Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further...

50
Replicator Dynamics 1

Transcript of Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further...

Page 1: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Replicator Dynamics

1

Page 2: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Nash makes sense (arguably) if…

-Uber-rational

-Calculating

2

Page 3: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Such as Auctions…

3

Page 4: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Or Oligopolies…

Image courtesy of afagen on Flickr. CC BY NC-SA Image courtesy of longislandwins on Flickr. CC-BY

4

Page 5: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

But why would game theory matter for our puzzles?

5

Page 6: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Norms/rights/morality are not chosen; rather…

We believe we have rights!

We feel angry when uses service but doesn’t pay

6

Page 7: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

But…

From where do these feelings/beliefs come?

7

Page 8: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

In this lecture, we will introduce replicator dynamics

The replicator dynamic is a simple model of evolution and prestige-biased learning in games

Today, we will show that replicator leads to Nash

8

Page 9: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

We consider a large population, N, of players

Each period, a player is randomly matched with another player and they play a two-player game

9

Page 10: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Each player is assigned a strategy. Players cannot choose their strategies

10

Page 11: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

We can think of this in a few ways, e.g.:

• Players are “born with” their mother’s strategy (ignore sexual reproduction)

• Players “imitate” others’ strategies

11

Page 12: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Note:

Rationality and consciousness don’t enter the picture.

12

Page 13: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Suppose there are two strategies, A and B.

We start with:

Some number, NA, of players assigned strategy A

And some number, NB, of players assigned strategy B

13

Page 14: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

We denote the proportion of the population playing strategy A as XA, so:

xA = NA/N

xB = NB/N

14

Page 15: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

The state of the population is given by

(xA, xB) where xA ≥ 0, xB ≥ 0, and xA + xB = 1.

15

Page 16: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Since players interacts with another randomly chosen player in the population, a player’s EXPECTED payoff is determined by the payoff matrix and the proportion of each strategy in the population.

16

Page 17: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

For example, consider

the coordination game:

a > c

b < d

And the following starting frequencies:

xA = .75

xB = .25

a, a b, c

c, b d, d

A

B

A B

17

Page 18: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Payoff for player who is playing A is fA

Since fA depends on xA and xB we write fA(xA, xB)

fA(xA, xB) = (probability of interacting with A player)*UA(A,A)

+ (probability of interacting with B player)*UA(A,B)

= xA*a + xB*b

= .75*a + .25*b

18

Page 19: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

We interpret payoff as rate of reproduction (fitness).

19

Page 20: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

The average fitness, f, of a population is the

weighted average of the two fitness values.

f(xA, xB) = xA*fA(xA, xB) + xB*fB(xA, xB)

20

Page 21: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

How fast do xA and xB grow? Recall xA = NA / N First, we need to know how fast does NA grows Let NA = dNA/dt Each individual reproduces at a rate fA, and there are NA of them. So: NA = NA * fA(xA, xB) Next we need to know how fast N grows. By the same logic: N = N * f(xA, xB) By the quotient rule, and with a little simplification…

21

Page 22: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

xA = xA * (fA(xA, xB) – f(xA, xB))

Current frequency of strategy

Own fitness relative to the average

This is the replicator equation:

22

Page 23: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

xA = xA * (fA(xA, xB) – f(xA, xB))

Current frequency of strategy

Because that’s how

many As can reproduce

Own fitness relative to the average

This is our key property.

More successful strategies grow faster

Growth rate of A

23

Page 24: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

xA = xA * (fA(xA, xB) – f(xA, xB))

If: xA > 0: The proportion of As is non-zero fA > f: The fitness of A is above average Then: xA > 0: A will be increasing in the population

24

Page 25: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

The steady states are

xA = 0

xA = 1

xA such that fA(xA,xB) = fB(xA,xB)

25

Page 26: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

a, a b, c

c, b d, d

A

B

A B Recall the payoffs of our (coordination) game:

a > c b < d

26

Page 27: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

f A (

x A,x

B)

– f B

(xA

,xB)

xA = 0 xA = 1

= “asymptotically stable” steady states i.e., steady states s.t. the dynamics point toward it

27

Page 28: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

What were the pure Nash equilibria of the coordination game?

28

Page 29: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

a, a b, c

c, b d, d

A

B

A B

29

Page 30: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

0 1

30

Page 31: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

And the mixed strategy equilibrium is:

xA = (d – b) / (d – b + a – c)

31

Page 32: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

0 1 (d – b) / (d – b + a – c)

32

Page 33: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Replicator teaches us:

We end up at Nash

(…if we end)

AND not just any Nash

(e.g. not mixed Nash in coordination)

33

Page 34: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Let’s generalize this to three strategies:

R

P

S

34

Page 35: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Now…

NR is the number playing R

NP is the number playing P

NS is the number playing S

35

Page 36: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Now…

xR is the proportion playing R

xP is the proportion playing P

xS is the proportion playing S

36

Page 37: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

The state of population is

(xR, xS, xP) where xR≥0, xP≥0, xS≥0,

and xR + xS + xP = 1

37

Page 38: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

R P S

P

R

S

0

1

-1

-1

-1

1

1

0

0

For example, Consider the Rock-Paper-Scissors Game: With starting frequencies: xR = .25 xP = .25 xS = .5

38

Page 39: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Fitness for player playing R is fR

fR(xR,xP,xS) = (probability of interacting with R player)*UR(R,R)

+ (probability of interacting with P player)*UR(R,P)

+ (probability of interacting with S player)*UR(R,S)

= .25*0 + .25*-1 + .5*1

= .25

39

Page 40: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

In general, fitness for players with strategy R is:

fR(xR,xP,xS) = xR*0 + xP*-1 + xS*1

40

Page 41: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

The average fitness, f, of the population is:

f(xR,xP,xS) = xR*fR(xR,xP,xS) + xP*fP(xR,xP,xS)

+ xS*fS(xR,xP,xS)

41

Page 42: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

xR = xR * (fR(xR,xP,xS) – f(xR,xP,xS))

Current frequency of strategy

Own fitness relative to the average

Replicator is still:

42

Page 43: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

(xR=.33, xP=.33,xS=.33)

xR=1

xP=1

xS=1

xR=.5, xP=.5 xS=.5, xP=.5

xR=.5, xS=.5

43

Page 44: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

44

A

CB

Image by MIT OpenCourseWare.

Page 45: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Notice not asymptotically stable

It cycles

Will show this in HW

45

Page 46: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

R P S

P

R

S

0

2

-1

-1

-1

2

2

0

0

46

Page 47: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

47

A

B C

Image by MIT OpenCourseWare.

Page 48: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

Note now is asymptotically stable

Will solve for Nash and show this is what dynamics look like in HW

48

Page 49: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch. 3 Some notes: Can be extended to any number of strategies Doesn’t always converge, but when does converges to Nash We will later use this to provide evidence that dynamics predict behavior better than Nash

49

Page 50: Replicator Dynamics - MIT OpenCourseWare · PDF filedynamics look like in HW . 48. For further readings, see: Nowak Evolutionary Dynamics Ch. 4 Weibull Evolutionary Game Theory Ch.

MIT OpenCourseWarehttp://ocw.mit.edu

14.11 Insights from Game Theory into Social BehaviorFall 2013

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.