Author Topic: Survival of the flattest  (Read 3614 times)

Offline Greven

  • Bot Destroyer
  • ***
  • Posts: 345
    • View Profile
Survival of the flattest
« on: April 13, 2006, 08:56:01 PM »
Okay I will make it short because it is really late at my location...

I read something interesting on the net about the survival of the flattest, if that rings a bell.

My main concern on DB was always that highy fit hand written bots always did de-evolve into something less fit, and thereby the new less fit bot got the overhand in the sim and outcompeted the hand written bots, and in the end the end all bot became extinct. Which for me always seemed very strange. But some research in avida seems to indicate that this is possible when mutations rates are to high.

In short:
that low-performance individuals can outcompete high-performance individuals if they are more robust to mutations. Were robust means, that the DNA (code whatever) are more likely to get mutations which are not fitness decreasing.

Mmmhh, dont really know if we can but something like this onto DB. But somehow it seems to fit better into my DB view. I mean it is perfectly likely that the highly tuned hand written bot will continue to get degrading mutations, often because they have so much interaction between the genes and they are so compacted and structured vs. normal evolved bots, were the code often is in a messy state....

Just some random rabble... Hope I manage to get something meaning out of it.
10010011000001110111110100111011001101100100000110110111000011101011110010110000
011000011000001100010110010111101001110100110010111100101000001000001111001011101
001101001110011011010011100011110100111000011101100100000100110011010011100110110
010110000011100111101001110110111101011101100110000111101001101001110111111011101
01100100000111010011010001100001110111010000010001001000010100001

Offline Numsgil

  • Administrator
  • Bot God
  • *****
  • Posts: 7742
    • View Profile
Survival of the flattest
« Reply #1 on: April 13, 2006, 09:26:35 PM »
This has alwways been shvarz's opinion I believe, and I tend to agree with it.  If your bots are devolving your mutation rates are too high.  We've been saying it for quite a while, you haven't been ignoring us have you

Offline Welwordion

  • Bot Destroyer
  • ***
  • Posts: 325
    • View Profile
Survival of the flattest
« Reply #2 on: April 13, 2006, 09:27:08 PM »
Well is that not obvious? I mean if you destroy a wheel in a clockwork the whole clock ceases to function, however an organism works more flexible and can either compensate for certain loses or produces an effect trough the combination sum of multiple attempts , such when one gene is deactivated you just weaken but not destroy they effect.
But mainlyits just simply that organisms are used to defects and evolve forms that are stable to defects.
« Last Edit: April 13, 2006, 09:27:49 PM by Welwordion »

Offline Greven

  • Bot Destroyer
  • ***
  • Posts: 345
    • View Profile
Survival of the flattest
« Reply #3 on: April 13, 2006, 09:41:43 PM »
I know that they have been much talk about this, but it wasnt enough for me, and didnt really explain why the less fit bots didnt got lesser ...

Well then, now we have a name for the "to high mutation rates"... Survival of the flattest, interesting article by the way.
10010011000001110111110100111011001101100100000110110111000011101011110010110000
011000011000001100010110010111101001110100110010111100101000001000001111001011101
001101001110011011010011100011110100111000011101100100000100110011010011100110110
010110000011100111101001110110111101011101100110000111101001101001110111111011101
01100100000111010011010001100001110111010000010001001000010100001

Offline EricL

  • Administrator
  • Bot God
  • *****
  • Posts: 2266
    • View Profile
Survival of the flattest
« Reply #4 on: April 14, 2006, 12:33:10 PM »
So, there is evidence that most biological mutations leading to variation are not external events but rather a consequence of different kinds of DNA replciation errors.   Mutation probability is not equal across the genome in such cases and is in large part determined by how codons are encoded.  The degenerate nature in which different BPs are combined to encode different codons which transcribe for the same amino acid allows for the same genotypic expression to have different mutation probabilities depending upon the encoding.   This means that the mutation probability at particular loci is at least in part, susceptable to selection upon the different codon encoidings (and on other encoidng artifacts as well such as repeat sequences, etc) and there is recent research showing that it is selection upon mutation probability that has much to do with creating stable genome elements as well as areas which are more inclined to produce genotypic variation where such variation is advantageous.  My term for this, which may actually mean something else in acedemic circles, is 'Genotypic Plasticity".

What this says to me is that an ALife system which provides only externally defined, fixed mutation probabilities and does not allow mutation probability to vary for different parts of the genome and be selected upon is incomplete.  In fact, one could argue (I would) that the *primary* source of meaningful mutation and hence variation in nature is not being simulated and that the result is analogous to attempting to evolve in a highly radioactive environment.  If the radiation is too high, all things complex wither.  If too low, the pace of mutation is insufficient to achieve meaningful variation in finite time.  Genotypic Plasticity allows for both I.e. high varation and fast paced mutation where such is advantageous, low mutation probability where such is not...

-E
Many beers....

Offline Endy

  • Bot Overlord
  • ****
  • Posts: 852
    • View Profile
Survival of the flattest
« Reply #5 on: April 15, 2006, 05:12:54 AM »
I think DB actually is one of the better alife sims in this regard. Since it's all math based evolution can arrive at the same solution via a variety of methods. I think we should actually factor robustness in as part of fitness during evolution. In normal combat mode its no use, but with evolution involved the fitness landscape radically changes.

The bots do seem to like some of what we give them. Vital dna that they can't improve or kills them if they mess with, is kept. If the trait/identifier is selected for/against by a group it can also be self-reinforcing.
« Last Edit: April 15, 2006, 05:14:32 AM by Endy »