Views on Krep
tony@ontek.com (Tony Sarris)
Date: Fri, 3 Dec 93 13:33:21 PST
From: tony@ontek.com (Tony Sarris)
Message-id: <9312032133.AA12147@ontek.com>
To: cg@cs.umn.edu, interlingua@ISI.EDU
Subject: Views on Krep
What a wonderfully written -- and I think quite accurate -- identification
and description by Pat Hayes of what I also believe are two schools of
thought on the subject of Krep. I don't believe either school represents the
ONLY way to do meaningful knowledge representation, but we at Ontek
certainly lean toward the broader and deeper representation of knowledge,
particularly so that someday automated agents can be used to help in
interpreting knowledge and planning actions to be taken, not merely in
exchanging knowledge that depends almost exclusively on human agents
for its interpretation and use.
There may arguably be some cases where exchanging individual 'otologies'
for interpretation and use by someone (specifically some HUMAN) on the
other end is perfectly adequate. In some cases a computer can process
[traditional] program-like functions specified in the 'ontology' and use the
results of that processing for some useful purpose. But I also agree that
that's not really 'knowledge representation' -- at least not in the sense of
having an artificial intelligence. This approach is a step in the right
direction, and may provide some real, short-term benefits. But let's not
pretend that it's the long-term solution to the many knowledge acquisition
and exploitation bottlenecks that Lenat and others have studied as part of
CYC.
The problem is that to accomplish much in this world, you not only have to
exchange information and understand what the other party means, you have
to come to some sort of agreement about what to do with the information,
i.e., some action to take as a result of analyzing the information. If you
think of the exchanging of 'otologies' as some sort of automated version of
Wittgenstein's language games, then all we're doing is providing some sort
of faster, more efficient way for us to all have our own little world views
and say "fine, YOU do what YOU want". What do we do when we have as
many (if not more) computerized agents than people, all off doing their
own thing and largely depending on humans to act as mediators? I guess
that's one vision of distributed systems, but it's not mine.
One of the challenges facing many of our customers is enterprise
integration. Getting engineering and manufacturing to exchange their
'otologies' of PART, for example, might be useful. What would really be
useful though would be reconciling those disparate views so that we
humans, and maybe even certain automated engineering and manufacturing
applications, understand the nature of the differences between the two
views and why they are the way they are. This level of knowledge
representation depends on the ability to represent deep and commonsense
knowledge about domains such as engineering and manufacturing, and what
it means to be PARTS, MATERIALS, FUNCTIONS, FORMS, MACHINES,
PROCESSES, DESIGNS, SCHEDULES, etc.
I don't think you have to represent everything about all things for all time
to do anything [an excuse I've heard used by several people who would
rather do nothing!]. But work desperately needs to be conducted in the area
of basic otologies. There are still always going to be some differences in
viewpoints (and different associated 'otologies') -- probably all the way
down to the lowest levels. The more common, generic knowledge we
represent, though, the easier it is to create specializations or at least
explications for the differences. In other words, it is also possible then to
make sense of one view in relation to some other view.
I actually think Fritz Lehman advocates this approach, too, and I think that
might have been his starting basis for criticizing all the emphasis placed
on 1) syntax (and the 'syntax wars' that rage), 2) building numerous ad hoc,
and unrelated otologies without shared foundations, and 3)
communicating agents that get a little too autonomous for their own good
(i.e., those that seem to take the old Burger King adage "have it your way"
to the extreme).
In any case, way to go Pat Hayes! Perhaps you'd consider writing
something up (formally) on this very topic?
Tony Sarris
tony@ontek.com