Introduction to Game Theory

Strategic Thinking When Others Are Thinking Strategically

1. What is Game Theory?

Definition

Game Theory is the mathematical study of strategic interaction among rational decision-makers.

It provides tools for analyzing situations where the outcome for each participant depends not only on their own decisions but also on the decisions of others.

Key Characteristics
  • Multiple agents making decisions
  • Strategic interdependence: my payoff depends on others
  • Rationality: agents maximize their utility
  • Common knowledge: everyone knows the game structure
Why It Matters for AI
  • Multi-agent systems and robotics
  • Competitive learning environments
  • Adversarial settings (security, games)
  • Mechanism design (auctions, markets)
The Key Insight
"I think that you think that I think..."

Game theory formalizes this recursive reasoning about what rational agents will do.

2. Historical Foundations

1944: The Beginning

John von Neumann and Oskar Morgenstern publish "Theory of Games and Economic Behavior" - founding the field.

1950: Nash Equilibrium

John Nash proves that every finite game has an equilibrium point. This becomes the central solution concept in game theory.

1994: Nobel Prize

Nash, Harsanyi, and Selten receive the Nobel Prize in Economics for their contributions to game theory.

Today: AI & Multi-Agent Systems

Game theory is essential for multi-agent AI, reinforcement learning, autonomous systems, and mechanism design.

John Nash (1928-2015)

Mathematician whose work on equilibrium points revolutionized economics and game theory.

"The best result will come from everyone in the group doing what's best for himself AND the group."

3. Types of Games

Cooperative Games

Players can form binding agreements and coalitions.

  • Focus on coalition formation
  • How to divide payoffs fairly?
  • Example: Business partnerships
Non-Cooperative Games

Players make independent decisions (our focus).

  • Each player maximizes own utility
  • No binding agreements possible
  • Example: Prisoner's Dilemma
Zero-Sum Games

One player's gain = another's loss.

  • Strictly competitive
  • Sum of payoffs = 0
  • Examples: Chess, Poker, Tennis
Non-Zero-Sum Games

Players can both win or both lose.

  • Mixed motives: compete and cooperate
  • More realistic for real-world
  • Examples: Trade, Arms Race
Simultaneous Games

Players choose actions at the same time.

  • Normal/Strategic form representation
  • Don't observe opponent's choice
  • Example: Rock-Paper-Scissors
Sequential Games

Players move in order, observing previous moves.

  • Extensive form (game tree)
  • Subgame perfect equilibrium
  • Example: Chess, Bargaining
Our Focus in This Lecture

We will focus on non-cooperative, simultaneous, finite games in strategic (normal) form. This is the foundation for understanding Nash Equilibrium and strategic reasoning.

4. Interactive: What Would You Do?

Scenario Challenge
🚗 Traffic Intersection

You and another driver arrive at an intersection at the same time. There's no traffic light.

Your options:

5. Real-World Examples

💰
Auctions

Bidding strategies, winner's curse, optimal auction design

🛡️
Cybersecurity

Attacker vs defender, resource allocation, intrusion detection

🚘
Autonomous Vehicles

Multi-car coordination, merging, intersection protocols

📶
Network Routing

Selfish routing, congestion games, price of anarchy

🤖
Multi-Agent AI

Cooperative robots, competitive games, emergent behavior

📊
Market Competition

Pricing strategies, entry deterrence, oligopoly

6. Game Theory vs MDP: Reasoning Under Uncertainty

Where Does Game Theory Fit in AI?

Both MDPs and Game Theory deal with decision-making under uncertainty, but they address different sources of uncertainty. MDP handles uncertainty from the environment, while Game Theory handles uncertainty from other rational agents.

Aspect MDP (Lecture 12) Game Theory (Lecture 13)
Source of Uncertainty Environment dynamics (stochastic transitions) Other agents' strategic choices
Number of Agents Single agent vs Nature/Environment Multiple rational agents (2+ players)
Environment Behavior Probabilistic but non-strategic Strategic and reactive to your actions
Solution Concept Optimal Policy (maximize expected utility) Nash Equilibrium (no unilateral deviation)
Key Question "What should I do given the world's probabilities?" "What should I do given what others will do?"
PEAS Framework for Game Theory

Recall the PEAS framework: Performance, Environment, Actuators, Sensors

Performance Measure

Payoff/utility function ui(s) — what each player wants to maximize

Environment

Other players + game rules. The "world" includes rational opponents!

Actuators

Strategy selection — choosing an action from strategy set Si

Sensors

Game structure, payoff matrix (common knowledge), possibly opponent history

Environment Classification

How does a strategic game classify under the standard AI environment properties?

Observable
Partially Observable

You know the game structure, but NOT what action opponents will choose (simultaneous games)

Deterministic
Strategic (Non-Deterministic)

Outcome depends on others' choices — uncertainty from rationality, not randomness

Agents
Multi-Agent

Multiple rational decision-makers, each with their own goals (competitive or cooperative)

Static vs Dynamic
Static (Simultaneous Games)

Environment doesn't change while you deliberate (one-shot games)

Discrete vs Continuous
Discrete (Finite Games)

Finite set of strategies and outcomes — can be represented as a matrix

Known vs Unknown
Known (Complete Information)

All players know the game structure and payoffs (common knowledge assumption)

Key Insight: Strategic Uncertainty

In MDP, uncertainty comes from nature (random transitions). In Game Theory, uncertainty comes from other agents who are also trying to maximize their payoffs. This makes the problem fundamentally harder — the "environment" is thinking back at you!

Back to Lecture Overview Next: System Model Components