As artificial systems (of hardware, software and networks) continue to grow in size and complexity, the engineering traditions of rigid top-down planning and control are reaching the limits of their applicability. In contrast, biological evolution is responsible for the apparently unbounded complexity and diversity of living organisms. Yet, over 150 years after Darwin's and Mendel's work, and the subsequent "Modern Synthesis" of evolution and genetics, the developmental process that maps genotype to phenotype is still poorly understood. Understanding the evolution of complex systems - large sets of elements interacting locally and giving rise to collective behavior - will help us create a new generation of truly autonomous and adaptive artificial systems. The Generative and Developmental Systems (GDS) track seeks to unlock the full potential of in silico evolution as a design methodology that can "scale up" to systems of great complexity, meeting our specifications with minimal manual programming effort. Both qualitative and quantitative advances toward this long-term goal will be welcomed.
The genotype is more than the information needed to produce a single individual. It is a layered repository of many generations of evolutionary innovation, shaped by two requirements: to be fit in the short term, and to be evolvable over the long term through its influence on the production of variation. "Indirect representations" such as morphogenesis or string-rewriting grammars, which rely on developmental or generative processes, may allow long-term improvements to the "genetic architecture" via accumulated layers of elaboration, and emergent new features. In contrast, "direct representations" are not capable of open-ended elaboration because they are restricted to predefined features.
While complex genotypes may not be required for success in simple environments, they may enable unprecedented phenotypes and behaviors that can later successfully invade new, uncrowded niches in complex environments; this can create pressure toward increasing complexity over the long term. Many factors may affect environmental (hence genotypic) complexity, such as spatial structure, temporal fluctuations, or competitive co-evolution.
Today's typical numbers of generations, sizes of populations, and components inside individuals are still too small. Just like physics needs higher-energy accelerators and farther-reaching telescopes to understand matter and space-time, evolutionary computation needs a boost in computational power to understand the generation of complex functionality. Biological evolution involved 4 billion years and untold numbers of organisms. Nature could afford to be "wasteful", but we cannot. We expect that datacenter-scale computing power will be applied in the future to produce artificially evolved artifacts of great complexity. How will we apply such resources most efficiently to "scale up" to high complexity?
The GDS track has recently added a new focus: defining quantitative metrics of evolved complexity. (Which is more complex - a mouse, or a stegosaurus?) The evolutionary computing community is badly in need of such metrics, which may be theoretical (e.g., Kolmogorov complexity) or more practical. Ideally, such metrics will be applicable across multiple problem domains and genetic architectures; however, any efforts will be welcomed. We encourage authors to submit papers on these quantitative metrics, which will be given special attention by the track chairs this year.
The GDS track invites all papers addressing open-ended evolution, including, but not limited to, the areas of: