# KEHOME/knowledge/applications/SemanticWeb/OpenCyc/SeriousProblems.html
# Sep/1/2003
Serious problems with Cyc knowledge structure
Serious problems with Cyc knowledge structure
Based on a few days of using MKE to browse
the OpenCyc knowledge base,
I have identified some key problems
which may prevent OpenCyc from becoming
a useful tool for ordinary people.
- too many properties converted to classes
- kb is "cluttered" with "internal" Cyc properties
- hard to find Cyc word for English concept
- hard to find Cyc context for questions/statements
- some property lists are encoded as executable functions
- need an English-like user interface
too many properties converted to classes
Many AI researchers say it's good, and necessary for efficient inferencing.
My intuition says it's bad, and promotes inefficient inferencing.
These "extra" classes should be generated
only when/if the need (context) arises.
kb is "cluttered" with "internal" Cyc properties
It's hard to find ordinary, useful properties.
Cyc properties should be segregated in a separate branch of the hierarchy.
Even with my current, smaller MKE KBs, this is a serious practical problem.
Cyc is currently a great tool for AI researchers,
but bad for ordinary people.
hard to find Cyc word for English concept
Will WordNet really help?
hard to find right Cyc context for questions/statements
I'd like to hear about your new algoritms for finding the right context.
some property lists are encoded as executable functions
I guess this promotes efficiency, but is hardly suitable for ordinary people.
need an English-like user interface
Should have Natural Language and/or MKR interface.