Multi-Agent Systems

by Albert Major

Core Concepts of Adaptive Agents

Defining Adaptation and Cognition

• Adaptive Behavior: An agent is considered adaptive if it adjusts its behavior based on environmental conditions or the behavior of other agents.
  • Rule-Following: Even simple rule-following agents are considered adaptive if those rules respond to changes. Complex learning (like neural networks) is not required for a definition of adaptation in this context; simple "if-then" logic suffices.
• Reflexive Agents: These agents operate on deterministic rules. If a specific condition is met in the environment, the agent immediately performs a specific behavior (similar to a biological knee-jerk reflex).
• Utility-Based Cognition: Agents attempt to optimize a specific utility function. They weigh values (e.g., package value) against costs (e.g., energy cost, collision penalties) to make decisions.

Subsistence

• Definition: The ability of agents to maintain and support themselves to survive.
• Mechanism: This is often driven by agent-environment interactions (eating resources) or agent-agent interactions (predation).
• Example: In a Predator-Prey model, subsistence is the drive to maintain energy levels above a threshold to avoid death and gather enough energy to reproduce.

Perturbation and Resilience

• Perturbation: An event or shift that changes the established dynamics or patterns of a model.
  • Spatial Perturbations: Changes in the location or availability of resources (e.g., grass is consumed in a specific area).
  • Temporal Perturbations: Events happening over time, such as seasonal weather changes (droughts) that affect resource regrowth.
• Resilience: The capacity of a system to recover from a perturbation and return to a stable state.
• Quantifying Resilience (Entropy/Complexity):
  • Resilience Curve: When a system is perturbed, the "entropy" (interpreted here as system complexity/chaos) increases.
  • Recovery Time: Resilience is measured by how quickly the system returns to its normal baseline level of entropy after the spike.
  • High Resilience: Characterized by a short time to peak entropy and a fast return to normal levels.
  • Low Resilience: Characterized by a long recovery time or a failure to ever return to the baseline state.

Specific Models Discussed

1. Predator-Prey (Wolf-Sheep) Modifications

• Context: Used to explore system stability and oscillating populations.