Librarians
Position paper for Workshop 4: AOIS at CAiSE 99
"Hiro is about to ask the Librarian whether he knows that Lagos is dead. But it's a pointless question. The Librarian knows it, but he doesn't. If he wanted to check the Library, he could find out in a few moments. But he wouldn't really retain the information. He doesn't have an independent memory. The Library is his memory, ..."
Neal Stephenson, "Snow Crash"
Just as in the citation above, where a man asks an artificial system, this view proceeds from the point that it is possible with documents lying in libraries to model cooperative problem solving behaviour in hybrid societies. We will bring artificial agents and natural agents together that they will work cooperatively to solve problems for which they use a common knowledge base - the library. Some aspects have to be considered for this: the documents, for example, must be readable from all members of the hybrid society and the members must be able to communicate with each other. In this paper we will concentrate only on the first aspect.
All current models that describe documents (e.g. Appelt (1990), Bryan (1988), Knuth (1984)) say that you can have three views on documents. At first every document has a content. This content can consist of text, pictures, sound, or sequences of pictures. The seperate contents are in the content portions at which every content portion can have only one content type. Thus a mixture of text and pictures in one document needs at least two content portions.
The second view on a document is the logical view. In this view you specify the construction of the document. By assigning the content portions to the single construction elements you generate the logical dependencies between the parts of the content.
To improve readability of a document especially for a human reader the outer form of a document is specified in the layout view. In this view you assign the layout structure to the document which is also linked via the content portions. The layout describes the representation of the document parts and the whole document itself. This view will be ignored in this paper.
To have an efficient access on documents they are collected in libraries wherein they are structured by criteria. Therefore a library is a structured collection of documents and nothing is changed in the documents itself. We can say that a library is a structured collection of document contents.
The final node in this aggregation of documents is the library. With the help of the infon logic from Devlin (1991) it is possible to describe formally the environment of cognitive agents formally and the flow of information between them. We will use the infon logic to model a less complex world than the real world. The world of documents.
Cognitive agents dwelling in document worlds get their knowledge from the document body of the library. This poses probally some problems concerning the reliability of the content. A detailed document that figures this out is Schmidt (1998), wherein this problem for the data of measurements is solved. Is the whole transfer function of a sensor producing the content known then it is sufficent to keep the denotation of the sensory data instead of the data itself. In cases where this is not possible the data must be kept and saved in the document, probably with an existing denotation.
Because we cannot expect in predictable time that a full automatically evaluation of sensory data is possible we have to deal with working groups that consist of humans and machines. In cases of a full automatically evaluation there have to exist specialized agents which are able to handle the manifold contents. In the above case the various sensory data. So we have to handle with (cognitive) agencies whose members must be able to communicate with each other. This communication shall be modelled by the exchange of documents.
The variety of agents in such agencies comes from two coordinates. A horizontal distribution of knowledge results from the various spheres of competence. This can be modeled through modularity of knowledge. From an computer science point of view there exist examinations concerning this topic that will be considered in our context (cf. Wachsmuth 1989). A vertical distribution results from the depth of knowledge. The knowledge ist finer grained at lower levels. So we have to deal with questions of granularity. This topic is not yet examined very well from a computer science view. An aid to solve this could be the documents of Hobbs (1985) and Heydrich (1995).
Appelt, Wolfgang: Dokumentenaustausch in Offenen Systemen, Springer Verlag, Berlin etc., 1990.
Bryan, Martin: SGML, Addison-Wesley, 1988.
Devlin, Keith: Logic and Information, Cambridge University Press, 1991.
Fuhr, Norbert: Information Retrieval, Skriptum zur Vorlesung, Universität Dortmund, 1997.
Heydrich, Wolfgang: Mereonymie, Preprint Universität Hamburg, Germanisches Seminar, Von-Melle-Park 6, Hamburg (o.J.); erschienen als: Zur Semantik der Teil-Ganzes-Relationen, in C. Eschenbach und W. Heydrich (Hrsg.): Parts and Wholes. Integrity and Granularity, Schriftenreihe des Graduiertenkollegs Kognitionswissenschaft Hamburg, Report 49, 1995.
Hobbs, Jerry R.: Granularity, IJCAI Proceedings of the International Joint Conference on Artificial Intelligence, p.432-435, 1985.
Knuth, Donald E.: The TEXbook, Addison-Wesley, 1984.
Schmidt, Karl-Heinrich: Wissensmedien für kognitive Agenten, Habilitationsschrift, Technische Fakultät der Universität Bielefeld, 1998.
Wachsmuth, Ipke: Zur intelligenten Organisation von Wissensbeständen in künstlichen Systemen, Habilitationschrift, Universität Osnabrück, 1989.