Introduction to Game Theory Yale BraunsteinSpring 2007
General approach • Brief History of Game Theory • Payoff Matrix • Types of Games • Basic Strategies • Evolutionary Concepts • Limitations and Problems
Brief History of Game Theory • 1913 - E. Zermelo provides the first theorem of game theory; asserts that chess is strictly determined • 1928 - John von Neumann proves the minimax theorem • 1944 - John von Neumann & Oskar Morgenstern write "Theory of Games and Economic Behavior” • 1950-1953 - John Nash describes Nash equilibrium
Rationality Assumptions: • humans are rational beings • humans always seek the best alternative in a set of possible choices Why assume rationality? • narrow down the range of possibilities • predictability
Utility Theory Utility Theory based on: • rationality • maximization of utility • may not be a linear function of income or wealth It is a quantification of a person's preferences with respect to certain objects.
What is Game Theory? Game theory is a study of how to mathematically determine the best strategy for given conditions in order to optimize the outcome
Game Theory • Finding acceptable, if not optimal, strategies in conflict situations. • Abstraction of real complex situation • Game theory is highly mathematical • Game theory assumes all human interactions can be understood and navigated by presumptions.
Why is game theory important? • All intelligent beings make decisions all the time. • AI needs to perform these tasks as a result. • Helps us to analyze situations more rationally and formulate an acceptable alternative with respect to circumstance. • Useful in modeling strategic decision-making • Games against opponents • Games against "nature„ • Provides structured insight into the value of information
Types of Games • Sequential vs. Simultaneous moves • Single Play vs. Iterated • Zero vs. non-zero sum • Perfect vs. Imperfect information • Cooperative vs. conflict
Zero-Sum Games • The sum of the payoffs remains constant during the course of the game. • Two sides in conflict • Being well informed always helps a player
Non-zero Sum Game • The sum of payoffs is not constant during the course of game play. • Players may co-operate or compete • Being well informed may harm a player.
Games of Perfect Information • The information concerning an opponent’s move is well known in advance. • All sequential move games are of this type.
Imperfect Information • Partial or no information concerning the opponent is given in advance to the player’s decision. • Imperfect information may be diminished over time if the same game with the same opponent is to be repeated.
Key Area of Interest • chance • strategy Imperfect Information Non-zero Sum
Matrix Notation Notes: Player I's strategy A may be different from Player II's. P2 can be omitted if zero-sum game
Prisoner’s Dilemma & Other famous games A sample of other games: Marriage Disarmament (my generals are more irrational than yours)
Prisoner’s Dilemma NCE Prisoner 2 Blame Don't Blame 10 , 10 0 , 20 Prisoner 1 Don't 20 , 0 1 , 1 Notes: Higher payoffs (longer sentences) are bad. Non-cooperative equilibrium Joint maximumInstitutionalized “solutions” (a la criminal organizations, KSM) Jt. max.
Games of Conflict • Two sides competing against each other • Usually caused by complete lack of information about the opponent or the game • Characteristic of zero-sum games
Games of Co-operation Players may improve payoff through • communicating • forming binding coalitions & agreements • do not apply to zero-sum games Prisoner’s Dilemma with Cooperation
Prisoner’s Dilemma with Iteration • Infinite number of iterations • Fear of retaliation • Fixed number of iteration • Domino effect
Basic Strategies 1. Plan ahead and look back 2. Use a dominating strategy if possible 3. Eliminate any dominated strategies 4. Look for any equilibrium 5. Mix up the strategies
Plan ahead and look back Opponent Strategy 1 Strategy 2 Strategy 1 150 1000 You Strategy 2 25 - 10
Use strategy 1 If you have a dominating strategy, use it Opponent Strategy 1 Strategy 2 Strategy 1 150 1000 You Strategy 2 25 - 10
Eliminate strategy 2 as it’s dominated by strategy 1 Eliminate any dominated strategy Opponent Strategy 1 Strategy 2 Strategy 1 150 1000 You Strategy 2 25 - 10 Strategy 3 160 -15
Look for any equilibrium • Dominating Equilibrium • Minimax Equilibrium • Nash Equilibrium
Maximin & Minimax Equilibrium • Minimax - to minimize the maximum loss (defensive) • Maximin - to maximize the minimum gain (offensive) • Minimax = Maximin
Maximin & Minimax Equilibrium Strategies Opponent Strategy 1 Strategy 2 Min 150 Strategy 1 150 1000 You Strategy 2 25 - 10 - 10 Strategy 3 160 -15 -15 Max 160 1000
Definition: Nash Equilibrium “If there is a set of strategies with the property that no player can benefit by changing her strategy while the other players keep their strategies unchanged, then that set of strategies and the corresponding payoffs constitute the Nash Equilibrium. “ Source: http://www.lebow.drexel.edu/economics/mccain/game/game.html
Is this a Nash Equilibrium? Opponent Strategy 1 Strategy 2 Min 150 Strategy 1 150 1000 You Strategy 2 25 - 10 - 10 Strategy 3 160 -15 -15 Max 160 1000
Boxed Pigs Example Cost to press button = 2 units When button is pressed, food given = 10 units
Decisions, decisions... Little Pig Press Wait Press 5 , 1 4 , 4 Big Pig Wait 9 , -1 0 , 0
Time for "real-life" decision making • Holmes & Moriarity in "The Final Problem" • What would you do… • If you were Holmes? • If you were Moriarity? • Possibly interesting digressions? • Why was Moriarity so evil? • What really happened? • What do we mean by reality? • What changed the reality?
Mixed Strategy Safe 1 Safe 2 $10,000 Safe 1 $ 0 $100,000 $ 0 Safe 2
Player #2 Moriarty Strategy #1 Strategy #2 Canterbury Dover Player #1 0 50 Strategy #1 Canterbury Payoff (1,1) Payoff (1,2) Holmes 0 100 Dover Strategy #2 Payoff (2,1) Payoff (2,2) The Payoff Matrix for Holmes & Moriarity
Where is game theory currently used? • Ecology • Networks • Economics
Limitations & Problems • Assumes players always maximize their outcomes • Some outcomes are difficult to provide a utility for • Not all of the payoffs can be quantified • Not applicable to all problems
Summary • What is game theory? • Abstraction modeling multi-person interactions • How is game theory applied? • Payoff matrix contains each person’s utilities for various strategies • Who uses game theory? • Economists, Ecologists, Network people,... • How is this related to AI? • Provides a method to simulate a thinkingagent
Sources • Much more available on the web. • These slides (with changes and additions) adapted from: http://pages.cpsc.ucalgary.ca/~jacob/Courses/Winter2000/CPSC533/Pages/index.html • Three interesting classics: • John von Neumann & Oskar Morgenstern, Theory of Games & Economic Behavior (Princeton, 1944). • John McDonald, Strategy in Poker, Business & War (Norton, 1950) • Oskar Morgenstern, "The Theory of Games," Scientific American, May 1949; translated as "Theorie des Spiels," Die Amerikanische Rundschau, August 1949.