roy@arcsun.UUCP.UUCP (03/13/87)
Dear abby (sigh). And I thought you had all the answers! Just to summarize the responses (wow.. so many of them too!) 1. justification mechanisms are not good enough yet, ergo expert systems do not need a justification capability. This is missing the point. Just because current justification mechanisms (jms) simply print a trace of its reasoning is not an argument against the utility of jms per se. perhaps the work on "deep model" reasoning will come up with good jms. >When they ask for "the reason why" should I have written >a huge explanation database instead of relying on the >programming language internal logic control??????? (Michaelson) No, but *when* they ask why, you should get the (current state of the art) explanation module that was keeping track of the reasoning system to spit out what it has. Pretty difficult to do in prolog unless you build some kind of es shell on top of it. 1b. Humans do not backtrack over a line of reasoning. Humans dont justify themselves. An interesting comment by B. Nevin. >....Instead, we reconstruct what such a line of reasoning > might plausibly be. It's called rationalization. To me, a doctor who says "S**t, I prescribed x... better cover myself" is one who is rationalizing his/her decisions. (but at least s/he is providing a justification for the decision (:->)) Even if the expert is reconstructing the reasoning, it is based on the knowledge of the field, and it is difficult (for me) to argue that the "rationalization" wasn't a trace of the line of reasoning since you dont have access to the reasoning in the first place. I dont ask my doctor to always explain herself, but if she was not able to when i did, i would leave pretty quickly. 2. the term "expert system" is not well defined. I couldn't agree more with this more. Three terms are often used interchangeably "expert system, rule-based system, knowledge-based system". A program that behaves as an expert (i.e. makes expert-like decisions) cannot be considered an expert system. Is SPSS (the statistical package written in fortran) an expert system.. it sure performs functions similar to an expert statistician (relative to me, anyway). A program that only has a clear knowledge/control separation cannot be called an es. any system written on top of a spreadsheet has a clear knowledge/control separation. >Knowledge-based system technology is a programming methodology, which >facilitates the incorporation of "human or expert" knowledge. Hence, the >criterion that explanation facilitiy is a must for a knowledge based >system (or an expert system once you add the expert's knowledge) is >to be questioned. [...users don't like rule printouts, they like >"a more robust ENGLISH translation and "nice graphics" (Sriram) I dont see how your (pretty broad) definition of a knowledge-based system negates the need for an explanation facility (if kb-system in your reality == expert system). The second comment simply supports my view (cf 1) >...it seems to me that disputes over whether explanation is "needed" >before you can call it an expert system are missing the point... (Coffee) Wish I had said that. 3. depends on what the es will be used for. es will be more accepted if they have an explanation facility. I guess when i think of expert systems' use, i usually think in terms of it being used as a consultant or advisor (cf "our expert is overworked, and getting old" stories). Using an "expert system" in "production" seems analogous to human experts writing a set of instructions for use when they are not available. Would consulting the set of instructions constitute a session with the expert? Putting a justification mechanism if/when needed is another way of saying that the facility is a "luxury" and not really necessary. I think that perhaps I have a very tight view of the term "expert system" and its use. Thanks for the feedback, roy