[FoRK] Software reuse is information theoretically hard?

Jef Allbright <jef at jefallbright.net> on Fri Nov 2 12:45:06 PDT 2007

On 11/2/07, Jeff Bone <jbone at place.org> wrote:
>
> Cf.
>
>    http://reddit.com/goto?rss=true&id=t3_5zn95
>
> ...huh.  That's interesting.  There's something deep here.  Where's
> our fully-developed theory of compositional complexity?

Thanks!  This goes to the heart of an argument, er, discussion I was
having with a friend last week about the effective extensibility of
genetic algorithms.  My point was that a particular GA can work well
within a particular problem domain, with recombination of code
corresponding to synergies already discovered, but must "reinvent
itself" itself at some point in order to continue growing (as for a
hypothetical "general" machine intelligence.)  At this transition
point, the probability of "reinvention" falls to a level similar to
the preceding one. I would expect to see the same distribution in all
evolutionary processes. My problem was how to mathematically model my
intuition, which was tending toward max entropy and Zipf`s law.

Stuff like this is why I hang out in these places.

- Jef

More information about the FoRK mailing list