Re: Question on grounding!

   [Sakda]    
    >I have been observing examples on DAML-S and other implementations for >a
    >while. I have some observation that I would like to ask for your 
    >opinion.

    >1)       I realized that almost all the examples available now, all the
    >input and output of any remote methods to be invoked by the web 
    >services are
    >in primitive types.

    >Setting aside the problem of interoperability of
    >serialization in each environment like java or .NET, in the semantic 
    >level,
    >how would we be able to describe complex type data like String[], 
    >DataSet,
    >ArrayList. Because the semantic and data structure of these rich data 
    >types
    >are encoded by the developers and it is extremely difficult to describe
    >semantics of these data types.

I have a feeling that what you mean by "semantics of these data types"
is not their algebraic properties, but what they denote in the real
world.  E.g., the denotation of an ISBN (International Standard Book
Number) is a particular book.  (Actually, it means more, because the
publisher is encoded as well, but ignore that.)

It may be "extremely difficult to describe" the meaning or denotation
of a name or predicate, but it may not be necessary to describe it.
Suppose that agent 1 is trying to get agent 2 to do something
involving a particular book (not an individual book, but a book in the
sense that "_Lolita_" is a book).  Agent 2 may have published a
protocol in which at one point agent 1 is to send it the ISBN of the
book in question.  All that agent 1 needs to know about ISBNs in this
context is that every book has exactly one of them, no two books have
the same one, and if the ISBN for book B is sent to agent 2, then
agent 2 will put it on North Korea's banned-book list.  (Agent 2, it
turns out, is the Dear Leader's personal secretary.)  

This is where hard-headed XML hackers freak out, because if you're
having trouble describing the semantic of ISBN's, how are you going to
describe the semantics of "North Korea's banned-book list"?  Relax.
Note that I didn't describe the semantics of ISBN's at all; I just
stated the facts that agent 1 needs to know in order to plan to send
one to agent 2.  If agent 1's purpose was to get the book banned in
North Korea, then no further facts are required.  If agent 1's goals
were more indirect, then it would have to have more information about
how its goals (ban _Lolita_ everywhere?) might be advanced by having
the book banned in North Korea.

Moral: An agent doesn't need to have the semantics of data structures
"described," let alone characterized in terms of necessary and
sufficient conditions for them to denote something or other.  It just
needs to know what will happen when it uses one in a given context.

-- 
                                   -- Drew McDermott
                                      Yale Computer Science Department

Received on Thursday, 20 November 2003 08:29:58 UTC