DAI-List Digest Friday, 30 August 1991 Issue Number 51 Topics: DPS vs. MAS More on DPS vs. MA Systems 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: Prof. Misbah Deen Date: Thu, 22 Aug 91 14:51:15+010 Subject: DPS vs MAS We (Keele) wish to express our appreciation to all those who have contributed to the debate on DPS vs. MAS, originally initiated by Farhad Haidari of Keele. Having seen the contributions so far, we have formed the following opinion. Comments on this will be welcome. Pure DPS and Pure MAS seem to lie at the two extremes of the same spectrum. A pure DPS is a special purpose system, built typically from scratch, to solve a specific problem. The problem is divided into tasks, and special task performers (agents) are designed to solve these tasks for that problem only. All interaction (cooperation, coordination if any) strategies are incorporated as an integral part in the design of the system. It is a kind of top-down designed system, since agents are designed to conform to the requirements specified at the top. DPS may also be described as a closed system, in the sense that the interfaces of its agents are closed, i.e., the agents are not intended to participate in other DPSs except the one for which they are designed. In contrast in a pure multiagent system, the agents are autonomous, potentially preexising, and typically heterogeneous. Although an agent here can also be a special task performer, it has an open (in the sense of standard) interface, which is accessible to everyone (using the right standard). It can therefore participate in any multiagent systems wherever its skill is called for, the open interface facilitating the necessary interagent interactions. An agent may act differently in different situations, cooperating, competing or whatever, depending on the requirements. So an MAS can act as a DPS when required, but it can also act differently when required. An MAS can also be viewed as a bottom-up designed system, since in principle agents are designed first (they can preexist), and the solution strategy (top) for a given problem is specified later. Most systems, however, are likely to lie away from the two extreme points of the spectrum. ------------------------------------------------------------------------ Date: Thu, 22 Aug 91 10:26 CDT From: Michael N. Huhns Subject: More on DPS vs. MA Systems Based on the discussion so far it seems that DPS/MA systems can be viewed from at least three valid perspectives, and there are different definitions and characterizations from each perspective. Perspective 1: the viewpoint of the designer(s) of the system. Jeff Rosenschein has presented a definition from this perspective, where he states that "a `pure' MA system is one where the agents have been designed separately, by different designers." Perspective 2: the viewpoint of an observer of the system, who is trying to characterize it, or the problems it is solving, as either DPS or MA. Les Gasser and Ed Durfee have presented definitions from this perspective, based on the competition and cooperation that occurs in the system and whether or not there is a common problem or goal being solved by the system, or whether the goals require collective action. Perspective 3: the viewpoint of an observer of an agent in the system, who is trying to characterize how autonomously the agent can or does behave. My contributions to the discussion have come from this perspective. Interestingly, all contributors to the discussion seem to agree that the DVMT represents a DPS system, because 1) it was constructed by one design team, 2) it had a single overall problem it was trying to solve and whose solution required collective activity, and 3) its problem solvers had insufficient autonomy to do anything other than work on vehicle tracks.