Re: using SWARM to model software evolution

[glen e. p. ropella]

> Do you think its possible to use SWARM to model the evolution of a body
> of software, i.e., like leaves on a tree versions of the original trunk of
> software are modified and released, experience changes and are released
> again and again and again.  Can the versions on the outermost branches be
> considered independent agents assuming a life of their own and can SWARM
> model the tree's life?  Does this make sense?
> 
> Rick Karcich

Cool idea!  Would we want to make the leaves the agents with
trajectory histories that make up the branches?  Or would we want to
make *any* version an agent, where any parent of a given
active/fertile version may go inactive/sterile.  If each version is an
agent, then the "active" ones can reproduce (mate and/or mutate) to
form new versions.  Versions will be capable of mat/mutating several
times and going sterile at any point.

That type of framework could give rise to a version tree-like structure.
But, the problem would then be in finding the parameters that resulted
in real version trees.  Like, we know that most versions don't mate
very often (unless we're in a development where there're lots and lots
of developers).  And we know that most versions reproduce via lossy
self-reproduction (or non-random mutation).

This is a cool idea.  It would require evolving agents (via genetic
algorithms or genetic programming [I've secretly wanted to fold GP
into Swarm for a while, now....I just didn't have an app for it])
and a fitness function.

The trouble is that most sofware projects imagine themselves as being
top-down designed.  (Most of us know they're not, really...but, to
convince management of that is impossible. [grin])  And to make the
model resemble a real software project would require some instantiation
of this ill-effective top-down controller.... *in addition to* the
fitness function.

Definitely worth exploring.

glen