Game Theory : Minority Games

Published:
Python

Introduction

Game theory provides a valuable framework for analyzing strategic interactions in various real-world scenarios. One such model is the Minority Game, which simulates situations where the optimal choice is the one that fewer participants select. This dynamic finds applications in diverse areas, such as financial markets, traffic systems, and competitive sports like Formula One.

This project explored different forms of the Minority Game, developed simulation algorithms, and introduced inductive strategies to enhance agents’ decision-making. The aim was to provide a detailed theoretical analysis, followed by a step-by-step implementation and evaluation of different game versions.

What is a Minority Game?

A Minority Game is a type of game where agents aim to choose options such that they belong to the minority group. If an agent selects an option chosen by a smaller number of agents, they receive a reward. This setup reflects many real-world scenarios where resources or opportunities are limited, and participants benefit from taking a less crowded path.

Components of a Minority Game

Theoretical Analysis: Nash Equilibrium in Static Games

Before moving to simulations, the Nash Equilibrium was explored for the static version of the Minority Game. This theoretical analysis examined the scenarios in which no agent would benefit from unilaterally changing their decision.

In a simple normal form game representation, two Nash equilibria were identified: where agent ( A_i ) chooses the opposite option from the rest of the agents. This duality creates uncertainty, similar to classic game theory problems like the “Battle of the Sexes,” indicating that predicting the majority’s choice is inherently challenging.

Simulation of Static and Repeated Games

1. Static Game Simulation

In the static version, agents do not have access to historical data. They select their option randomly, weighted by the minority threshold ( T ). This reflects scenarios where agents make a one-time decision without prior knowledge of other agents’ choices.

Results and Observations

Simulations showed that, as expected, the average percentage of agents choosing each option matched the minority threshold. This behavior aligns with the idea of agents making weighted decisions in the absence of inductive strategies.

2. Repeated Game Simulation

In a Repeated Game, agents can participate in multiple rounds, allowing them to gain insights into the overall behavior of other agents. However, without inductive strategies, the decision-making logic remains simple, and agents do not adapt their behavior over time.

Results and Observations

When agents repeated the same decision-making strategy across multiple rounds, the mean proportion of agents selecting each option remained consistent with the static game results. The spread and score distributions followed expected trends, with minor variations due to randomness.

Proposal of Inductive Strategies

Inductive strategies are decision-making methods where agents learn and adapt based on past experiences. In this project, several inductive strategies were proposed to allow agents to make more informed choices.

Proposed Strategies

Implementation and Simulation of Inductive Game Version

The inductive game version introduced new complexities in the agent’s class structure, allowing them to evaluate and switch between strategies using a softmax selection process. This dynamic decision-making setup aimed to explore how agents with memory and adaptive capabilities could outperform non-inductive strategies.

Results and Analysis

A Real-World Minority Game Example: Formula One Pit Strategies

One notable application of Minority Games in the real world is Formula One pit stop strategies. Drivers can choose to pit once or twice during a race. If the majority of drivers opt for a single pit stop, those choosing two can leverage fresher tires for faster laps, and vice versa.

However, the dynamics are more complex due to factors like:

Conclusion

This project explored the theoretical and practical aspects of Minority Games, highlighting how strategic adaptation can significantly impact agent success. By simulating different game versions and introducing inductive strategies, the project illustrated the power of adaptive learning in dynamic environments.

Previous Project Practical Aspects of Fairness In AI