nroberts

I was reading Darwin's Dangerous Idea and I came across a discussion of the beginnings of life...from the ooze. Well, the problem is that DNA is incredibly complex. 10 billion years is not enough time for it to occur, statistically, given the rules of chemestry and physics as starting points and allowing the components to float around in a puddle.

But, there are many theories that account for it by starting smaller...as one would expect. Basically, there are some molecules that will replicate and are very simple. They in turn would have evolved into the more complex and stable forms. However, there's no fitness function on molecules. Or is there?

Given a completely random chance that a given string of molecules will copy itself we STILL end up with order from non-order. I have begun experimenting with this idea on my computer. Last night I developed a simple genetic algorithm with no fitness function. It just randomly selected from the old generation and copied into the new.

Starting with a completely random set of strings, within 100 generations we will arrive at around 3 distinct strings from 100. Adding mutation that number changes drastically depending on how well the strings copy themselves. So, for instance, a mutation rate of 50% never really arrives at anything, even given a billion generations to evolve in, but a mutation rate of 1% will arrive at around 5-10 distinct strings at the end of a 100 generation cycle.

This just shows that completely random selection can resolve to a pattern of order. The likelihood of randomly starting with a uniform density is very rare. That you'd still have it at the next generation even more so...and so on. What this doesn't show is that stable copying could have evolved.

The next step I wish to go through is to base the mutation rate on the strings themselves, just as it works in Chemestry. The G and C components of DNA if strung together copy more successfully that other strings of molecules. If I impose something like this on the genetic algorithm, I expect that the pattern will quite normally resolve to sequences having these characteristics.

None of this is ground breaking of course...it's all based on other people's work. But it's interesting to me at any rate so I thought I'd share.