Bots and Simulations > Evolution and Internet Sharing Sims
I have an idea
-venom-:
have you ever seen that movie called "evolution" were a meteor brings alien life to our planet and it quickly evolves into more and more complex things? well I was thinking is there a way to hype up mutations and reproduction enof to increase evolutionary speed similar to the movie??
Numsgil:
Evolution is a function of the mutations/generation ratio. If it's too high, mutations overpower any natural selection and you have a mutation meltdown. The DNA just turns to non coding goo. If it's too low, evolution lacks the novelty to break out of local minima (something that's short term beneficial but long term is a dead end).
In addition, the idea that evolution is a nice linear process from a to b is extremely tempting but inaccurate. It's more like a blind drunk guy trying to walk home. It's extremely susceptible to dead ends. On Earth, most major advances in complexity were in response to a previous massive extinction event. Quite often you need to wipe the slate clean and start over.
And last, there's an undeniable element of time involved. There's a finite speed that molecules can be moved from place to place with osmosis. This sets a finite pacing for natural selection to act that can't really be sped up.
So in short, increasing the mutation rate won't work. Increasing the generational turnover will help, but only up to a point. If you could somehow dilate time (for real life) or massively increase the cycles/sec (on a computer), you could speed it up, but both are rather hard to do. And even then, you could run a simulation forever, and not get anything more complex than when you started because evolution is extremely prone to finding dead ends that it can't navigate out of (for computer people, evolution is basically like an imperfect greedy search done in massive parallel).
The movie, in case you're wondering, is extremely naive in its treatment of biology. Not just evolution.
EricL:
The problem is that most mutations are delaterious or neutral. Advantageous mutations are rare and take time to fixate in a population (function of conveyed advantage, generation time and lifetime fucundity) and generally only convey a small relative advantage. Delaterious mutations on the other hand are generally 100% deadly and (this is key) advantageous mutations don't help you defend against mutations. Delterious mutations are essentially random death from on high. Evolution requires a low enough deleterious mutation rate that advantageous mutations have "time" to become relatively widespread in a population relatively quickly relative to the delaterious mutation rate. If the deleterious mutation rate is too high, advantageous mutations can't fixate before they are destroyed.
Let me try an example. Assume 500 out of 1000 mutations are delaterious. 499 are neutral and 1 is advantageous. (rather optomistic). Thus, the chances a mutation will adversely impact you is 50%. Now assume the birth mutation rate is 100%. Namely, all births mutate. That's a pretty high mutation rate. This means 50% of all offspring will have lower fitness, 49.9% will have the same fitness and one out of 1000 will have greater fitness than it's parent.
With a fixed population (like in our sims) average lifetime fucundity is 1 given asexually reproducing organisms (it has to be or the population would increase or decrease over time) but that is an average. Some die without giving birth. Some give birth more than once but one or more offspring die before reproducing. On average, one offspring for each parent survives to reproduce. At 50% deleterious mutations, if we assumed mutations were the only thing impacting reproductive sucess, then on average, every orgamism has two offspring, one of which dies at boirth and one of which survives to reproduce. On average. If we assume preditors and accidents and just being plain unlucky, the ratio is higher. Lets assume on average, a bot has 4 offspring in it's life, one of which survives to reproduce. Half are killed at birth due to deleterious mutations and one of the four is killed by something else before he can reproduce. Lets also assume an advantagous mutation improves a bot's chances of surviving to reproduce by 10% (also very opotomistic) but for simplicity sake, if a bot lives that long, he reproduces 4 times. Namely, an organism with an advantageous mutation will have on average, a 10% better chance of surviving to reproduce relative to those without it.
What is the probability that when advantageous mutation occurs, it can survive long enough to fixate in the population? Given the high 100% mutation rate, the chances are bascially 0. The chances are 55% an organism born with an advantageous mutation will survice to reproduce (by definition, it was born without a delterious mutation, so it's one of the two offsrping that on average, are neutral or advantageous and since it is in fact one of the 1 in 1000 with an advantageous mutation, it has a 10% better chance of surviving than it's neutral sibling, but only 10%. It can still get eaten, fall in a hole, starve, etc.). But of the 4 offspring it will have on average if it survies to reproduce, half will die at birth due to delterious mutations. Their inherited advantage helps them there not at all. Of the two that don't die at birth, the chances of each surviving to reproduce is similarly 0.55. Thus, the chances an advantageous mutation survives to the third generation is the probability of the first surviving multiplied by the probabilitiy of at least one of the second surviving: (.55) * (.55 + (.55/2)) = 45%. Already less than 50/50. This is because both things must happen for the mutation to still be in the population at this stage: the original bot with the mutation must have survived to reproduce AND one or both of it's neutral offspring must also have survived. The chances of surviving to the 4th generation are lower still and so on. I may not have the math exactly right, but you get the idea. The high mutation rate prevents even highly advantageous mutations from surviving long enough to fixate. It's like trying to survive in a nuclear reactor.
So, bottom line, you need a low enough mutation rate so that when an advantageous mutation does happen, it can spread in the population before delterious mutations wipe it out...
-venom-:
oooo lots of knowledge
ok I realize that hyping up the muts wont work.
and I do realize more complicated isn't always better.
but I was just wonderin if there was a way to speed up evo (other then faster cycles)
for porposes like zerosims and the likes
EricL:
--- Quote from: -venom- ---but I was just wonderin if there was a way to speed up evo (other then faster cycles)
for porposes like zerosims and the likes
--- End quote ---
I think utilizing self-regulating shepard bots as a means to optomize selective pressures towards certain goals holds a lot of promise....
Navigation
[0] Message Index
[#] Next page
Go to full version