home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!crdgw1!ge-dab!puma.ATL.GE.COM!puma!sjameson
- From: sjameson@fergie.dnet.ge.com (Stephen M Jameson)
- Newsgroups: comp.ai
- Subject: Re: Case-Based Reasoning as Problem Solving
- Message-ID: <SJAMESON.92Jul22085652@fergie.dnet.ge.com>
- Date: 22 Jul 92 12:56:52 GMT
- References: <1992Jul15.212602.9280@msuinfo.cl.msu.edu>
- Sender: news@puma.ATL.GE.COM (USENET News System)
- Reply-To: sjameson@atl.ge.com
- Organization: General Electric Advanced Technology Labs
- Lines: 66
- In-Reply-To: wallingf@cps.msu.edu's message of 15 Jul 92 21:26:02 GMT
-
- In article <1992Jul15.212602.9280@msuinfo.cl.msu.edu> wallingf@cps.msu.edu (Eugene Wallingford) writes:
-
- > Our disagreement concerns usage of the term "problem solving" with
- > respect to case-based reasoning. I have made the statement that
- > case-based reasoning can be viewed as a general methodology for
- > problem solving. (Of course, it can also be viewed more broadly.)
- > My colleague objects to this characterization on the grounds that
- > the use of cases from a memory -- particularly a memory that is
- > indexed automatically by the CBR system as it goes about its
- > business -- may not reflect knowledge of the domain.
-
- If the cases don't reflect knowledge of the domain, then what do they reflect?
- I think your colleague's argument may stem from the old (relatively) dispute
- about deep versus shallow knowledge. Deep knowledge consists of underlying
- knowledge about how the domain works (such as the basic theories of physics),
- while shallow knowledge consists of sets of examples which are believed to
- characterize the behavior of the domain (such as a set of examples from a
- high school physics text). Many people are uncomfortable with the claim that
- by knowing a set of examples of how the domain operates, you really "know" the
- domain, and in general they are justified. However, in many domains, it is
- either not possible to characterize the deep knowledge in a computationally
- usable way, or the computation required to make use of is it in most cases
- excessive, so one must resort to the use of shallow knowledge, with occasional
- use of deep knowledge to "interpolate" to fill in the gaps between the known
- cases (see the work of Leake).
-
-
- > That is,
- > the indexing of the memory may seem "opaque" to a knowledgeable
- > problem solver because such indexing may not conform to some
- > knowledge that the "expert" using in reaching a solution.
-
- This argument has merit only in the cases where there is no (practical)
- objective means of determining the correctness of the solution developed by the
- CBR system. This is the case in many domains, medicine in particular, where
- the only practical judge of the correctness is comparison with the opinions of
- medical "experts" [of course, you could proceed on the basis of the system's
- recommendations and see how well it works; your lawyer will love you for it].
- On the other hand, in a domain where the cost of occasionally determining
- "ground truth" for comparison, or of accepting and investigating occasional
- wrong answers is not inordinate, there is a practical objective means of
- verifying the correctness of the problem-solver, and it becomes less important
- that the CBR system solve problems the same way human experts do.
-
- In general, however, the reason why CBR has been pushed as a problem-solving
- methodology (yes, that is what it is) is that many people (such as Schank)
- believe that it conforms closely to the way humans actually solve many types of
- problems. There is considerable evidence (see the work of Koton, for example)
- that this is true in the medical domain, and it is quite clearly true in the
- legal domain (see Rissland and Ashley), and so CBR seems an appropriate
- paradigm for producing automated problem-solvers in these areas.
-
- Steve Jameson
- sjameson@fergie.dnet.ge.com
- General Electric Aerospace
- Advanced Technology Laboratories
- Moorestown, New Jersey
- ****************************************************************************
- Spring is here, spring is here,
- Life is skittles, life is beer,
- I think the loveliest time of the year
- Is the Spring,
- I do,
- Don't you?
- 'Course you do!
- -- "Poisoning Pigeons in the Park", Tom Lehrer
-