5. Agents & Agency
Agents in Agency
-
A key challenge in agent organizations is creating ways for agents to evaluate the roles within an organization so they can decide whether to participate. An agent needs to consider if it wants to take on a role and if it has the abilities to perform it as required.
-
In many cases of Gen-2 MAS, agents are built from the start to match the behavior expected by the agency. In these cases, the agency model is often built into the design of the agents. However, more complete approaches cannot assume that all agents are known in advance. They require agency-aware agents that can think about their own goals and preferences and then decide or negotiate whether to join an agency.
-
By defining how interactions can happen in an environment, multi-agent agencies can set global goals without depending on how individual agents are designed. A role description in an agency model identifies a position to be filled by a participant, which contributes to part of the agency's objectives and follows its interaction rules.
-
From the agency's perspective, it usually does not matter which agent takes a role, as long as the agent has the needed abilities and meets its policies. However, each agent may perform the role differently because they have their own goals and use their own reasoning to decide how to act. This means the overall results can vary. The ability to assign different agents to roles as needed gives the agency flexibility to adapt to changing goals and environments.
-
For agents to join an agency, there must be a clear description of what is expected of them, along with an admission process to check if the goals and requirements of both sides align. AgencyGrid approaches role enactment in two built-in ways:
- As a process of communication with the agency and on the outcomes, such as when an agent adopts the role’s objectives as its own.
- As a compatibility check to ensure an agent’s goals align with the role’s goals before allowing it to take the role.
-
Agents that take on roles must be able to perform them effectively. This means they should be able to:
- Carry out the functions defined by the role or required by its relationships with other roles, including using the resources available to the role.
- Communicate on behalf of the role with agents playing other roles.
- Decide which of their own plans and activities can be used to achieve the role’s objectives.
-
AgencyGrid gives agents an interface to the agency through a system called a governor. This interface blocks any actions not allowed by the role definition or policy specifications, ensuring that organizational rules are followed.
-
From the agent’s perspective, a role description outlines the knowledge and skills needed to perform it. If the agency's specification is very detailed, agents need to operate strictly within the designed environment to exactly match those requirements. This works better in closed systems. However, this approach is not flexible enough for different kinds of agents, capabilities, or styles of enacting a role. It hides the agent’s individuality from the organization. In open systems with many different kinds of agents, they are usually designed to operate independently of any one agency.
-
From the agency's perspective, the key concern is how an agent’s attitude affects its role performance. Agents can be selfish, altruistic, honest, dishonest, or show other social attitudes. These attitudes affect how they plan and prioritize goals, which influences whether and how they meet role objectives and follow role norms. Some agents may fully commit to role goals over their own goals, while others may focus on role goals only after their personal goals are met. The relationships between agent goals, role objectives, and role norms all affect the agency's overall performance.
-
Agents join an agency when there is some benefit for both the agent and the agency. How an agent performs its role can vary:
- In a social enactment strategy, the agent focuses on fulfilling the agency's objectives before pursuing its own goals.
- In a selfish enactment strategy, it works on its own goals first.
- Many other variations are possible, depending on how agent plans and role objectives influence each other. Agency norms also shape behavior, as they can limit or change how agents achieve their personal goals.