. Hacking Evolution .             

In a wide range of domains, biological systems have adaptive capacities that far surpass those of man-made systems. It seems reasonable to suspect that there are certain core computational problems that can be tackled much more efficiently than traditionally supposed, and that many of the more remarkable forms of adaptation in the natural world proceed from the discovery, in nature, of extraordinarily efficient ways of solving these core problems. I’m interested in identifying these “core efficiencies”, and in harnessing them to tackle some of the pressing challenges that computer scientists face when engineering adaptive systems.

Over the past few years I’ve focused on the genetic algorithm. I find it striking that this simple, biologically-plausible model of evolution routinely procures good, often great, solutions to a wide range of poorly understood optimization problems. By virtue of imitation, the genetic algorithm seems to harness something of the core computational efficiency underlying the remarkable adaptive capacity of natural evolution.

I’ve developed a new hypothesis about the nature of this core efficiency. My account, called the generative fixation hypothesis, departs from the reigning hypothesis in genetic algorithmics—the building block hypothesis—at a fundamental level. The potential impact of the generative fixation hypothesis extends beyond genetic algorithmics to (amongst others) the fields of evolutionary computation, optimization, machine learning, and evolutionary biology.