On Mon, 14 Mar 2005, Dennis Gorelik wrote:

> From my point of view CYC in on the same level of intelligence as
> MS Word. Well, probably MS Word is even more intelligent.
> At least MS Word works and produce nice and intelligent results (not
> super-intelligent though).
> Does CYC have any practical use at all?

Speaking with my own opinion, and not that of my employer Cycorp, I would 
say that the Cyc approach to AI at a meta level is the same as that taken 
by other individuals and groups creating an AI.  The approach is to 
identify the hard AI problem first and solve it, then move on to the rest 
of the required behaviors.  In the mid-1980's Doug Lenat identified 
commonsense knowledge as the hard AI problem, that once solved would 
enable otherwise brittle domain-specific expert systems to work together 
robustly and be easily extendible.  Cyc's commonsense knowledge is 
organized into an ontology suitable for symbolically representing the full 
range of human thoughts.  Cyc's current behavior is fact entry and 
question answering, with additional web based tools for rule creation and 
knowledge base browsing.  The recent release of OpenCyc 0.9 contains three 
times as many facts as the prior release.  A new product, ResearchCyc, is 
available for academic/commercial research (without a fee), that contains 
even more content (e.g. the lexicon) and the full set of tools.

Presently the greatest practical use of Cyc in my opinion is it's 
ontology, which may be applied to organize concepts and relations in any 
application.  We have provided specialized ontology exports to government 
entities for a few years, and have recently published the entire OpenCyc 
ontology in OWL (the Web Ontology Language) format on our web site.

Our government sponsors have provisions for Cyc applications to be tested 
this year with user organizations, so that will be a test of practical 
use.

Beyond question answering, a wider range of intelligent behavior is 
possible.  My own ambition for example, is to enable Cyc to be curious, to 
take the initiative, to seek required knowledge and improve its own 
behavior using a hierarchical goal-oriented command structure.

Dennis, I have toured your web site and have the following points about 
features that Cyc has that you might have to eventually incorporate into 
your own architecture:

1. Distinction between individuals (CityOfAtlantaGA) and types of 
individuals (USCity)

2. Type taxonomy of concepts so that specific types have one or more 
supertypes (including a taxonomy of relationships) 

3. Context.  This is the ability to group assertions that have shared 
assumptions, and to exclude wrong contexts when answering questions  

4. Meta assertions, in which a concept in a relationship is itself a 
relationship between other concepts

5. First order rules - "every bat has two wings", in which quantified 
variables concisely represent what would otherwise be a large number of 
ground facts

6. Complex objects, notably events and situations.  Cyc treats these 
according to the philosophy of Donald Davidson, all actions are rich 
events that entail various actors (e.g. a robbery event has a robbed 
entity even if you are only told about the robber).  Cyc has a rich 
vocabulary of actor roles for thousands of situation and event types.

6. Truth maintenance.  This is Cyc's ability to automatically retract all 
the associated facts when a concept is removed from the knowledge base

7.  Lexicon.  This the association of natural language (mostly English) 
words and multi-word strings with concepts, and in the case of event and 
situation concepts the lexicon contains verb frames (e.g. how are the 
sentence subject and object related?)

8.  Inference.  Forwards deductive inference automatically asserts new 
derived facts when certain antecedent facts are asserted (using marked 
forwards-firing implication rules).  Backwards deductive inference uses 
rule back-chaining to answer queries when the ground facts cannot be 
simply looked-up.  There are numerous special cases to optimize.  For 
example many relationships are transitive (greaterThan).

8a.  Temporal inference - the ability to reason about time intervals, time 
points (e.g. What World War II movies starred John Wayne?)

8b.  Modal inference - represent and answer questions about assertions 
with modal logical operators, e.g. "John ought to get enough sleep."

8c.  Abduction - when directed, hypothesize the best answers to queries

Cheers.
-Steve

-- 
===========================================================
Stephen L. Reed                  phone:  512.342.4036
Cycorp, Suite 100                  fax:  512.342.4040
3721 Executive Center Drive      email:  [EMAIL PROTECTED]
Austin, TX 78731                   web:  http://www.cyc.com
         download OpenCyc at http://www.opencyc.org
===========================================================

-------
To unsubscribe, change your address, or temporarily deactivate your 
subscription, 
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]

Reply via email to