized architectures*-i.e., how spontaneous systems need not always be random and engineered systems need not always be directly designed-a hybrid concept insufficiently explored so far. I will illustrate this goal with a spatial multi-agent model of *programmable* and *reproducible* morphogenesis that integrates self-assembly (SA) and pattern formation (PF) under the control of a nonrandom gene regulatory network (GRN) stored inside each agent of a swarm. The differential properties of agents (division, adhesion, migration) are determined by the regions of gene expression to which they belong, while at the same time these regions further expand and segment into subregions due to the self-assembly of differentiating agents (SA is to "self-sculpting" what PF is to "self-painting"). This model offers a new abstract framework, which I call *Embryomorphic Engineering* (coined after Neuromorphic Engineering) to explore the developmental and evolutionary link from genotype to phenotype that is needed in many emerging computational disciplines, such as artificial embryogeny and collective robotics.
Artificial Embryogeny, Spatial Computing, Evolutionary Development, Complex Systems, Bio-Inspired Engineering, Self-Organization
Doursat, R.. (2009)
Facilitating Evolutionary Innovation by Developmental Modularity and Variability.
GECCO'09: Genetic and Evolutionary Computation Conference. ACM.