> ...not able to get the complex behavior by selecting only the fittest to survive. Instead, he ended up having to kill off the least fit of the species and leave the rest to proliferate and develop interdependencies.
How is "selecting only the fittest to survive" aka killing the least fit different from "kill off the least fit of the species" aka selecting only the fittest to survive they sound like exactly the same thing thing.
I probably could have stated it more clearly like this: "selecting only the fittest ONE to survive," and "killing off the least fit ONE."
The difference was in what the composition of the surviving population was - in the extreme case where only the fittest one survives, there's only one species that makes it to the next round. In the case where the least fit one is killed off, there are n-1 species still in the mix with their relationships intact. You can of course extend it to keeping more than one of the fittest, or killing more than one of the least fit.
I was just re-reading that section of the book to try to give you a better answer and I think I must have mis-remembered his original experiment a bit (sorry about that). The original evolution model that failed to exhibit the behavior was more complex than just a survival of the fittest (he was playing with mutations and connections between species in a kind of complicated way). The part about killing off the least-fit species really was the key to getting complex behavior though, so at least that part of my original comment was right.
The eventual solution was to simplify the model so that the least-fit single species would go extinct and be replaced by a new randomly generated one each round. The fitness of a species was related to its connections, so if one species was to go extinct it could trigger others to cross the fitness threshold and go extinct as well. This led to cascades of extinctions with a power law distribution (like the pattern observed in real extinctions on Earth). The mass extinctions would eventually stabilize at a point where all species were above the fitness threshold, but the "gene pool" had been seeded with lots of new randomly generated species at that time, so there were periods of rapid mutation punctuating long stable periods where not much changed.
His book is pretty interesting, although it's sometimes a little over the top with self-congratulation. I'm reading another one now called "Self-Organized Criticality"[1] which is a much more rigorous treatment of the idea that I think will be a bit less sensational and more applicable to real problems. Bak's book is excellent if you want to get excited about making simple models do complex things though.
How is "selecting only the fittest to survive" aka killing the least fit different from "kill off the least fit of the species" aka selecting only the fittest to survive they sound like exactly the same thing thing.