Evolutionary programming as an approach to artificial general intelligence has been successful only in creating trivial programs.
What we learn in engineering, programming and even life usually reinforces elegance, simplicity and efficiency. It is not surprising that we approach evolutionary programming and related algorithms in AGI with what amounts to Occam program searches.
But perhaps the compactness of Occam programs is the very reason we can't find them. Perhaps we need to consider un-Occam programs - those that are deliberately inefficient, at least in size. For example by unrolling loops, removing subroutine hierarchies, and so on. Something like spaghetti code or Brooklyn code.

Posted 9/15/07