Cyc was originally intended to be a repository of the kind of
information that would normally be found in an encyclopedia. McCarthy
challenged any encyclopedia knowledge base to say whether Wellington
heard of Napoleon's death-- and later whether Napoleon heard of
Wellington's death.
.
That Napoleon died in 1821 and Wellington in 1852 and that Napoleon was the Britain's most important prisoner when he died, and Wellington was a British high official is in any encyclopedia. Common sense can then answer the questions, but the answers are not in the encyclopedia.
These considerations persuaded Lenat to aim his encyclopedic effort at common sense itself. This issue is of great importance to the HPKB project. A successful knowledge base must contain far more than the specific facts about the problem domain, without an understanding, and representation of the common sense knowledge required, no non-trivial question can be answered.
On a later occasion, Cyc was challenged by questions about setting the table for dinner. It was asked how many spoons would be needed for three people to have soup for lunch. It turned out that the formalism chosen for the problem could not be readily modified to know that only one table was required but that each diner would require his own spoon. It isn't clear whether this is a deficiency of Cyc or just a deficiency in how the Cyc team had implemented the facts about dining. When Cyc was programmed to correctly answer that three spoons were necessary, for three people to eat soup, it also answered that three tables were necessary. The problem here was that the defaults that were entered were too general.
There is a great temptation to write general purpose defaults, that cover a large number of cases. Without this generality defaults are not useful, however, writing defaults is a dangerous business, fraught with problems. Our expertise with the above systems will enable us to make suggestions for challenges that will catch such overly general defaults.