As was explicitly stated in the introduction to theSIGART Newsletter, the purpose of theissue was to provide a forum within which researchers and practitionersof knowledge elicitation could interact, comment, and report on thecurrent techniques and tools being employed at that time. Dr. Fuldamakes a bold statement when he says, “…the newsletter’scontributions to the literature on knowledge elicitation are not anymore helpful than the rest of the literature in the brand-new field:that is, not at all.” Hiscomments, as intended, do not only apply to the special issue but toevery conference, journal, and book created on the subject of knowledgeacquisition.
According to Dr.Fulda’s beliefs, the otherjournals identified in his review, including theInternational Journal of Man-MachineStudies, KnowledgeAcquisition, Data and KnowledgeEngineering, and the IEEETransactions on Knowledge and Data Engineering, arealso worthless contributions to theadvancement of intelligent systems. I find this hard to believe, as domany of my colleagues. Fulda’s statement is also indirectly targeted atthe industries, academic institutions, and government offices that havededicated (or wasted, in his vision)their resources to developing the systems defined in thesepublications.
The actual number of special issues printed was increased toaccommodate the large volume of anticipated back order requests. Lastcheck with the ACM order department showed that 50 percent of theinventory has been depleted. Therefore, his opinions about this issueand recommendations for readership are not supportive of the generalconsensus of the research community, which leads me to question hisintent in reviewing a publication two years after the releasedate.
The special issue was partitioned into two sections and abibliography to allow a sampling of knowledge elicitation research frommore than one format. The first section consisted of medium-sizedmanuscripts to provide focused discussions of certain applications,tools, or techniques. Space was limited and the length of the papers wasrestricted to between five and ten pages. Much of the material presentedin this section was original (as was evident during the review process)and there were several papers notaccepted because they were published elsewhere. Dr.Fulda’s comment that the “contributors offer little beyond whathas already been published” is an oversight on his behalf and canbe discounted due to the large amount of original material presented inthe issue.
The “short pieces” to which Fulda referred made up thesecond part of the issue and were referenced as special research notes.These notes were a compilation of extended abstracts, informativestatements, and related summaries describing various knowledgeacquisition work. Their purpose was to illustrate the diversity ofapplications and the various types of approaches the community wasus ing. Granted, the articles could not disclose much detail about thereported research; however, they were meant (as stated) to initiatedialogue among those researchers working within similar beliefstructures. From the correspondence I received after publication, it wasmy interpretation that the layout of this section worked quitewell.
Dr.Fulda’s review lost focus on the specialissue and started concentrating on the faults of the knowledgeacquisition community. He over-generalized when he commented thatcomputer scientists do not think in terms of interpersonal tasks. Thisstatement may be correct in a different context, but to those involvedin the development of intelligent and expertsystems--to which this work wasaddressed--it reveals a shortcoming in hisunderstanding of their capabilities. The review further downplayed thefield of knowledge acquisition when Fulda stated that it “is stillat square one.” Knowledge acquisition, in general, is fairly rougharound the edges, and researchers all try to smooth out these edges byproposing new methods, learning from past mistakes, and publishing theirresults; hence we call them researchers.
As Fulda’s discussion of knowledge acquisition progressed, itmatter-of-factly introduced “a well-established branch ofAI” that is called computer induction. He has failed, however, toprovide a definition of the limitations and goals for understanding whatcomputer induction is and how it may be compared with knowledgeengineering. He has also not stated whether the computer learns from afixed and standardized store of information (such as a database) or froma highly dynamic and unstructured entity such as a human being. It istrue that automated knowledge acquisition systems have been advancingquite rapidly for some time now and, as Fulda should know, many of thesesystems were originally developed using person-to-person interviewingmethodologies. Due to the vast amount of paper that can be generated bythese traditional techniques, the proper management of this informationis best suited to computerized procedures.
Dr. Fulda should give more consideration to identifying a flaw orimproperly applied truism in a particular approach or model for reducingthe knowledge acquisition bottleneck, rather than universallydepreciating the efforts of others trying to do so. I hope he enjoysreading the book Readings in knowledgeacquisition: current practices and trends [1] as muchas he has the special issue.
