Hi IC,
Sorry to see you go. Because I think the problems you have with the computer evolution example are common (such programs take advantage of what is more formally known as genetic algorithms), I'll explain a little anyway.
I'm not going to refresh my memory by going out on the web, but the description at
NewScientist.com in the
Radio emerges from the electronic soup article is of an incredibly primitive experiment. Transistors on circuit boards fitted with programmable switches? Not since the 1980s at the latest, so I have no idea why they were doing this in 2002.
The article feels funny to me for another reason, and that is because it is so similar to an earlier experiment performed in the US using PGAs (Programmable Gate Arrays), which is how one would actually perform such an experiment today. The outcome in the US was precisely the same. Experimenters used genetic algorithms driving the programming in PGAs in an attempt to obtain an oscillator design. The winning 'organism' did produce an oscillating signal, but only because it was picking up electromagnetic emissions from the oscillator in a nearby computer. In other words, in the exact same way as this more recent result, an oscillator was created by designing a radio receiver. This was written up in EE Times a while back, but the article doesn't seem to be turning up in a search of their archives today.
That the experimenters say they don't know the mutational path taken that resulted in a radio receiver means that they kept very poor records, another alarm bell for me.
Anyway, a genetic algorithm is just a simulation of the evolutionary process. There has to be a final goal, and in this case it was an oscillator. There has to be an evaluation algorithm which assesses who well each 'organism' (design version) satisfies the final goal. The winning designs have their 'genes' (design parameters) mixed in some way while producing 'offspring' designs in order to simulate sexual reproduction. Then the offspring are assessed for how well they measure up against the final goal and the process is repeated until there is a winner, in this case a working oscillator.
Simulation of anything by a computer program, be it evolution or the weather or of particle interactions or of blackjack hands, is not a process which involves interaction by humans. Humans only write the programs and set up the initial conditions. Meteorologists do not design tomorrows weather when they simulate weather patterns in order to create a forecast. Particle physicists are not designing what the particles will do when they simulate particle interactions to figure what will happen in the next experiment. And electrical engineers are not designing the circuits that simulations of the evolutionary process carry out.
Evolution is just selection followed by descent with modification. You can see that my description of genetic algorithms follows this process precisely. First designs are selected to be bred for the next generation. Then the design parameters from the breeding population are mixed (often between breeding pairs, but of course a computer program doesn't have biology's constraints and can mix the parameters in many different ways) to produce offspring which differ from the parents, which is descent with modification.
Genetic algorithms are a excellent demonstration of the ability of the evolutionary process to create novelty, in this case using a radio receiver to create an oscillator.
--Percy
Edited by Percy, : No reason given.
™ Version 4.2
