03/12/2023

1 2

3 4 1
 03/12/2023

5 6

“2/3 of the average” game

• Everyone writes down a number between 0 and 100
• Person closest to 2/3 of the average wins
• Example:
• A says 50
• B says 10
• C says 90
• Average(50, 10, 90) = 50
• 2/3 of average = 33.33
• A is closest (|50-33.33| = 16.67), so A wins

7 8 2
 03/12/2023

Rock-paper-scissors Matching pennies (~penalty kick)

Column player aka.
player 2
(simultaneously)
chooses a column

0, 0 -1, 1 1, -1
Row player
L R
aka. player 1
chooses a row 1, -1 0, 0 -1, 1
L 1, -1 -1, 1
A row or column is -1, 1 1, -1 0, 0
R -1, 1 1, -1
called an action or
(pure) strategy
Row player’s utility is always listed first, column player’s second

Zero-sum game: the utilities in each entry sum to 0 (or a constant)

Three-player game would be a 3D table with 3 utilities per entry, etc.

9 10

“Chicken” Rock-paper-scissors
• Two players drive cars towards each other
• If one player goes straight, that player wins
• If both go straight, they both die

S D

D S
D S 0, 0 1, -1 1, -1
D 0, 0 -1, 1 not zero-sum -1, 1 0, 0 -1, 1
S 1, -1 -5, -5 -1, 1 1, -1 0, 0
11 12 3
 03/12/2023

Dominance Prisoner’s Dilemma

• Player i’s strategy si strictly dominates si’ if
• for any s-i, ui(si , s-i) > ui(si’, s-i)
• si weakly dominates si’ if
• for any s-i, ui(si , s-i) ≥ ui(si’, s-i); and
• for some s-i, ui(si , s-i) > ui(si’, s-i)
• Pair of criminals has been caught
• District attorney has evidence to convict them of a
-i = “the player(s) minor crime (1 year in jail); knows that they
other than i” committed a major crime together (3 years in jail)
but cannot prove it
• Offers them a deal:
– If both confess to the major crime, they each get a 1 year reduction
– If only one confesses, that one gets 3 years reduction

strict dominance 0, 0 1, -1 1, -1 confess don’t confess

weak dominance -1, 1 0, 0 -1, 1 confess -2, -2 0, -3

-1, 1 1, -1 0, 0 don’t confess -3, 0 -1, -1
13 14

“Should I buy an SUV?” Mixed strategies

purchasing cost accident cost • Mixed strategy for player i = probability distribution over player i’s (pure)
strategies
• E.g.,1/3 , 1/3 , 1/3
cost: 5 cost: 5 cost: 5 • Example of dominance by a mixed strategy:

cost: 3 cost: 8 cost: 2

cost: 5 cost: 5 1/2 3, 0 0, 0

Usage:
1/2 0, 0 3, 0 σi denotes a
-10, -10 -7, -11 mixed strategy,
1, 0 1, 0 si denotes a pure
-11, -7 -8, -8 strategy

15 16 4
 03/12/2023

Iterated dominance “2/3 of the average” game revisited

• Iterated dominance: remove (strictly/weakly) dominated strategy, repeat 100
• Iterated strict dominance on Seinfeld’s RPS:
dominated

(2/3)*100
dominated after removal of
(originally) dominated strategies
(2/3)*(2/3)*100
0, 0 1, -1 1, -1
0, 0 1, -1 …
-1, 1 0, 0 -1, 1
-1, 1 0, 0
-1, 1 1, -1 0, 0
0

17 18

19 20 5
 03/12/2023

Two-player zero-sum games revisited

• Recall: in a zero-sum game, payoffs in each entry sum to zero Mixed Strategy
– … or to a constant: recall that we can subtract a constant from
anyone’s utility function without affecting their behavior
• What the one player gains, the other player loses • A mixed strategy for a player is a probability
distribution over some (or all) of her strategies.
• The strategy we have studied so far, i.e., taking
some action for sure, is called a pure-strategy.
Note: a general-sum k-player
0, 0 -1, 1 1, -1 game can be modeled as a zero-
sum (k+1)-player game by adding
• When the outcome of the game is uncertain, we
assume that each player maximizes expected value
a dummy player absorbing the
of her payoff.
1, -1 0, 0 -1, 1 remaining utility, so zero-sum
games with 3 or more players  Expected utility theory (von Neumann and
have to deal with the difficulties of
Morgenstern, 1944)
-1, 1 1, -1 0, 0
general-sum games; this is why
we focus on 2-player zero-sum
games here.

