Re: Wolfram Alpha

Bob and Enrico,

Bob, it's good to hear from you again.  I remember that project
very well, but we were really doing knowledge representation
and AI -- and so are the Powerset gang.  I agree with Enrico
that Wolfram Alpha is doing something simpler that could be
called "low-hanging fruit":

EF> It can be understood that the Mathematica theorem prover is
 > not connected in the current version of Wolfram Alpha. Anyway,
 > even when it will be connected, we are talking here about
 > equational reasoning over real numbers, not really KR stuff.

Stephen Wolfram explicitly said Mathematica is the basis for
the Alpha project.  But he also said that they're not doing AI,
at least not in the current version.  See his blog:

    http://blog.wolfram.com/2009/03/05/wolframalpha-is-coming/

Following is his short summary of their approach:

SW> Of course, getting computers to deal with natural language
 > has turned out to be incredibly difficult. And for example
 > we're still very far away from having computers systematically
 > understand large volumes of natural language text on the web.
 >
 > But if one's already made knowledge computable, one doesn’t need
 > to do that kind of natural language understanding.  All one needs
 > to be able to do is to take questions people ask in natural language,
 > and represent them in a precise form that fits into the computations
 > one can do.

I'm sure that much of that computation includes equational reasoning
with numbers.  But he also says

SW> I wasn't at all sure it was going to work. But I'm happy to say
 > that with a mixture of many clever algorithms and heuristics, lots
 > of linguistic discovery and linguistic curation, and what probably
 > amount to some serious theoretical breakthroughs, we're actually
 > managing to make it work.
 >
 > Pulling all of this together to create a true computational knowledge
 > engine is a very difficult task.  It's certainly the most complex
 > project I've ever undertaken. Involving far more kinds of expertise
 > -- and more moving parts -- than I've ever had to assemble before.
 > And -- like Mathematica, or NKS -- the project will never be finished.

Since Mathematica supports symbolic manipulation of nearly every
major mathematical system and since it also has a logic-programming
engine as the base processor, I suspect that they've implemented
a wide range of methods -- including what is usually called a
deductive database.  The borderline between that and what is
called knowledge representation is porous or nonexistent.

RS> You may remember that this is the sort of approach IBM ...
 > used in the mid 1990's...

That approach was closer to Powerset than to Wolfram Alpha:

  1. We used a parser that had been developed by Michael McCord
     (who retired from IBM a while ago, and his parser was released
     on IBM AlphaWorks).

  2. We were using conceptual graphs as the knowledge representation,
     and the IBM-CSLI verb ontology was based on the set of conceptual
     relations in the appendix of my 1984 book and extended for the
     2000 book.

  3. We were participating in the ANSI workshops on ontology and
     the Heidelberg workshop in 1998, which included a significant
     subset of the cast of characters that have been active in
     ontology circles for the past decade.

RS> IBM failed to see the value of the project and canceled it.
 > I had worked for IBM for thirty years, thought that was enough
 > and retired.  All of the people who worked on the project
 > eventually left IBM....  Martin along with Livia Polanyi were
 > original employees at PowerSet which was recently bought by Microsoft.

I am very well aware of all those developments.  And given all the
multimillions of dollars that Xerox, the venture capitalists, and
now Microsoft have put into what has become the Powerset technology,
one could say that IBM was justified in canceling the project.

I certainly wish the Powerset people well, and I'm happy for them
that they still have a job.  But I suspect that Wolfram will turn
a profit from their approach sooner than Powerset (now MSFT) will.

John


---------------------------------------------------------------------
To unsubscribe, e-mail: cg-unsubscribe@...
For additional commands, e-mail: cg-help@...