jberger@asterix.drev.dnd.ca (Jean Berger) (11/07/90)
Would someone help me in clarifying in an unambiguous manner the real difference between two kind of expert tasks namely: - Data Interpretation and - Diagnosis By unambiguous manner I mean an easy way to PARTITION tasks domain concerning respectively Diagnosis and Data Interpretation in a unique manner. Sometimes it might be hard to distinguish Data Interpretation from Diagnosis tasks, what are the basic fundamental elements discriminating each other. Hope that may help some people to discriminate clearly between the two expert tasks. Thank you. -- Great Minds Think Alike ======================================================================== jean berger Eng. Defence Research Establishment Command and Control Division Combat System Section C.P. 8800 Courcelette, Quebec Canada G0A - 1R0 PHONE: (418) 844-4645 FAX: (418) 844-4538 E-MAIL: ARPA: jeanb@quebec.drev.dnd.ca =========================================================================
byland@iris.cis.ohio-state.edu (Tom Bylander) (11/09/90)
In article <1990Nov6.184658.29097@asterix.drev.dnd.ca> jberger@asterix.drev.dnd.ca (Jean Berger) writes: > > Would someone help me in clarifying in an unambiguous manner > the real difference between two kind of expert tasks namely: > > - Data Interpretation and > - Diagnosis I would say that diagnosis is a subclass of data interpretation, i.e., a fault hypothesis is an interpretation of the observations. Diagnostic reasoning, however, is often associated with data gathering and test generation, both of which are probably considered separate from data interpretation. Data interpretation is also closely related to abduction, which is finding explanations for data. Perhaps it would be best to consider explanations as a subclass of interpretations, making abduction a subclass of data interpretation. So I would roughly put the relationship this way. Abduction is data interpretation in which the interpretation is an explanation of the data. Diagnosis is abduction where the data are abnormal observations and the explanation is a (composite) fault hypothesis. Tom Bylander byland@cis.ohio-state.edu
sticklen@cps.msu.edu (Jon Sticklen) (11/10/90)
From article <85816@tut.cis.ohio-state.edu>, by byland@iris.cis.ohio-state.edu (Tom Bylander): > In article <1990Nov6.184658.29097@asterix.drev.dnd.ca> jberger@asterix.drev.dnd.ca (Jean Berger) writes: >> >> Would someone help me in clarifying in an unambiguous manner >> the real difference between two kind of expert tasks namely: >> >> - Data Interpretation and >> - Diagnosis > I think that Tom Bylander captured most of what I would say to this question too. But I would add one other discriminant between "data interpretation" and "diagnosis." Diagnosis should not be considered as an isolated task. The bottom line when we do diagnostic problem solving is not that we want to find out "what is wrong" but rather that we want to "fix what is wrong." Diagnosis is problem solving aimed at characterizing what is wrong, but the output of diagnositic problem solving is then used to fix what is wrong. If we are doing classification-based diagnosis, then the diagnositic categories for any real world problem will surely have therapy utility. In fact, that is what diagnostic categories are - equivalence classes which map situations into appropriate therapies. So at root I agree with Bylander that diagnosis is a subset of data interpretation. What I am adding is why its a subset. And that why is that the output of diagnostic problem solving must have utility for fixing something that is broken. ---jon--- Jon Sticklen AI/KBS Lab - CPS Dept Michigan State University
mtanner@gmuvax2.gmu.edu (Michael C. Tanner) (11/13/90)
This is mostly a response to <byland@cis.ohio-state.edu> and <sticklen@cpswh.cps.msu.edu>. I wish somebody would define data interpretation. You normally have to interpret data in order to do diagnosis. You may choose to view the result of diagnosis as an interpretation of data. But unless there's some technical definition of data interpretation that I'm missing, I don't think one is a subset of the other. Both are really ways of "finding out what's true". I don't think that's the same as "inference to the best explanation" (IBE) (similar to abduction). If you have the prior assumption that there's something wrong, you might attempt to find out what's true (i.e., what's wrong) by diagnosis, which you may choose to view as a problem of inferring the best explanation. So I disagree with Bylander when he says diagnosis is a kind of data interpretation and data interpretation is abduction. But I agree with Sticklen (and I think Bylander too) that diagnosis carries the explicit assumption that something is wrong and the diagnostic answer will say what's wrong. IBE, and data interpretation, and (especially) finding out what's true, are too ill-specified to be of any direct use. All problem solving is done with some purpose, and that purpose introduces some pragmatic considerations. I don't think you can just interpret data. You have to interpret it with some use in mind for the result. Diagnosis is a possible use, i.e., you might interpret data in order to use it for diagnosis. Diagnosis intoduces the consideration that something is wrong and acceptable answers will specify faults and perhaps explain how the faults produce the misbehavior that made you aware of the problem. And as Sticklen points out, the goal of treatment or repair (a subgoal of which might be diagnosis) introduces the consideration that the diagnostic answers will relate to possible treatments. So my answer (modulo the real meaning of data interpretation) is: diagnosis is a kind of problem solving and data interpretation isn't. -- Michael C. Tanner Assistant Professor CS Dept AI Center George Mason Univ. Email: tanner@aic.gmu.edu Fairfax, VA 22030 Phone: (703) 764-6487