Message Passing Systems

An alternative communication system for agents.

Message passing systems differ from Blackboards in the following ways:

• Agents tend to know more about what each other.
• Knowledge is used to direct messages that a most likely to do the desired task.
• Each agent possesses sufficient knowledge so that its expertise can be employed without communication with other agents.
• Processed information then passed on.

A typical Message Passing System posses the following components:

Problem solving agents

-- domain specific experts.

System agents

-- command interpretation, error handling etc.

Facilities

-- built on functions that agents can use pattern matching.

Agent description database

-- updated in system.

Kernels

-- one per processor to handle I/O, message routing etc.

The problem solving agent will maintain models of other agents. It will typically model the following information about an agent:

• Name, Class and Location (Address).
• Relationship or role between the two agents.
• The agent's function or skills.
• The goal's or outcome of the agent
• The plan of the agent -- How it achieves its goals.