Suppose that someone got an AI program working so that it was able to
learn a rudimentary natural language. There is no reason why that
program should be expected to understand everything. That would not
make sense. Imagining that this program had only learned a little, it
would not be able to disambiguate every phrase that might be thrown at
it. So the test that it should be able to disambiguate anything is not
a valid test. A qualified test then, is, after it was given some
instruction in different subjects where some common word and words
were used in different ways, could it 'understand' these different
uses of the words as they pertain to the different subject domains?
This qualified test would be difficult for most children but the
ability to follow a discussion using terms that the child had learned
should turn up some confirmatory evidence that the child had learned
something about that subject. And with more practice the child would
become better skilled at disambiguating the phrases that it had been
taught.  This kind of testing then can stand as the disambiguation
test for innovative language-based AI projects in our age.
Jim Bromer


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to