Sunday, September 1, 2002

Self-Organizing Circuit Reinvents Radio [Slashdot] What is really interesting in this story is not addressed in the posting or in the original New Scientist article, although it comes up in the /. discussion. Really. "Evolutionary" programming does randomized hill-climbing of a fitness function. It is very difficult to craft a fitness function that downgrades all undesirable designs, however perverse. The critical difference between evolutionary programming and biological evolution is that the fitness function in evolutionary programming is an arbitrary human construct intended to steer mutations toward a goal. In the present case, the fitness function was some measure of the ability to produce oscillations; it didn't say anything about all the other requirements that are usually implicit in an engineering design. In other words, the concept of an "oscillator" in engineering is far more than "there is a periodic voltage variation between these two wires." The fitness function in biological evolution is whatever environmental conditions determine. (These conditions include what components are already available for building upon from earlier evolution.) The assumption that one can use "evolutionary" means to reach a desirable goal is based on an unfortunately prevalent naïve teleological reading of biological evolution. 3:17:54 PM