On 9/15/00, at 5:33 AM, James Rogers wrote:
>I am not sure that GA is actually useful for most classes of business
>problems, or at least, I think there are methods that give nearly as good
>results in a more deterministic fashion under the constraints of what
>businesses want. While GA sounds good for offline, non-realtime problem
>solving domains, it has never seemed well-suited for environments where
>one needs to adapt to rapidly changing conditions in realtime. Also,
>businesses are really uncomfortable replacing humans with machines in
>these spaces if the business algorithms cannot be explained in excruciating
>detail, which they can with people and plain old software (POS :^).
The same considerations you discuss are relevant to many non-business
problems. I'm thinking of medical diagnostic software in particular.
No doctor (and presumably some patients) is going to trust a diagnostic
program that cannot clearly explain its reasoning.
The AI community realized this early on, and it's a standard feature in
research and commercial expert systems.
On the other hand, people make megabuck decisions every day based on
data mining techniques -- including neural nets and genetic programming --
that are known to work, but may be hard to understand or seem counter-intuitive.
One approach managers can use to sleep better at night is to not to rely on
a result unless it's been confirmed with two different model types. Of
course, this nearly doubles the mining cost.... TANSTAAFL.
-- David Lubkin.
______________________________________________________________________________
lubkin@unreasonable.com || Unreasonable Software, Inc. || www.unreasonable.com
a trademark of USI:
> > > > > B e u n r e a s o n a b l e .
______________________________________________________________________________
This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:38:06 MDT