A theorem showing that in some reasonable sense, fuzzxy methods are
indeed the best way of handling numerical data is presented in the
following paper:
Robert N. Lea and Vladik Kreinovich,
``Intelligent Control Makes Sense Even Without Expert Knowledge: an
Explanation'',
{\it Reliable Computing}, 1995, Supplement (Extended Abstracts of
APIC'95: International Workshop on Applications of Interval Computations,
El Paso, TX, Febr. 23--25, 1995),
pp. 140--145.
This paper can be downloaded (in LaTeX or Postscript) from
http://cs.utep.edu/interval-comp/apictoc.html, click on the
corresponding paper.
Similar results have been proven by Kosko, you may want to look
into his latest book for results and references.
Vladik