Intrafield Taxonomy

Several taxonomies have been presented previously for the related field of DAI. One example is Decker's division mentioned above [7]. In this same survey of DPS, Decker presents four dimensions of DAI:

• Agent granularity (coarse vs. fine);
• Heterogeneity of agent knowledge (redundant vs. specialized);
• Methods of distributing control (benevolent vs. competitive, team vs. hierarchical, static vs. shifting roles);
• and Communication possibilities (blackboard vs. messages, low-level vs. high-level, content).

Along dimensions 1 and 4, multiagent systems have coarse agent granularity and high-level communication. Along the other dimensions, they can vary across the whole ranges. In fact, the remaining dimensions are very prominent in this article: degree of heterogeneity is a major MAS dimension and all the methods of distributing control appear here as major issues.

More recently, Parunak has presented a taxonomy of MAS from an application perspective [8]. From this perspective, the important characteristics of MAS are:

• System function;
• Agent architecture (degree of heterogeneity, reactive vs. deliberative);
• System architecture (communication, protocols, human involvement).

A useful contribution is that the dimensions are divided into agent and system characteristics. Other overviews of DAI and/or MAS include [6, 9, 10].

The taxonomy presented in this article is organized along the most important aspects of agents (as opposed to domains): degree of heterogeneity and degree of communication. Communication is presented as an agent aspect because it is the degree to which the agents communicate (or whether they communicate), not the communication protocols that are available to them, that is considered. All the other aspects of agents in MAS are touched upon within the heterogeneity/communication framework. For example, the degree to which different agents play different roles is certainly an important MAS issue, but here it is framed within the scenario of heterogeneous non-communicating agents (it arises in the other two scenarios as well). Domain issues are discussed separately in Section 4.2.

The three combinations of heterogeneity and communication considered in this article (homogeneous non-communicating agents; heterogeneous non-communicating agents; and heterogeneous communicating agents) are presented in order of increasing complexity and power (see Figure 2).

  
Figure 2: The major categories of the intrafield taxonomy and how they relate to the major dimensions. Single agent systems are more centralized and complex than multiagent systems.

Many of the issues that arise in the earlier scenarios also apply in the later scenarios. Nevertheless, they are only mentioned again in the later scenarios to the degree that they differ or become more complex. Notice that single agent systems are presented as being more complex than multiagent systems. The intuition behind this presentation will become clear as the article proceeds.

The complexity (and accompanying power) of the different scenarios is important, especially from the perspective of system engineering. In general, system builders should stay within the simplest possible multiagent scenario. Using a scenario with more power than is needed only makes programming and debugging unnecessarily complex and difficult. In a sense, this recommendation is the system designer's version of Occam's razor: it allows anybody trying to reverse-engineer the system to safely use Occam's razor!

The multiagent scenarios along with the issues that arise therein are summarized in Table 2. The techniques that currently exist to address these issues are described in detail in Sections 6 - 8.

  
Table 2: Issues arising in the various scenarios