Erann Gat <gat@jpl.nasa.gov> wrote:
+---------------
| Let me tell you a story.
| When I was a graduate student I did my master's thesis on something called
| "the problem of referential transparency".  It has to do with avoiding
| unwanted conclusions in AI systems based on formal logic.  The classic
| example is, "Mary knows Bill's phone number" and "Bill's phone number is
| the same as John's phone number."  From this one does not want to conclude
| that Mary knows John's phone number, but under standard formal logic this
| conclusion is unavoidable if one models these two statements as:
| 
| knows(Mary, phone-of(Bill))
| phone-of(Bill) = phone-of(John)
+---------------

It seems obvious that those two statements are in fact *not* a
good model of the situation of interest. They should be instead:

	knows(Mary, belongs-to(phone-of(Bill), Bill))
and:
	phone-of(Bill) = phone-of(John)

While from those two you *can* conclude that:

	knows(Mary, belongs-to(phone-of(John), Bill))

you cannot conclude that:

	knows(Mary, belongs-to(phone-of(John), John))
nor:
	knows(Mary, belongs-to(phone-of(Bill), John))
nor:
	knows(Mary, phone-of(Bill) = phone-of(John))


-Rob

-----
Rob Warnock, PP-ASEL-IA		<rpw3@rpw3.org>
627 26th Avenue			<URL:http://rpw3.org/>
San Mateo, CA 94403		(650)572-2607