Multi-Agent Systems

by Albert Major

Introduction to Agent-Based Modeling

Multi-Agent Systems (MAS) Definitions and Criteria

• Definition Criteria: For a system to be considered a Multi-Agent System, there must be a possibility for interaction between agents.
• Examples analyzed against the definition:
  • Human Liver: Considered a MAS. While not every cell interacts with every other cell globally, they interact locally with neighbors and exchange information (e.g., oxygen exchange).
  • CSAI Program: Considered a MAS. Students and staff are agents; while you may not interact with everyone, the possibility for interaction exists.
  • Particle Accelerator: Not considered a MAS in this context. Particles are not modeled as having "sensing" capacities or agency to make autonomous decisions based on their environment, unlike agents in an ABM.
• Emergent Behavior: It is False that MAS always exhibit emergent behavior. It is possible to create systems where agents do not interact significantly or where no complex pattern arises from their interactions.

Modeling and Simulation Concepts

• What is a model?

  • A simplified representation of a process, object, or event.
  • All models are inherently "wrong" because they are simplifications, but they are useful for understanding complex phenomena.

• Example: The Atkinson-Shiffrin Memory Model (1950s) uses boxes to represent short-term and long-term memory. It does not perfectly represent neural biological reality, but it provides a useful conceptual framework for cognitive psychology.

Image of Atkinson-Shiffrin memory model

↑ not important ↑

• What is a simulation?
• A computer-based execution of a model as it unfolds over time.
• Used to generate predictions and validate models by comparing output to real-world data.
• The ABM Approach: The goal is to create models -> run simulations -> evaluate output by varying conditions or comparing to reality.

Agent-Based Modeling (ABM) Core Components

• Agents: Autonomous individuals with properties (variables) and actions (behaviors).
  • Note: In tools like NetLogo, inanimate objects (like the ground/environment) can also be treated as agents with properties.
• Environment: The "world" in which agents exist.
• Interactions: The dynamics occur through Agent-Agent interactions and Agent-Environment interactions.

Introduction to NetLogo

NetLogo is the most widely used software for Agent-Based Modeling. It bridges the gap between novice users (easy to run existing models) and experts (coding complex custom models).

Key Interface Components:

1. Interface Tab: Visualizes the model. Contains buttons (to run code), sliders (to change parameters like population size or density), and monitors (to track data).