21 22

Best-response strategies How to play matching pennies

• Suppose you know your opponent’s mixed strategy Them
• E.g., your opponent plays rock 50% of the time and scissors 50% L R
• What is the best strategy for you to play?
• Rock gives .5*0 + .5*1 = .5
L 1, -1 -1, 1
Us
• Paper gives .5*1 + .5*(-1) = 0
R -1, 1 1, -1
• Scissors gives .5*(-1) + .5*0 = -.5
• So the best response to this opponent strategy is to (always) • Assume opponent knows our mixed strategy
play rock • If we play L 60%, R 40%...
• There is always some pure strategy that is a best response • … opponent will play R…
• Suppose you have a mixed strategy that is a best response; then
every one of the pure strategies that that mixed strategy places • … we get .6*(-1) + .4*(1) = -.2
positive probability on must also be a best response • What’s optimal for us? What about rock-paper-scissors?

23 24 6
 03/12/2023

Matching pennies with a sensitive target Matching pennies with a sensitive target
Them Them
L R L R
L 1, -1 -1, 1 L 1, -1 -1, 1
Us Us
R -2, 2 1, -1 R -2, 2 1, -1
• If we play 50% L, 50% R, opponent will attack L • What if we play 60% L, 40% R?
• We get .5*(1) + .5*(-2) = -.5 • Opponent has choice between
• What if we play 55% L, 45% R? • L: gives them .6*(-1) + .4*(2) = .2
• Opponent has choice between • R: gives them .6*(1) + .4*(-1) = .2
• L: gives them .55*(-1) + .45*(2) = .35 • We get -.2 either way
• R: gives them .55*(1) + .45*(-1) = .1 • This is the maximin strategy
• We get -.35 > -.5 • Maximizes our minimum utility

25 26

Let’s change roles General-sum games

• You could still play a minimax strategy in general-sum
Them games
L R • I.e., pretend that the opponent is only trying to hurt you
L 1, -1 -1, 1 • But this is not rational:
Us
R -2, 2 1, -1 0, 0 3, 1
• Suppose we know their strategy
von Neumann’s minimax
1, 0 2, 1
• If they play 50% L, 50% R,
theorem [1928]: maximin • If Column was trying to hurt Row, Column would play Left, so
• We play L, we get .5*(1)+.5*(-1) = 0
value = minimax value
• If they play 40% L, 60% R, (~LP duality) Row should play Down
• If we play L, we get .4*(1)+.6*(-1) = -.2 • In reality, Column will play Right (strictly dominant), so Row
• If we play R, we get .4*(-2)+.6*(1) = -.2 should play Up
• This is the minimax strategy • Is there a better generalization of minimax strategies in zero-
sum games to general-sum games?
27 28 7
 03/12/2023

29 30

Nash equilibria of “chicken”

S D

D S
D S
D 0, 0 -1, 1
S 1, -1 -5, -5
• (D, S) and (S, D) are Nash equilibria
– They are pure-strategy Nash equilibria: nobody randomizes
– They are also strict Nash equilibria: changing your strategy will make
you strictly worse off
• No other pure-strategy Nash equilibria

31 32 8
 03/12/2023

Nash equilibria of “chicken”…

D S
D 0, 0 -1, 1
S 1, -1 -5, -5
• Is there a Nash equilibrium that uses mixed strategies? Say, where player 1
uses a mixed strategy?
• Recall: if a mixed strategy is a best response, then all of the pure strategies
that it randomizes over must also be best responses
• So we need to make player 1 indifferent between D and S
• Player 1’s utility for playing D = -pcS
• Player 1’s utility for playing S = pcD - 5pcS = 1 - 6pcS
• So we need -pcS = 1 - 6pcS which means pcS = 1/5
• Then, player 2 needs to be indifferent as well
• Mixed-strategy Nash equilibrium: ((4/5 D, 1/5 S), (4/5 D, 1/5 S))
– Expected utility -1/5 for each player

