From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Wed Sep 23 1998 - 16:29:23 MDT
Robin Hanson wrote:
>
> and the key assumption is that "the phenotypic optimum changes suddenly and
> then remains fixed during the bout of adaptation studied." The paper then
> models a hill-climbing search to a peak. In such searches, one makes big
> moves early on, and then smaller and smaller moves as one approaches the
> peak.
>
> As genes get fixed along the way, the genotype ends up reflecting this
> distribution of moves; some genes embody very large moves and others
> very small moves. The paper notes that you don't get this effect "if the
> phenotypic optimum perpetually drifts, moving away from the population at
> about the same rate that the population evolves to `keep up.'"
>
> The bottom line is that this paper *assumes* punctuated equilibrium,
> and so is not evidence in favor of it.
Punctuated equilibrium is something that needs to be explained because of
geological evidence. Postulating it is not a problem. In fact, I would say
the major problem with punctuated equilibrium is explaining it to the
loony-toon creationists.
As I understand it, the innovation is in explaining the punctuated equilibrium
in terms of an exponentially distributed mixture of large and small tweaks.
The default explanation is large environmental shifts that increase the
selectivity pressures. Another interesting explanation is a continually
churning pool of small mutations with some synergetic combinations. The pool
keeps churning until the mutation has a sufficiently large chance of meeting
up with its synergetic "mate", and then the mutation rapidly becomes dominant
throughout the pool. (This one is also based on computer simulations; the key
percentage is 14% of the population, or the reciprocal of e squared.)
Personally, I think that all three explanations describe real phenomena.
Punctuated equilibrium, as the fossil-record phenomenon, neither requires nor
prohibits single big-win mutations. It is not even necessarily relevant to
the trajectory of a seed AI, since most theories have nothing to do with the
shape of the "hill" one is attempting to climb, or revolve around resistance
barriers that aren't relevant to intelligent design. I noticed that paper
because it explained punctuated equilibrium in a way that exported fairly well
to breakthrough/bottleneck AI trajectories. It was the first explanation I
had seen with that property.
--
sentience@pobox.com Eliezer S. Yudkowsky
http://pobox.com/~sentience/AI_design.temp.html
http://pobox.com/~sentience/sing_analysis.html
Disclaimer: Unless otherwise specified, I'm not telling you
everything I think I know.
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:49:36 MST