Author Topic: SciAm article about how diversity of animal forms arises  (Read 2023 times)

Offline Trafalgar

  • Bot Destroyer
  • ***
  • Posts: 122
    • View Profile
SciAm article about how diversity of animal forms arises
« on: May 05, 2008, 10:12:54 PM »
There's is a very interesting and very detailed article on Scientific American's site about how and why the great diversity of animal forms arises despite the genes being so similar. Essentially, genes have 'enhancers' which determine when and where they turn on, and genes can have more than one enhancer.

http://www.sciam.com/article.cfm?id=regulating-evolution

Offline goffrie

  • Bot Builder
  • **
  • Posts: 65
    • View Profile
SciAm article about how diversity of animal forms arises
« Reply #1 on: May 06, 2008, 08:02:29 PM »
SciAm looks a lot like "scam"

Anyways, interesting article.

Offline Moonfisher

  • Bot Overlord
  • ****
  • Posts: 592
    • View Profile
SciAm article about how diversity of animal forms arises
« Reply #2 on: May 10, 2008, 07:33:04 AM »
Nice article, haven't finished it yet, but what it describes is a lot like I had immagined it. All sorts of genes than can do all sorts of things in different situations getting shut down or activated or reused...
I was very happy to read about the "conditions" being able to fire several genes. I'm still playing with the idea of making a MOD  for DB where the mutations would be building or breaking down a neural network like structure.
I have a lot of small ideas on how to make it work, but one idea was that outputs could have a condition output and only get fired when the condition was met, and to let these conditions make connections to anything.
(It wouldn't actualy build a neural network as we know them, but the way it works would be sort of the same.)
Basicaly I'm hoping to create a more "fluid" and redundant way for actions to evolve.
I don't realy know how normal genes work and I'm thinking this is probably closer to how brains evolve, but I don't realy care, I looove neural networks    and I just want to see evolving code making faster progress, I don't realy care if it's the same way nature does it. I can immagine that dna does have more binary actions and is just so long and redundant that in the end you reach a complexity that hopefully encourages positive outcomes from mutations or just has a lot to fall back on.
But I'm not patient enough for something like that to arise, and the way conditions work in DB seem a litle too... human... I definately think inline conditions was a huge improvement where evo sims are concerned, but I'm still not quite convinced.
I know conditions probably don't work like a neural network in nature (In single cells anyway), but the point of trying to build it this way is to force the redundancy and high amount of conditions to arise faster... it sounds right to me that actions would evolve to have a very high amount of litle conditions that seem irrelevant to the action itself.

So far I've tryed to evolve a neural network structure using point mutations, and it actualy seems to be working, lots of funny litle behaviors, and the way it navigates makes no sence, but works very well.
It basicaly moves sideways all the time because of a broken output and uses .up and .setaim to stear, so it aproaches enemies sideways and occasionaly turns to shoot at them.
This is mostly because theres very litle food in the sim and the most imporatnt form of behaviors that seem to realy make a difference are :
- Not killing the alge (Since odds are very low that you'll find one again in your lifetime)
- Nudging the alge around in random directions without loosing controll of it. (Makes it harder for others to steal the alge)
- Reproducing when in pain. (This seems to be the best defence against attacks so far, creating some confusion and increasing odds that one of your own will end up with the alge) (It also makes some shell, but thats not very interesting)
- Shooting around the alge. (Not sure why it's so important, but saw less wastefull bots that had very good alge controll and only shot occasionaly, but it was never realy a success, I'm guessing the shots where good for keeping others away, and apparently worth the energy, or maybe the less wastefull behavior just never got lucky, who knows...)
- If getting realy big while feeding on an alge they also make a very big child in order to reduce it's shot strength (Or it could risk killing the alge)

Bottom line is it seems that with the low amount of food the most important behaviors seem to be the ones that help you keep whatever alge you found, everything else comes in second. It's also important to find the alge, so spreading out seems to play a factor, there was one specie wich was good at covering the entire field finding alge that where ending up near borders or corners, but at one point a new color suddenly seemed to become very successfull and virtualy starved out the entire left side of the field. It seemed the only difference was that it had very good controll over the alge and was very carefull never to kill it. Since then the spectrum of colors is now very varied again and slowly the spreading behavior is resurfacing as severely mutated bots accidently kill an alge somewhere.

Anyway I got way off topic, just saying I like the article, will read the rest soon.