33 34

The presentation game

Matching Pennies Again
Presenter
Put effort into Do not put effort into
presentation (E) presentation (NE)
Pay

Audience
attention (A) 4, 4 -16, -14 Player 2 Heads (p) Tails (1-p)
Player 1
Do not pay
attention (NA) 0, -2 0, 0 Heads 1 -1
(q) -1 1
• Pure-strategy Nash equilibria: (A, E), (NA, NE)
• Mixed-strategy Nash equilibrium:
Tails -1 1
((1/10 A, 9/10 NA), (4/5 E, 1/5 NE)) (1-q) 1 -1
– Utility 0 for audience, -14/10 for presenter
– Can see that some equilibria are strictly better for both players than other
equilibria, i.e., some equilibria Pareto-dominate other equilibria

35 36 9
 03/12/2023

How to Find Equilibrium? How to Verify Equilibrium?

• If a player takes both “Heads” and “Tails” with positive probability,
she must be indifferent between these two pure strategies, i.e., the
expected payoff derived by choosing Heads must be equal to that by
choosing Tails.
• -p+(1-p)=p-(1-p), hence p=0.5.
• q-(1-q)=-q+(1-q), hence q=0.5.
• Note that if p=0.5, Player 1 does not have a strict
incentive to change her strategy from q=0.5.
• Similarly, Player 2 does not have a strict incentive to
change his strategy from p=0.5, if q=0.5.
 Therefore, p=q=0.5 constitutes a mixed-strategy
equilibrium.
Q: Is this equilibrium reasonable/stable?
A: Yes (each player ends up randomizing two
strategies equally if the rival is smart enough).

37 38

Modified Matching Pennies Indifference Property

• Suppose the payoffs in the up-left cell changes as • Under mixed-strategy NE, Player 1 must be indifferent between
the following: choosing H and T:
 -2p+(1-p)=p-(1-p), hence p=0.4.
Player 2 Heads (p) Tails (1-p) • Similarly, Player 2 must be indifferent between choosing H and T:
 2q-(1-q)=-q+(1-q), hence q=0.4.
Player 1
• You can easily verify that (p,q)=(0.4,0.4) indeed becomes a mixed-
Heads 2 -1 strategy NE.

(q) -2 1
Tails -1 1
(1-q) 1 -1

39 40 10
 03/12/2023

Reporting a Crime Mixed-Strategy Equilibrium

• A crime is observed by a group of n people. Each
person would like the police to be informed but
prefers that someone else make the phone call.
• Players: The n people.
• Strategies: “Call” or “Don’t call”
• Payoffs:
0 if no one calls.
v if someone else calls but she does not.
v-c if the player calls.
Game Theory
• There is a symmetric mixed strategy equilibrium in
which each person calls with positive probability
p<1.
• In any such equilibrium, each person’s expected
payoff to calling is equal to her expected payoff to
not calling.
 v-c=v(1-Pr{no one else calls})
 c/v=Pr{no one else calls}
• Notice that the probability that no one else calls is
independent of n, but constant!
41 Game Theory 42

41 42

Implication Look for any equilibrium

Pr{no one calls}
=Pr{i does not call}xPr{no one else calls}
=(1-p)c/v •Dominating Equilibrium
• You can show that 1-p is increasing in n, i.e., the •Minimax Equilibrium
probability that person i does not call increases as n
becomes larger. •Nash Equilibrium
• This implies that the probability that no one calls is
also increasing in n.
 The larger the group, the less likely the police are
informed of the crime!

Game Theory 43

43 44 11
 03/12/2023

More Mixed strategies

Maximin & Minimax Equilibrium

• Minimax - to minimize the maximum loss (defensive)

• Maximin - to maximize the minimum gain (offensive)
• Minimax = Maximin x=probability to take
action R

y=probability to take
action S
x y 1-x-y 1-x-y=probability to take
no action P
no
No Nash equilibrium for pure strategy
46

45 46

Tennis Example

They have to be equal if expected payoff

independent of action of player 2

47

47 48 12
 03/12/2023

49 50

• Slides are obtained from the presentations of :

• Vincent Conitzer , Normal form games
• Bernhard Nebel, An Introduction to Game Theory Part I.
Strategic Games

51 52 13
 03/12/2023

53

14