Author Topic: Controlled Evolution  (Read 2621 times)

Offline bacillus

  • Bot Overlord
  • ****
  • Posts: 907
    • View Profile
Controlled Evolution
« on: April 30, 2008, 08:29:49 PM »
I had an interesting idea today: Given lots of variables that are randomly stored at birth, could you make a bot that forcibly evolves itself without resulting in useless/detrimental mutations? The trick would somehow be to slow down mutation once ideal conditions have been reached.
"They laughed at Columbus, they laughed at Fulton, they laughed at the Wright brothers. But they also laughed at Bozo the Clown."
- Carl Sagan

Offline EricL

  • Administrator
  • Bot God
  • *****
  • Posts: 2266
    • View Profile
Controlled Evolution
« Reply #1 on: April 30, 2008, 09:09:11 PM »
The various versions of Lionfish do this very effectively with something like 20 different varibles it modifies itself.  It owned IM as a non mutating bot for weeks and weeks since it could adapt to newcomers by having selection favor bots with specific values in specific ranges without risking deleterious mutations which would destroy functionality whole sale.

This line of thought is why I would like to see local-specific mutation probabilities where different sequences could have different mutation rates and these rates would themselves would be exposed to selection.  My current thinking is that we simply create a new command which does nothing other than influence the mutation rate of the N base pairs following it, at least for point mutations.  Things get stickier for other mutation types.  In real world biology, the mechanics of things like sequence copy mutations are largely a function of DNA copying and repair machinery which is itself exposed to and the result of selection.  Selection has favorred machinery which makes imperfect copies in specific ways at specific frequencies when specific codons are encountered.  This allows geneomes adapt so as to code themselves so as to focus high/low mutation rates to where it does the most good or the least harm.  Things like the DNA sequence coding for the specific poison a Cone Snail produces for example can mutate 10's of thousands of times more frequently than the rest of the genome, a result of selection favorring this do to co-evolution with preditors whose anti-posions mutate just as frequently.

Eventually, I think this is where DB will head.  I.e. the simualtor itself will perform no mutations at all and only provide organisms the most primitive of DNA copying machinery that they utilize qat a much higher level for reproduction.  The simulator copy code would itself introduce no errors but organisms would as they reproduce, taking control of their own mutation probabilities and mechanisms.

Afterall, the worst thing a genome can do in a changing world is to stay the same....  
Many beers....