February 09, 2005

Computers Use Darwinian Algorithms to Design New Gear

There may not be a lot of flying cars yet, but it’s definitely the future.

There’s this wild thing going on in the design world. Computers generate a bunch of designs. The least successful at achieving the specific design goal—for example, a radio antenna—are discarded; the most successful are combined with each other and produce new designs. It’s natural selection applied to product design. It’s all done inside of a computer, so thousands of generations of designs can be generated quickly.

This bizarre approach is actually generating usable products. And the wild thing is—sometimes people have no idea why the products work. From Technology Review:

Evolutionary algorithms, also known as genetic algorithms or GAs, take their cue from biological evolution, which can turn a crawling reptile into a soaring bird without any kind of forward-looking blueprint. In sexual reproduction, the shuffling of each parents genescombined with random genetic mutationcreates organisms with new characteristics, and the less fit organisms tend not to pass on their genes to succeeding generations. Evolutionary algorithms work much the same way, but inside a computer. When Lohn creates a new antenna, for example, he starts off with a population of randomly generated designs and grades their relative performance. Designs that come close to preset goals win the right to intermingle their properties with those of other successful candidates. Designs that disappoint go the way of the archaeopteryx: oblivion.

This gizmo is an antenna:

Breeding antennas takes time, of course. Most designs are downright awful, and it takes a large number of computing cycles to find decent performers. Still, when youve got a computer that can generate and test 1,000 generations an hour, interesting ideas do emerge*. Lohn, a PhD who hasnt taken a course on electromagnetism since his undergraduate years, expects to have at least one of his teams antenna designs go into space this year as part of NASAs Space Technology 5 mission, which will test a trio of miniature satellites. His favorite computer-designed antenna: a corkscrew contraption small enough to fit in a wine glass, yet able to send a wide-beam radio wave from space to Earth. It resembles nothing any sane radio engineer would build on her own.

Evolutionary algorithms are a great tool for exploring the dark corners of design space, Lohn says. You show [your designs] to people with 25 years experience in the industry and they say, Wow, does that really work? The slightly spooky answer is that yes, they really do, as Lohn established after months of testing.

Pretty freaky, huh?

...says NASAs Lohn, There are two schools of thought. One says I just need something that does X, Y, and Z, and if evolution gives me X, Y, and Z, thats all I care about. The other school wants to know whats in there and how it works. We cant really help those people, because we frequently see evolved designs that are completely unintelligible.

That’s hype, of course. If we really wanted to figure out how one of these designs worked, we’d figure it out. Studying such a design that at first seemed unintelligible would surely be a great way to learn new things.

The faster computers get, the more of this stuff we’re going to see.

(Via GeekPress. )

03:58 AM • Blogroll The Big Picture! • Email This to a Friend

Categories: Science & Technology
Most recent comments by: David Goldberg • Vik Rubenfeld • Dave Goldberg •

Replies: 3 comments