Posted by David Rea on January 09, 2004 at 07:35:50:
In Reply to: Re: More evolution! posted by Jeroen on January 09, 2004 at 06:25:20:
Hmmmm.
Very interesting thread (more so than the one above this, which I shall delete shortly).
I agree that these mechanisms could do interesting things within the population, but my concern is that the core deficiencies wouldn't be addressed. I.e., we'd be skipping over the real problem and fixing something more esoteric.
There definitely is an apparent problem with stagnation, although I'm unsure whether it's a natural outgrowth of an evolutionary system slowly stabilizing, or something that could be remedied. Was the 2nd million years after the Cambrian explosion less interesting than the first million years?
The scheme I've been thinking about, and which has been discussed in earlier threads, is a mechanism for preserving "good genes" intact. In other words, snip points could be made less likely to occur inside of popular phrases. I'm imagining a tree structure of the poems, where snip points are more likely to occur at junctions rather than inside the branches. The more deeply nested a phrase is, the more it would be protected from snipping.
Of course, that immediately means poems couldn't be fixed length anymore. (For anyone new reading this, the problem with variable length poems is that people tend to vote for coherency, so the population tends to become dominated by shorter and shorter poems.) The solution is to weight in favor of an optimal length, which means fitness moves to floating point and voting would trigger a formula more complex than +1/-1. But that's easy.
Under this scheme popular phrases would be more likely to accumulate in single poems, and all kinds of insertions, inversions, and deletions would be possible as well.
Now back to viruses: were this mechanism in place it would be fairly easy to add a viral sub-species, because the tree structure is built on the idea of insertion points. Plus the length restriction goes away.
But even MORE interesting would be to move some of the reproductive rules into the genome, and see if viruses evolve on their own....
--David
P.S. My big concern with all of these ideas is that they might require larger populations and thus even more voting, and enthusiasm seems to have waned. A workable plan might be to:
1) Move DP to a much more robust server, one that could handle a slashdotting.
2) Implement an email list for "poem smackdown a day". I.e., sign up and get one email a day asking for a vote.
3) Ask my contact at Slashdot to post a story with the new version.
:
: : 1) A low level of unguideed random words from dictionary should be counted and dealt with as non-viral mutation during the breeding process, IMHO. To much would spoil any evolving structure, but stagnation over-phrasing (which is a subjective measurement, anyway) would be prevented with the right amount.
: : 2) If viral strings were introduced, then perhaps they should be more volotile than the standard poem entity. How about the viral process being:
: : a) Basic viral string A-B-C-D infects poem section 1-2-3-4-5-6-7-8-9-10 to create 1-2-3-A-B-C-D-8-9-10,
: : b) This poem does well (as do others with A-B-C-D, perhaps), increasing the 'fitness' of this viral string,
: : c) If, when a new viral infection is 'deemed' to occur, A-B-C-D is the one chosen from the fragments floating about, and among its hosts, the 1->10 poem is chosen as 'source' through prominent/significant success,
: : d) New 'infection' has a (very) small chance of being 3-A-B-C, A-B-C-D-8-9 or even B-C-D (we can allow transposed viral segments to vary in size, even if we're keeping the hosts constant), perhaps this creates a new viral identity along the way, creating a parallel evolution of viral strings.
: : Under this method, viruses in dead poems don't count towards fitness count. I don't think it would be constructive to have 'killer' viruses floating around that kill hosts (it completely reverses the selective pressures I envisage), so perhaps these might be consided 'symbiotes' instead of viruses. (Tok'Ra and Goa'Uld, anyone? :)
: : Going in the other direction, destructive viral infection, if used, could be tuned to allow most (say 75-90%) of the poems to be progressively immune to 'infection', and an infected version of a poor-performer is free to breed with the integrated virus incorporated indelibly (i.e. off-spring viability does not count as far as viral family fitness is concerned, it's now part of the genetic structure) and only the infected item yields 'particles' on death.
: : Sorry, too many thoughts, not enough practicality, but you're welcome to my views if you want them... :)
: I agree that this is virus design is far superior to the simple inserts I proposed. The big problem (I'm sorry, the correct phrase would be "challenge for Dave", I suppose) with this kind of scheme is the seperate virus-fitness-count you introduce.
: Every time an infected poem gets a point, you have to check them for viruses and reward them as well. This might get complicated if the virus is cut by a snip point or mutates or etc.
: Luckily we only have to have bright ideas, no solutions.