DAI-List Digest Wednesday, 14 August 1991 Issue Number 47 Topics: DPS vs. MA, (cont.) Cooperation, Trust Overview of MAAMAW-91 Workshop 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: Jeff Rosenschein Date: Tue, 13 Aug 91 11:45:19 +0300 Subject: DPS vs. MA, again My perspective on the difference between DPS and MA systems is that in DPS, the system designer can *depend* on agents helping one another with information, actions, etc. if that would be an effective way to build the desired system, while in Multiagent research, the system designer simply cannot *depend* on agents helping one another. It's more a matter of fundamental overall control of the interaction environment, rather than the details of who has what goals, or how helpful the agents are. In a DPS system, the agents might not help each other, but that was at some level the designer's choice. In a MA system, the agents might help each other (and might even have identical goals), but at some level that was not the designer's choice, it was not really under his control. In other words, I believe the distinction is in the options available to the agent builder at the design stage, not (necessarily) in the agent's behavior at run-time. You could happen upon a group of agents interacting, and from their goals, beliefs, and behavior, not know whether or not they were centrally designed. It's my hope that the MA perspective helps builders of DPS systems consider more autonomous, "competitive" attitudes for their centrally designed agents---when that ends up being an effective way to build the kinds of systems they want to build. --Jeff ------------------------------------------------------------------------ Date: Tue, 13 Aug 91 11:55:02 BST From: canon.co.uk!steve (Steve Marsh) Subject: Cooperation, Trust A couple of thoughts about what's been said in the DAI-List recently... > The actual technique is to consider a utility function for any >arbitrary coalition's action, consider how the coalition's payoff is >divided among members (we use the Shapley value to divide payoffs >among coalition members), then allow each agent to selfish join >whatever coalition he wants (including, of course, the coalition where >he is the only member). Perhaps this mixing of selfish and cooperative >coalition formation is a pertinent example of Ed Durfee's message >about the DPS-MA spectrum: "Most interesting problems have agents that >are partially adversarial and partially cooperative all at the same >time." This technique struck me as extremely useful for many areas in this field, not least where the concept of a clique or a group was concerned. The idea of a utility function covers it quite well - in a multiagent world, what kind of payoffs can we (as a group) obtain from letting you (as an individual) join and help us? Is more of a payoff attainable if, for example, you are a group or clique yourself (also, if 'you' are a group, will we be more inclined to 'trust' you since you already trust each other?) Cooperation to me is a function of more than one variable, not least: - Additional benefit to group of 'letting' another individual/group join - Cost of letting individual (or group) join - Benefit to potential joiner ("do I want to join this group?") - This is, if I read it right, what your work with Ran Levy concentrates on? - Benefits (to group and individual) of any previous coalitions (Thus giving us a need for the concept of identity of individuals and a memory of some sort) I'd be interested to hear how different payoff functions, etc. drive the agents (pursuers) in terms of cooperation. Steve ------------------------------------------------------------------------ From: Anand Rao Date: Wed, 14 Aug 91 17:02:46 EST Subject: Overview of MAAMAW-91 Workshop For those of you in the community who did not make it to the MAAMAW-91 workshop here is a brief workshop report: MAAMAW-91 held at Kaiserslautern from August 5-7 The Third European Workshop on Modeling Autonomous Agents and Multi-Agent Worlds was held in a wonderful place called Kaiserslautern in Germany. The conference attendance was limited to around 60 participants, which facilitated good interaction. There were 13 paper presentations, 4 invited talks, 1 panel discussion, a poster session, and 5 entrants for the Multi-Agent Olympics (system demonstrations). The papers presented at the workshop seemed to fall under three categories: (A) top-down design of various aspects of multiagent systems, (B) bottom-up emergent behaviour of multiagent systems, and (C) bridging the gap between approaches (A) and (B). A majority of the papers (8) were in Category A indicating that the community at large is still pre-occupied with top-down design. However, some interesting papers were presented in Categories B (3) and C (2). Out of the 8 papers in Category A, 4 were formal/semiformal approaches to different aspects of MA systems. The paper by Chang & Woo discusses a protocol for negotiation based on speech act theory. Rao, Georgeff & Sonenberg discuss notions of social plans and joint intentions. Osawa and Tokoro provide a model for collaborative planning. Castlefranchi, Miceli & Cesta formalize the notion of social dependence. The remaining four papers in Category A provide architectures for designing MA systems. Three of these are general-purpose architectures, while the paper by Boissier and Demazeau provides a DAI architecture for general purpose vision systems. Burmeister & Sundermeyer give an architecture for problem-solving where intentions and perception play important roles. The paper by Ferguson discusses a three-layered architecture and an experimental testbed. Collinot & Hayes-Roth descibe a satisficing algorithm for control and analyse the performance of this algorithm. In Category B, the paper by Gambardella & Haex discusses simulation of physical objects, where the global physical behaviour emerges from the interaction of agents. Drogoul & Dubreuil provide an interesting solution to the N-puzzle problem using the eco-problem-solving or emergent model. Levy & Rosenschein give a game-theoretic solution to the pursuit problem. Papers by Wavish and Kiss & Reichgelt try to integrate both the top-down and bottom-up approaches. Wavish models symbolic behaviours and shows how emergent behaviours can be integrated with them to provide a rich model of behaviours. Kiss and Reichgelt use concepts from physical dynamics to give a semantics of desires. The panel discussion was on the dynamics of knowledge and organisation in MA systems. Numaoka gave a presentation on dynamic organisations, Dragoni on belief revision in MA context and Jennings on the formalisation of joint responsibility. Interesting discussions followed the presentations and Werner summarised the different issues involved in MA systems. The first invited talk was by Rosenschein & Kraus on focal points and attempts to formalize such a notion. Kiss talked about a layered architecture for the design of MA systems. The third invited talk was by Latombe who gave an overview of robot motion planning. Gasser argued for a bottom-up design of MA systems in his talk on why DAI systems work and why they don't work. The five entrants for the MA olympics included (a) simulation of an insect (using subsumption architecture) from U. of Hamburg, (b) Distributed ATMS by DFKI, (c) Emergent behaviour in a "Sheepdog" Simulation by Philips Research Labs, (d) Eco-problem-solving model of N-puzzle by U. of Paris & CERT-ONERA, and (e) MA system based on OPS5. The best prize was awarded to Drogul & Dubreuil for their demonstration of N-puzzle. On the whole the program chairmen Demazeau and Werner & local organizers Steiner and Muller had put together a wonderful workshop. The next workshop will be held near Rome, and Castelfranchi and Werner will be the program chairmen. Anand S. Rao AAII 1 Grattan Street Carlton Victoria 3053 Australia