Article by Klaus Æ. Mogensen
The man-made, evolutionary design process resembles natural evolution in that it pits a number of half-finished solutions against each other in a battle for the survival of the fittest. The solutions that come out on top are permitted to survive and multiply. The process is repeated hundreds of times until only the best solution remains – thus breaking away from the concept of conscious design propagated by lofty designers. Welcome to …
The title of this article might lead readers to think that the topic is the age-old discussion of Darwinist evolution versus intelligent design and creationism. And so it is, but with us in the role of intelligent designers, competing with evolutionary forces. We are used to the fact that the artificially created objects that surround us and which we use on an everyday basis are the result of carefully planned design, with the designer having made conscious decisions about how best to complete a given assignment. However, a new trend in the field of design (especially technological design) is starting to gain ground: letting evolutionary forces define the direction of the design through random mutations and natural selection – i.e. the core concepts of Darwin’s theory of evolution. With “Darwin as the designer”, we can find solutions to otherwise impossible problems and gain new insight into how things actually work. On the other hand, there is a risk that the approach will dumb us down. The idea is fundamentally very simple: if you cannot identify the best way to solve a problem, you can start by letting a number of half-finished solutions battle it out. The solutionsthat fare best are then allowed to “survive” and multiply, with the addition of minor mutations. Then you let these second-generation solutions compete with each other in the same way, and repeat the process through hundreds or thousands of generations. The mutations that provide the best solutions are allowed to continue and thrive, while unfit mutations simply die out, just like in nature. If the process functions as intended, you are left with a much better solution than the one with which you started.
As a simple example, think of software programs that play chess. These have a number of parameters that can feature numerous different values. It can be difficult to find precisely the combination of parameters that produces the best result – i.e. the program that is best at playing chess. You can test various versions of the software against human players, but this will take a very long time, and may not provide a particularly objective result (the program may have won because the human opponent was not having a good day). Instead, you can let the various versions of the software compete against each other in a series of virtual chess tournaments played exclusively on a fast and powerful computer. The “children” of the best version – featuring a variety of mutated parameters – progress to the next round, where they play each other in a new tournament, and the process continues until you are left with a configuration that is better than anything a human programmer could create.
Space ships, brain surgery and artificial intelligence
NASA, the American space agency, is just one of the organisations that has benefited from evolutionary design. For its missions to outer space, NASA uses what are known as ion engines to power its probes. Ion engines work by accelerating electrically charged particles (ions) in an electrical field and then firing them backwards out of the probe. The problem is that the accelerated ions have to pass through an electrically charged grid, which is therefore gradually broken down. On average, a grid of this kind lasts a little less than three years. To find a solution to this problem, Cody Farnell, an aerospace engineer, simulated a process of natural selection of mutated grids in order to come up with a grid that would last longer without compromising the effect. At the end of a process involving hundreds of generations, Cody Farnell was left with a grid design that could extend the lifetime of the ion engine to five years – an 80 per cent improvement.
In some cases, illnesses such as epilepsy and Parkinson’s disease can be controlled through the insertion of implants that use electrical impulses to stimulate nerve centres deep within the brain. Researchers into brain function have tested electrical pulses of different kinds – square or wave-formed, for example – but have been unsure about which shape worked best. A team of researchers at Duke University in the United States recently used evolutionary design involving simulated brain cells to identify the optimal wave form (known as a truncated Gaussian curve)., This requires less current and can therefore increase the service life of the battery in the implant by around 60 per cent. As changing the battery requires an expensive surgical procedure that is not wholly without risk, this is a significant improvement.
In a computer at Michigan State University, artificial life forms have been mutating and multiplying for generations. Their “DNA” comprises strings of computer code, and they live in an artificial environment consisting of cells with different amounts of “food”. Over the course of thousands of generations, these life forms have developed the ability to search out the areas that contain the most “food”. This requires a rudimentary form of intelligence and memory, which has thus arisen through random mutations rather than through conscious design. It also has practical applications, because the researchers have transformed their artificial life into software that can control a simple robot.
Benefits and drawbacks
The examples cited above are just a few, recent examples of the use of evolutionary design. The method can also be used to improve the lift capacity of an aircraft’s wings, to reduce the wind resistance of vehicles, to develop lighter and stronger bridges, and for many other purposes. In addition, evolutionary design can lead to new insight into how the world works. If, for example, the evolutionary process leads to a design that is radically different to that expected, you have to ask yourself “why does this work?” And if you can find the answer to this question, it can pave the way to all kinds of much more conscious design.
However, there are also drawbacks linked to evolutionary design. Generally speaking, you have to let the process of evolution take place in a simulated environment on a computer, and it is not certain that what works best in a (typically simplified) simulation will also work best in reality. Moreover, there is also a risk of over-specialised, evolutionary blind alleys. Chess software developed via evolutionary design may well be excellent at beating other chess programs that play using similar strategies, but utterly useless when faced with a human opponent who plays a little differently.
The biggest drawback is the fact that we do not always know why one design developed through artificial evolution is any better than the others. It works just fine, but we do not understand why. The computers of the future will be able to use evolutionary design to design everything better and more quickly than we can – but the process does not provide us with any great insight. The fairytale hero Aladdin was delighted to have the Genie of the Lamp to fulfil his every desire, but at the end of the day he was still an ignorant street urchin. And that is what we risk if we choose Darwin as the designer.
Denise Ngo: Rocket Scientists Use Darwinian Software to Evolve Better Ion Engine Designs, popsci.com May 2010. www.tinyurl.dk/19801
Paul Marks: Darwinian algorithm cuts the need for surgery, New Scientist July 2010. www.tinyurl.dk/19800
Catherine Brahic: Artificial life forms evolve basic intelligence, New Scientist August 2010. www.tinyurl.dk/19799