Multi-Agent Systems

by Albert Major

Agent-Based Models for Cognitive and Computational Social Science

1. Agents in NetLogo

• Definition of Agents: In the specific context of NetLogo, the definition of an agent is broader than in general theory.
  • Turtles: The typical agents that move and have behaviors.
  • Patches: The background grid squares (environment) are also considered agents.
  • Links: Connections between agents are treated as agents.
• Required Procedures: Every NetLogo model requires two specific procedures to function:
  • Setup: Initializes conditions, creates agents, and sets variables.
  • Go: Runs the model (executes steps).
• Note: Global parameters or specific counts (e.g., 100 turtles) are not mandatory requirements for every model structure.

2. Limitations of Traditional Cognitive Science Methods

• Traditional Methods:
  • Often rely on isolated experiments (e.g., measuring reaction times to stimuli, surveys, button presses).
  • Use analytic models (e.g., path analysis/regression) to describe outcomes based on static variables.
• The Problem with Isolation:
  • Human cognition is rarely isolated; it occurs during interaction.
  • Standard lab experiments often fail to capture feedback loops (e.g., how one person's behavior changes the environment, which then influences another person).
  • Scaling Issue: It is difficult to determine if a model of individual cognition applies when scaled up to a group. A group is not simply the same individual model multiplied; interactions change the dynamics.

3. When is Agent-Based Modeling (ABM) Useful?

ABM is a methodological tool used to overcome the limitations of isolated lab studies. It is particularly useful in the following scenarios:

• Interdependent Cognitive Functions: When an individual's decision depends on the actions of others (e.g., attention span influenced by neighbors, voting behavior influenced by peer discussions).
• Heterogeneity: When agents are not identical (unlike standard statistical assumptions). Agents can have different beliefs, roles, or decision-making rules.
  • Exogenous variability: Agents start with different parameters.
  • Endogenous variability: Agent parameters change over time based on experience.
• Temporal Dynamics: When the research question involves change over time, longitudinal effects, or long-term evolution (e.g., cultural transmission) that cannot be observed in a short lab session.
• Spatial Distribution: When physical location matters (e.g., navigation, foraging, search behaviors).
• Ethical Constraints: ABM allows researchers to simulate scenarios that are unethical to test on humans, such as the spread of violence or misinformation.

4. Calibration and Validation in ABM

• Calibration: The process of setting model parameters.
  • Can be derived from domain knowledge (theory).
  • Can be derived from empirical data (e.g., using real-world problem-solving speeds to set agent variables).