Posted by David on August 14, 2003 at 08:39:18:
In Reply to: Re: What is wrong with this picture? posted by Al on August 14, 2003 at 01:49:11:
Al,
The problem with a high mutation rate (and I already bowed to pressure and cranked it up in version 2) is that you end up with instability. Everything is so crappy in the early generations that it's tempting to throw more mutation in to liven things up a bit, but we'd regret it in the long run.
Most evolution does not in fact happen through mutation. At least, "mutation" in the sense of miscopied DNA. Most mutations are the result of recombination. Think how quickly we bred dog species: almost all of that is the result of crossover.
--David
: By mutation I mean adding/subtracting but most importantly, substitution (i.e. introducing new information to the pool). Biologically there are only 4 different markers and mutation happens quite frequently. The biological analogy is limited though because the vocabulary that these 4 letters produce is limited. Even so, that limited vocabulary builds in a redundancy which allows silent mutations to occur.
: High mutation rate does increase the amount of useless junk but with negative selective pressure that should be compensated for. It's what happens in nature - freaks tend not to survive. Crossover does not introduce new variation, merely recombines existing variation. While there is mutation in David's system, the crossover mechanism enables one short sequence to proliferate throughout a species (and even across species to an extent) and weights the outcome towards a gradual homogenisation of the total pool. Worse, v1 showed that the crossover used in that setup selected out long strings of meaning and hence actively discouraged the evolution of poems (coherent meaning which lasts the length of the whole string). I am unconvinced that the problem has been fixed in v2 - the speciation model used here seems only to be slowing the process down a little.
: High mutation rate drives variation. Without mutation the experiment becomes an exercise in creating poetry by shuffling the words you start with. With high crossover it becomes a more randomised experiment (along the lines of monkeys and typewriters) with a smaller word pool. It is certainly not evolution (which depends on surviving individuals being viable). The vast majority of these experimental matings could not survive under normal selective pressure (i.e. they are gibberish).
: The poems need to start on a simple level and gradually build in viable complexity, conserving it from generation to generation. That cannot happen under this model.
:
: : I'm not sure by what you mean with 'mutation'. Are you talking about adding/subtracting a letter, a word, a line? It's been my experience, with programs like Framsticks, that high mutation increases the useless junk in the "DNA" structure. Here, the DNA is not hidden within the body...we're actually reading it, and therefore it can't tolerate extra junk.