Agent-based models

Question 1

Name 3 properties of an agent based model

Answer 1

1. Includes agents that are intended to model people or other entities that gather information about the world, make decisions, and take actions
2. Agents are situated in a space or in a network and interact with each other locally.
3. They usually have imperfect or incomplete information about the world.
4. Often include randomness.
5. Useful for modelling dynamics of systems and relationships between individuals and system behaviour.

Question 2

Describe Schelling’s model

Answer 2

-A grid where each cell represents a house.
-The houses are occupied by two kinds of agents, red or blue with roughly equal numbers.
-About 10% of the cells are empty.
-An agent may be happy or unhappy, where they are happy if they have at least two neighbours like themselves, and unhappy if they have one or zero.
-Each step of the simulation, an agent is chosen at random, and checks whether they are happy. If they are happy, nothing happens. If they are unhappy, the agent chooses an unoccupied cell at random and moves.

Question 3

Describe the outcome of Schelling’s model

Answer 3

-From a random starting point, clusters of similar agents form almost immediately.
-The clusters grow over time until there are a small number of large homogenous clusters.

Question 4

Are agents racist in Schelling’s model?

Answer 4

Agents would be happy in a mixed neighbourhood, but they just don’t prefer being greatly outnumbered. This might be considered mild xenophobia.

Question 5

What are the parameters of the Schelling model?

Answer 5

p = threshold for number of agents of same type before moving (e.g. p=0.3 means 30% of the neighbours have to be the same type, otherwise they would move)
n = size of grid (e.g. n x n)

Question 6

What happens when you increase p in the Schelling model?

Answer 6

You increase segregation.

Question 7

What is sugarscape?

Answer 7

Sugarscape is a model of simple economy where agents move around on a 2-D grid, harvesting and accumulating “sugar”, which represents economic wealth.
Some parts of the grid produce more sugar than others, and some agents are better at finding it than others.

Question 8

What are the attributes of each agent in sugarscape?

Answer 8

Sugar: each agent starts with sugar (5-25)
Metabolism: each agent has an amount of sugar they consume per time step (1-4)
Vision: How much sugar each agent can see in nearby cells (1-6)

Question 9

Detail the time steps in sugarscape

Answer 9

Agents move one by one, in a random order.
Each agent follows the following rules:
1. The agent surveys k cells in each of the 4 directions, where k is the vision.
2. It chooses the unoccupied cell with the most sugar. (if a tie it picks the closer cell).
3. The agent moves to the selected cell and harvests the sugar, adding to its sugar amount and leaving the cell empty.
4. The agent consumes based on its metabolism and, and if the total sugar content at the end is negative, it starves and is removed.

Cells grow back sugar, but are capped.

Question 10

What are some of the insights from the sugarscape model?

Answer 10

Agents with areas with less sugar are likely to starve unless they have high initial sugar, and high vision.
Agents with high metabolism or low vision are
the most likely to starve.

Question 11

How does sugarscape model wealth inequality?

Answer 11

The distribution of wealth is lognormal, which is a heavy tailed distribution.
In reality, the distribution of wealth in most countries is lognormal.

Question 12

What is emergence?

Answer 12

An emergent property is a characteristic of a system that results from the interaction of its components, not from their properties.

For example waves in sugarscape, segregation in schelling model.

A property is emergent if we don’t have a reductionist explanation for it.

Question 13

What is the wave property in sugarscape?

Answer 13

The agents with the longest vision cross the valley between the peaks and propagate toward the high sugar areas. This can be seen as migration.

Question 14

Explain how traffic jams occur in traffic simulation.

Answer 14

Cars start equally spaced out. We expect them to accelerate until they reach the speed limit, or until their speed causes them to collide with another car, causing the car to stop.

The random noise is the reason that traffic jams occur in the simulation, as some cars go faster than others, and the spacing becomes uneven.

Jams tend to persist over time.

Question 15

What are some of the parameters in traffic simulation?

Answer 15

Minimum and maximum acceleration, noise, speed limit, number of cars

Question 16

What are some of the key findings from traffic simulation?

Answer 16

-As the number of cars increases traffic jams become more severe
-25 cars of fewer, with 0 noise, cars can reach maximum speed.
-With just 1% error/noise, capacity drops to 10 to reach maximum speed.

Question 17

What are the properties of boid agents?

Answer 17

Flock centring = move toward the centre of the flock
Collision avoidance = avoid obstacles, including other boids
Velocity matching = match velocity (speed + direction) with neighbouring boids

Boids only base this off local information based on FOV

Question 18

What are emergent properties of traffic jams and boids?

Answer 18

Traffic jams move backwards even though the cars are moving forward.
Boids behave as if they are centrally organised even though the animals in them are making individual decisions based on local information.