Title: Message



Yes.  So one has to pay attention to what layer or level of a system 
is 'evolving' or 'extending'. 
 
The 
interesting bit about the implementation is that its evolution independent of 
the interface 
is 
closed.  If changes in the interface don't change it (and shouldn't), then 
what is the source 
of its 
evolution?  Programmer?  Internal measurements?  On the other 
hand, the improvement 
in its 
speed does propagate to the containing system (it speeds up by some number 
determined 
by for 
instance, the number of calls to it).   This then is be 
propagated to the system within 
which 
that one is couched (think of a race condition).    The fun of 
modeling these is in setting 
up 
coupled conditions that create novel behaviors.  That is emergence.  
It isn't spooky behavior; 
just 
tedious to audit and a reason to be suspicious of data mining 
systems.
 
At the 
heart of these concepts is the notion of system and environment in which the 
environment 
is an 
evolvable system based on the activities of evolvable agents within it.  
That is the basic concept 
of 
feedback-mediated adaptation which may be random or directed.   Nota 
bene:  boundaryless 
systems never are because a system without boundaries goes to entropy at 
light speed.  This 
has 
seldom stopped any consulting group in the last twenty five years from making 
money 
on the 
concept though.  Now they will say 'permeable' and that is correct.  

 
Bottom 
up?  Maybe not.  It actually works in both directions which 
is something the so-called 
'hippies' and 'straights' had to finally cope with and maybe the 'open 
sourcers' and 
'proprietaries' will as well when the environment forces them both 
to evolve. ;-)
 
len

  
From: Chris Angus 
  [mailto:chris.angus@b...]

  I 
  think that these definitions work well in terms of an interface.  
  However, if one looks at an implementation X of an interface B, the 
  implementation might be said to 'evolve' if it changes without any change 
  to the interface B.  Such 'evolution' might be to improve performance in 
  some respect, with better algorithms surviving and poorer algorithms being 
  killed off.