Evolutionary design – getting the computer to evolve novel designs

The Economist has this article about how designers are using software to evolve better designs. The approach is called evolutionary design and it “enables a computer to run through tens of millions of variations on an invention until it hits on the best solution to a problem.”

The system takes the design blueprints as a genetic makeup and then evolves a population of different designs. Ones, that work well, are discarded and those that are promising are mixed with other promising designs. This process is repeated for perhaps millions of generations until some truly “inspired” designs are generated.

The article describes cases where evolutionary design has come up with novel designs that human designers had not thought of. For instance:

  • At the University of Sydney, in Australia, Steve Manos used an evolutionary algorithm to come up with novel patterns in a type of optical fibre that has air holes shot through its length. Normally, these holes are arranged in a hexagonal pattern, but the algorithm generated a bizarre flower-like pattern of holes that no human would have thought of trying. It doubled the fibre’s bandwidth.

The evolutionary algorithm (creating and mixing blueprints) is quite easy to develop (I once wrote one), the hard part is evaluating which designs are “good”. That is done using a software simulation of the designs and is the true constraint on this novel technology.

Still, very interesting.