Eliezer> Considering the infinitesimal amount of information that Eliezer> evolution can store in the genome per generation, on the Eliezer> order of one bit, Eric Baum wrote: >> Actually, with sex its theoretically possible to gain something >> like sqrt(P) bits per generation (where P is population size), cf >> Baum, Boneh paper could be found on whatisthought.com and also >> Mackay paper. (This is a digression, since I'm not claiming huge >> evolution since chimps).
Correction, I should have said, sqrt(N) bits per generation, where N is the length of the genome. Eliezer> That's for human-built genetic algorithms, not natural Eliezer> selection. Eliezer> For natural selection see Eliezer> e.g. http://dspace.dial.pipex.com/jcollie/sle/index.htm. (I Eliezer> don't buy some of the author's claims here, but the central Eliezer> principle of which he gives a heuristic explanation is Eliezer> something I've heard of before in evolutionary biology; I Eliezer> think it goes back to Kimura.) Natural selection does run on Eliezer> O(1) bits per generation. There's no real difference that I'm aware of. It did go back to Kimura, but I believe he was wrong. Mackay's paper talks about natural selection, but simply applying Baum-Boneh calculation. Eliezer> I furthermore note that gaining one standard deviation per Eliezer> generation, which is what your paper describes, is not Eliezer> obviously like gaining sqrt(P) bits of Shannon information Eliezer> per generation. Yes, the standard deviation is proportional Eliezer> to sqrt(N), but it's not clear how you're going from that to Eliezer> gaining sqrt(N) bits of Shannon information in the gene pool Eliezer> per generation. It would seem heuristically obvious that if Eliezer> your algorithm eliminates roughly half the population on each Eliezer> round, it can produce at most one bit of negentropy per round Eliezer> in allele frequencies. Maybe, but there are a lot of different allele frequencies. N of them to be precise. ... >> I'm claiming that the hard part-- discovering the algorithms-- was >> mostly done by humans using storage and culture. Then there was >> some simple tuning up in brain size, and some slightly more complex >> Baldwin-effect etc tuning up programming grammar into the genome in >> large measure, so we become much more facile at learning the stuff >> quickly, and maybe other similar stuff. I don't deny that if you >> turn all that other stuff off you get an idiot, I'm just claiming >> it was computationally easy. Eliezer> Arguably, in a certain sense it *must* have been Eliezer> computationally easy because natural selection is incapable Eliezer> of doing anything computationally *hard*; evolution can't sit Eliezer> back and design complex interdependent machinery with Eliezer> hundreds of interlocking parts in a single afternoon, like a Eliezer> human programmer. That's a critical and interesting claim, but I'm not sure its correct. Turn on ey gene on a wing, and a fly will grow a fully formed eye there. Evolution can make small changes, such as turning on a single gene somewhere, that create a complex, meaningful program. Is this obviously so different from what the programmer is doing? I'm not sure. How does evolution do this? Well, it has built compact programs that have solved lots of previous problems. The Occam hypothesis says, when you build compact programs that solve lots of problems drawn from a natural distribution, they will continue to solve new problems drawn from the distribution. So evolution has built subprograms and modules that are meaningful, and a small change can now invoke a large meaningful new program that actually does something interesting. I don't think its an accident that small changes continue to build new complex structures solving complex problems in ways that look like the new complex structures must have been designed. And I think this is similar from an interesting perspective to how the programmer is capable of being able to write complex programs that solve new problems. Basically, he's doing the same thing, utilizing pre-built modules that generalize because of Occam's razor. ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
