[comp.ai] Question re: conditions of Turing/other "understanding" tests

silber@sbphy.ucsb.edu (03/10/89)

We cannot, practically speaking, exhaustively test the understanding of
a given system.  
Can "optimal" statistical criteria + knowledge domain selections
for an "understanding" test be characterized?   (a highly domain
constrained test with a limited number of interchanges can coincide
with "understanding" (over the small set of interchanges) without
actually being understanding.

bwk@mbunix.mitre.org (Barry W. Kort) (03/10/89)

In article <1321@hub.ucsb.edu> silber@sbphy.ucsb.edu writes:

 > Can "optimal" statistical criteria + knowledge domain selections
 > for an "understanding" test be characterized?

You might consider the methods employed by the Educational Testing
Service (ETS) of Princeton, NJ, and similar institutions.  These
organizations devise standard intelligence tests, in the spirit of
IQ tests, SAT exams, the Stanford-Binet tests, the Iowa Test of Basic
Skills, Michigan Multiphasic Aptitude Tests, etc.

--Barry Kort