Bill Tozier wrote a very interesting article on how to be successful with evolutionary algorithms and machine learning, and how it applies to ExtremeProgramming.
http://groups.yahoo.com/group/extremeprogramming/message/45678
Excerpts:
"At every company I've ever visited, early on during my time there I have met a person in a position of decision-making power (call them the Expert) who says to me, 'Oh, one time I tried using <XXX>, but they don't work.'"
See, I think many modern folks are used to thinking in terms of general-purpose tools, and the consequent ability to apply them liberally anywhere and always. They misunderstand the words "general purpose" to mean just that: good for what ails ye. But personally I think it's a fallacy to believe that what makes a tool "general-purpose" is its ability to work, without adjustment or adaptation, in a variety of settings. That, in fact, is what's provably wrong and misunderstood with the notion as it applies to the sordid world of machine learning.
What makes a tool -- or any approach to solving problems [What difference, really?] -- "general-purpose" is the ease with which one can:
- examine the tool's structure
- understand the reason for that structure in getting you to the goal
- make changes to that structure to cope with special circumstances
In short, adapt the tool.
(Just go read the whole thing, there's too much good stuff to quote)
I'm surprised this is posted under the heading of an "EmergentProcess". I don't see how learning to use a tool is an emergent process. Elaboration might help me understand how this is a good name for this page.
Careful reading of the link and the material and the realization that all tools are not tools producing emergent artifacts might help you.
Related: ToolsProducingArtifact