1. Agent Based Modeling (ABM)
Stephen Kinsel

2. Outline What is ABM? Why use ABM? Applications Examples
Good Modeling Practices
Issues
Future of ABM

3. What is ABM? First: What is an agent?
An entity that functions continuously and autonomously in an environment in which other processes take place and other agents exist (Shoham)
General Characteristics:
Autonomy
Pro-activeness
Reactivity
“Social” Ability
autonomy: agents operate without direct intervention from the user, and have some sort of control over their actions
pro-activeness: agents not only react in response to the environment, but also exhibit goal-directed behavior
reactivity: agents perceive their environment and respond to changes
social ability: agents can interact with other agents

4. What is ABM?
Simulation modeling technique where a system is modeled as a collection of agents and the relationships between them.
Agents individually asses its situation in the environment and make decisions on the basis of a set of rules.
- Bonabeau
Even a simple ABM can exhibit complex behavior patterns and provide valuable insight about the dynamics of a real world situation
Rules created by the programmer. Their behavior can be anything you want, ranging from producing, selling, consuming, competing.

5. Agent Types (DeLaurentis)
Mobile
Agent
Autonomous
Agent
Adaptive
Agent
Yes
Yes
Yes
Does it run without continuous user input?
Can it change its behavior based on past experiences
Does it move? No
Agent
Does it have a set solution path?
Reactive
Agent
Require Assistant User?
Yes
Does it collect, filter & classify information?
Yes No
Does it care about the utility value?
Interface
Agent
Utility
Agent
Yes
Yes No
Info-gathering
Agent
Agents can possess more than one property
Goal-based
Agent

6. Agent Types in an Example
Traffic Control
Reactive: Police (enforce laws of road)
Info-gathering: Media (informs the public of traffic and accidents in major areas)
Autonomous: Disruptors (weather / accidents)
Goal-based: City Planners (would like the least number of accidents and greatest amount of flow through parts of town)
Adaptive: Drivers (may avoid roads that are known to be overcrowded during certain times of day)
Utility: Drivers (would like to minimize drive time / distance)

7. Why Use ABM? Captures Emergent Phenomena
As the components of a system interact with each other, and influence each other through these interactions, the system as a whole exhibits emergent behavior (Roetzheim)
This characteristic makes the output of a system difficult to understand and predict

8. Emergence Example Group of 10 – 40 people
Each member randomly chooses two people, person A and person B.
Members move themselves so that A is between themselves and B
Now move so that member is between A and B.

9. Why Use ABM?
Provides a Natural Description of a System composed of “behavioral” entities
Describes the system from the perspective of its constituent units’ activities more so than the system’s processes
Heterogeneous units

10. Heterogeneous Components of a System

11. Why Use ABM? Flexibility
What if the appropriate level of description or complexity is not known ahead of time?
Easy to add / subtract more agents
Tuning the complexity of the agents
changing behaviors, degree of rationality, rules of interactions, etc

12. Traditional vs. ABM simulation
ABM seeks “adaptive” rather than “optimizing” nature
Adapt: seek the rule and behavior set that lead to new capabilities
ABM does not emphasize analytical solutions (more qualitative than quantitative)

13. Areas of Application Flow Financial Markets Organizations Social
evacuation, traffic
Financial Markets
Organizations
organizational design, strategy
Social

14. ABM Generic Example - BOIDS
Separation: steer to avoid crowding local flockmates
Alignment: steer towards the average heading of local flockmates
Cohesion: steer to move toward the average position of local flockmates
Reynolds

15. ABM Generic Example - Evacuation
Stampede Situation
People become injured when they collide at a certain speed
As a consequence, leaving the room becomes difficult.
Stampede Situation w/ Column
A column in front of the door can avoid injuries.
It can increase the outflow well with less / no injured people
Helbing, Farkas, Vicsek

16. ABM Generic Example – Traffic Control

17. Good Modeling Practices
Choosing the language that is right for you and the problem
Goals of Good AB Programming
Project Management

18. Some Recommended Programs / Languages
StarLogo
Programmable modeling environment for new programmers
Swarm
For advanced programmers
Wide variety of tools
Languages
Basic
Easy to learn and use, but suitable for small projects
Pascal
Designed to be a first language for serious programmers, and easy to learn and is structured to encourage good programming habits
C, C++
Most commonly used among serious programmers.
Allows easy conversion between separate computers
OO languages make really large projects easier to program

19. Goals (Axelrod) Validity Would like to correctly implement the model
Is the model itself an accurate representation of the “real world”?
Problem: If there are unexpected results, is there necessarily a mistake?

20. Goals (Axelrod) Usability
Allow yourself and other users who follow to run the program, interpret its output, and understand how it works
Careful with different versions of the model
Extendability
Allow future users (including yourself) to adapt the program for new uses
New questions arise from models such as these

21. Project Management (Axelrod)
How can I achieve these goals?
Use long names for almost all variables
List all the variables at the start of the program
Write helpful comments
Fully label output
Develop upwardly compatible programs
Document versions of each code
Use commercial programs for most data analysis
Check microdynamics
Communication with other users
upwardly compatible: All new versions should include all useful features of old versions
check microdynamics: validating the program: difficult to confirm your code by checking out known results, because they probably don’t exist. Want to make sure your program is correctly handling the details

22. Issues with ABM Validity Human agents
Every model serves its own unique purpose
Must be built at the right level of description with the appropriate amount of complexity
Human agents
Complex Psychology
Irrational Behavior
Subjective Choices

23. Issues with ABM Qualitative vs. Quantitative
Varying degree of accuracy and completeness in input (data, expertise, etc)
Use qualitative data to learn about the system
Capturing the behavior of all constituent units
Lower level description can extremely computationally intense, and time consuming
Heterogeneous units

24. Future of ABM
New Software for Modeling
Research in BDI Architecture

25. References Shoham, Yoav, BDI agents: From Theory to Practice, 1993
Bonabeau, Eric, Agent-based modeling: Methods and techniques for simulating human behavior, 2002
Roetzheim, William H., Enter the Complexity Lab, Where Chaos Meets Complexity, 1994
Reynolds, Craig W., Flocks, herds and schools: A distributed behavioral model, 1987
Helbing, Dirk; Farkas, Illes; Vicsek, Tamas, Simulating Dynamical Features of Escape Panic, 2000
Axelrod, Robert, The Complexity of Cooperation, 1998