RE: Using Physical Descriptors

["Bullard, Claude L (Len)" <clbullar@ingr.com>]

We must be sure to note that analogical reasoning 
that proceeds by surface similarity is prone to errors (acquisition 
of superstitious rules) and that anytime analogical means are 
used, the learning (eg, new productions created and posted) 
is suspect and is always verified.  That is applying a 
production acquired by analogy requires a test to verify 
that the production results in the expected state.

For example, using the suface characteristics often 
grouped by "race" can result in a quick sort of some 
set of candidates.  It is not useful for determining 
further actions and in more cases than one should be 
comfortable with, can result in misidentification.

Our models must be critiqued in use as well as design.

Because deliberate analogical reasoning proceeds by looking for a 
production from a member of the same class, it is reasonable. 
Trying to apply a production from a different class, even 
one with surface similarity is a much less reasonable operation. 

Metaphorical classes are particularly thorny, so use of and 
evaluation of tropic statements should be limited in cases 
where the results of choosing a production are in some way, 
critical.  On the other hand, since feint often depends on 
misidentification, testing the object by asking it to 
perform productions of a similar class is not out of 
the question as a strategy for determining if isolating 
it is the next best move.  This emphasizes the importance 
of rapid and relevant feedback on the results of applying 
productions.

Len 
http://www.mp3.com/LenBullard

Ekam sat.h, Vipraah bahudhaa vadanti.
Daamyata. Datta. Dayadhvam.h