I have started working on a genetic algorithm which breeds object suitable for laser cutting. So far I’m only using artificial selection for the choice of most fit individuals. User is able to define the amount of time which the object will take to cut on a specific laser cutter. This value then serves as a fitness function (the closer particular phenotype is to desired cut time the more fit it is). Ultimately I’m planning to optimize for cost here. Generative procedure of the phenotype is what needs a lot of development too, but first I’m going to develop the entire GA setup.

Images above show that over time the whole population converges to a valid design solution. The breeding process, however, continues to look for even more suitable one. Thanks to that property software behavior becomes an illustration of the point which Steadman makes in Evolution of Designs – that quite different design solutions can be a valid answer to one well defined fitness condition. Aesthetics remain an ill defined problem, thus an artificial selection step where designer (or indeed a prosumer) can say which object he or she prefers to take forward is needed. This is why next step will be programming an artificial selection into the GA setup.

Most interestingly from designer’s point of view the same GA setup can work with entirely different generative procedures (entirely different phenotypes) if only the fitness objectives are the same. You can see my first try of the software underneath. I was dealing with a simple very simple phenotype generative procedure – stacking sections one on top of the other. What’s important is that It shows the same tendency as the more developed generative procedure – it converges to one design solution and then flips to a very different but still valid one.