DAI-List Digest Wednesday, 19 June 1991 Issue Number 39 Topics: Comments on "Task Decomposition in Multiagent Systems" CFP: Logic and Change Please send submissions to DAI-List@mcc.com. Send other requests, such as changes in your e-mail address, to DAI-List-Request@mcc.com. ------------------------------------------------------------------------ From: Long Qiegang Date: Mon, 17 Jun 91 21:03:17 GMT Subject: comments to Digest 38 Hello Netters, Below is my reaction to the message under the title "Task Decomposition in Multiagent Systems." Please feel free to comment on it. ======================================================== First, it is rather difficult to make preexisting stand-alone systems (for the sake of brevity, just call them agents) cooperative. In order to do so, we have to develop ANOTHER inference engine dedicated to cooperation for each agent. Inputs to this engine must include its view to the around world and itself (the controlled KBS). Each agent, therefore, needs to model itself and its partners in an abstract way. Normally, but also inevitably, to maintain such models in a dynamically changing world will cause a large amount of communication, which may become the bottleneck of the whole system. One practical way is to predefine the static information, such as agents' capabilities in each model, and only exchange the dynamic ones such as state (busy/free), intentions (plans), etc. Second, task decomposing is not totally equivalent to task distribution in essence. One way to decompose the task may bring about several ways to distribute subtasks if agents have overlapping abilities. Furthurmore, task decomposing is more dependent on the nature of the task, while distribution is more related with the other agents' state. What I am arguing is to predefine all the possible ways of decomposing a task and INTELLIGENTLY choose one for distribution according to the current model of the environment. Actually, I have got good results by this means. Thirdly, trying to explicitly map the specific situation to a certain cooperation type perhaps is impossible, at least is not such a good way as far as I concerned. The tricky point here is WHY DO YOU COOPERATE. If one has a task that cannot be dealt with locally, it must (has no other choice unless reject the task) try to get a kind agent for help -- this is task-sharing. On the other hand, if you get some results and believe (by consulting environment model) they may be useful for me to improve my own ones, and if you indeed send those good things to me, then I can say you're trying result-sharing with me. In brief, studying the motivations for cooperation is more important, you can then answer HOW and WHEN to cooperate consequently. The last aside, I also endorse the point concerning knowledge exchange. Perhaps we could benefit from the query languages which are widely used in DBS (???). With my compliments to you all, Long Qiegang Long@mickey.feup.rccn.pt ------------------------------------------------------------------------ Date: Wed, 19 Jun 91 10:22:08 +0200 From: Remo Pareschi Subject: CFP: Logic and Change CALL FOR PARTICIPATION Workshop during GWAI'91 (German Conference of AI) Bonn, Germany September 18/19, 1991 14.30-18.30 p.m. LOGIC & CHANGE ============== Logical approaches to Artificial Intelligence have the advantage of offering a declarative framework for the representation of knowledge. On the other hand, Classical Logic, which has been the generally assumed logical theory in AI during the past decades, was originally conceived for the static world of mathematics; therefore, it is not equipped to deal with the notions of "ACTION" and "CHANGE", which are often crucial in the problems that AI is meant to address. Many of the criticisms against the "logicist" view of AI find one of their main motivations in the fact that Classical Logic even rules out the possibility of a dynamically changing world. However, recently the AI community has started to consider alternative logical theories, which are better equipped to cope with the problems mentioned above. Among such theories are Modal Logic, Temporal Logic, Girard's Linear Logic and Bibel's Linear Proofs. THE WORKSHOP WILL FOCUS ON THE TOPICS OF ACTION AND CHANGE AND THE WAY LOGIC ATTEMPTS TO DEAL WITH THEM. SPECIAL CONSIDERATION WILL BE GIVEN TO LOGICS OR LOGICAL FORMALISMS WHICH ARE NOT ONLY ABLE TO MAKE STATEMENTS ABOUT ACTIONS AND CHANGES, BUT ADDITIONALLY OFFER A MORE explicit representation OF ACTIONS AND CHANGES. Concerned fields of application are among others: Planning, Reactive Concurrent Systems, Multiagent Worlds, Non-Monotonic Reasoning, Inheritance with Exceptions, Change and Logic Programming. The following is a list of already confirmed invited talks Jean-Yves Girard (Universite Paris VII): keynote speaker. Luis Farinas del Cerro (Universite Paul Sabatier, Toulouse): Automated Hypothetical Reasoning. Bertram Fronhoefer (Technical University, Munich): Change and Linear Proofs. Andreas Herzig (Universite Paul Sabatier, Toulouse): Change and Uncertainty. Reinhold Letz (Technical University, Munich): Global Variables in Logic Programming. Dale Miller (LFCS, University of Edinburgh and University of Pennsylvania): Multiple Conclusions Logic Programming and the pi-Calculus. Luis Monteiro, Antonio Porto (Uninova, Lisboa): Actions and Deductions. Remo Pareschi (ECRC, Munich): Linear Logic and Multiagent Models of Computation. Y. Jiang, Barry Richards (Imperial College, London): A Logic of Time and Change: IQ-Prolog Extended. Jacqueline Vauzeilles (Universite Paris-Nord): Generating Plans in Linear Logic: Actions as Proofs. If you are interested in contributing a talk, please send an extended abstract (2-3 pages) to: Bertram Fronhoefer Institute of Informatics Technical University Munich Postfach 202420 D--8000 Munich 2 Germany Tel.: +49-89-2105-2031 Fax: +49-89-526502 E-mail: fronhoef@informatik.tu-muenchen.de remo@ecrc.de E-mail submissions are welcome: in case you choose to submit by E-mail, please send your abstract to BOTH the addresses above. Please inform us even in case you plan to attend the workshop without giving a talk. Every E-mail will be acknowledged. The submitted abstracts will be refereed by the organizers. DEADLINES: July 15, 1991 Deadline for submissions August 15, 1991 Notification of acceptance ORGANIZERS: Bertram Fronhoefer Technical University Munich Remo Pareschi ECRC Munich Luis Farinas del Cerro Universite Paul Sabatier, Toulouse All workshop participants also have to register for the GWAI conference and have to pay the GWAI conference fees.