3 Representation of knowledge in SDML

3.4 Models

In SDML, models comprise a number of agents, which represent distinct entities (such as people or organisations) that are capable of making decisions and interacting with each other and the environment. Their behaviour is specified using rules and they communicate by placing information on databases. Agents have their own private rulebases and databases, and also have access to a public database contained in the model. The model also has its own rulebases, which can be used to represent the environment.

This modular approach of SDML offers a number of advantages over a single rulebase and database:

In the real world, entities often have private information that they do not wish to reveal to everybody else. Private databases are ideal for such information.
A large modelling task can often be simplified by dividing it into more manageable chunks. A single rulebase can be harder to find errors in and modify.
Modularity can also aid efficiency. Data can be located more quickly and fewer rules need to be considered for firing at once.

Additionally, this approach enables rules to write other rules, without resorting to imperative techniques. In SDML, agents have meta-level rulebases and object-level rulebases. Typically, meta-level rules are used for learning and generating strategies, whereas object-level rules utilise such strategies to make day-to-day decisions. For example, a meta-level rule may state that a particular strategy is used (represented by a consequent stating that a particular rule is in the object-level rulebase) in a particular situation (if certain antecedents are true). Conversely, meta-level rules can make decisions based on current or previous strategies. Since antecedents and consequents are represented as clauses, rules on rulebases can be manipulated as easily as clauses on databases.

It should be noted that all permanent databases in a model, whether public or private, have subdatabases corresponding to particular time periods. Similarly, permanent object-level rulebases have subrulebases, enabling meta-level rules to generate different object-level rules at different times.

The modularity of SDML is further enhanced by its object-oriented facilities, detailed discussion of which are beyond the scope of this paper. General rules can be specified in rulebases associated with agent types. Rules defined in general types are inherited by specific types and individual agents.

Efficient Forward Chaining for Declarative Rules in a Multi-Agent Modelling Language - 16 FEB 95

[Next] [Previous] [Up] [Top] [Contents]

Generated with CERN WebMaker