The Daily Telegraph covers an experiment to breed the perfect web page design through Darwinian natural selection.
Matthew Hockenberry and Ernesto Arroyo of Creative Synthesis, a non-profit organisation in Cambridge, Massachusetts, have created evolutionary software that alters colours, fonts and hyperlinks of pages in response to what seems to grab the attention of the people who click on the site.
Evolutionary algorithms have in the past been used to design aircraft wings and boat hulls. Because there are so many competiting and contributing factors to how well a certain design works, a human designer won’t find that optimum design which works the best. The idea of using an evolutionary algorithm is you start off with thousands of random and different designs. Each design then undergoes a form of natural selection – for aircraft wings, each design would be loaded into a computer model. The best designs are then “breeded” together to create the next generation of aircraft wings. This process is repeated thousands of times to produce an optimum wing.
The software treated each feature as a “gene” that was randomly changed as a page was refreshed.
After evaluating what seemed to work, it killed the genes associated with lower scoring features – say the link in an Arial font that was being ignored – and replaced them with those from higher scoring ones say, Helvetica.
Evolutionary algorithms are certainly a useful addition to the toolbox of engineers. There are of course limitations. Creationists sometimes argue “what use is half an eye?” as a refute to Darwinian natural selection. In this example, half an eye is still better than no eye at all: you might be able to see but you won’t get a sharp picture.
But at least you won’t walk into a wall. But say… why haven’t humans evolved wheels? “What use is half a wheel?” is a good question, and there isn’t a use for half a wheel. If any organism did evolve half a wheel, it would be selected against and therefore natural selection would never lead to a whole wheel.
I think this illustrates well why evolutionary algorithms can never be the end all; we’ll still need human input and design. But it’s great as a way to refine a design.