About 10 years ago I read a book by John Koza 
(http://www.genetic-programming.org/gpbook1toc.html) concerning a topic 
such as this.  I remember purchasing several VHS video tapes showing 
examples.  John was doing demos using LISP where the solution was 
derived using natural selection.  The whole concept of programming 
computers to solve problems by natural selection is a neat idea.  I 
happened to mention to a colleague recently that some developers work 
this way unknowingly.  They create a complex complete unit test and 
then keep trying different "solutions" to get the test to pass without 
thinking about the domain.  I actually like spending time developing a 
complete unit test first.  It helps me to understand the problem.

Not sure where the Koza tapes are now or if they are available today.

  - Steve

On Friday, August 15, 2003, at 01:02 PM, Alan Kay wrote:

> Hi Folks --
>
> An idea that has surfaced numerous times in various ways over the last 
> 40 years can be stated as a question:
>
> If unit tests are a good idea, then shouldn't we try to generate the 
> method code directly from them?
>
> In other words, there could be/should be a language in which one 
> programs in terms of the criteria to be achieved, and the system tries 
> to come up with code that meets those criteria. (The last one of these 
> that seemed really interesting was ART by Inference Corp (out of the 
> CMU context).)
>
> Anyone know of any interesting current work along these lines?
>
> Cheers,
>
> Alan
> -- 
>
>
--
"The real romance is out ahead and yet to come. The computer revolution 
hasn't started yet. Don't be misled by the enormous flow of money into 
bad defacto standards for unsophisticated buyers using poor adaptations 
of incomplete ideas."  -Alan Kay