Artificial life and artificial intelligence

40
Advanced Computational Modeli of Social Systems Lars-Erik Cederman and Luc Girardin Center for Comparative and International Studies (CIS) Swiss Federal Institute of Technology Zurich (ETH) http:// www.icr.ethz.ch/teaching/compmodels Artificial life and artificial intelligence

description

Artificial life and artificial intelligence. Today’s agenda. Complex Adaptive Systems Artificial intelligence Artificial life Genetic algorithms Foundation Genetic operators GeneIPD. Complexity theory. - PowerPoint PPT Presentation

Transcript of Artificial life and artificial intelligence

Page 1: Artificial life and artificial intelligence

Advanced Computational Modelingof Social Systems

Lars-Erik Cederman and Luc GirardinCenter for Comparative and International Studies (CIS)

Swiss Federal Institute of Technology Zurich (ETH)http://www.icr.ethz.ch/teaching/compmodels

Artificial life andartificial intelligence

Page 2: Artificial life and artificial intelligence

2

Today’s agenda

• Complex Adaptive Systems• Artificial intelligence• Artificial life• Genetic algorithms

– Foundation– Genetic operators

• GeneIPD

Page 3: Artificial life and artificial intelligence

3

Complexity theory

A model of the Internet

The Santa Fe Institute

“Boids”

Complex adaptive systems exhibit properties that emerge from local interactions among many heterogeneous agents mutually constituting their own environment

Page 4: Artificial life and artificial intelligence

4

Complex Adaptive Systems

A CAS is a network exhibiting aggregate properties that emerge from primarily local interaction among many, typically heterogeneous agents mutually constituting their own environment.

Emergent properties Large numbers of diverse agents Local and/or selective interaction Adaptation through selection Endogenous, non-parametric environment

Page 5: Artificial life and artificial intelligence

5

What is the mind?

• Marionettes (ancient Greeks)• Hydraulics (Descartes)• Pulleys and gears (Industrial Revolution)• Telephone switchboard (1930’s)• Boolean logic (1940’s)• Digital computer (1960’s)• Hologram (1970’s)• Neural networks (1980’s - ?)

Page 6: Artificial life and artificial intelligence

6

Bottom-up vs top-down approach

Intelligence Lifemodel this

model these

Page 7: Artificial life and artificial intelligence

7

Definition of life

• No universally agreed definition of life.• Typical features

– Self-organization– Emergence– Autonomy– Growth– Development– Reproduction– Adaptation– Responsiveness– Evolution – Metabolism

Page 8: Artificial life and artificial intelligence

8

Artificial life

• “By the middle of this century, mankind has acquired the power to extinguish life on Earth.”

• “By the middle of next century, it will be able to create it. Of the two it is hard to say which places the largest responsibility on our shoulders”

Chris Langton

Page 9: Artificial life and artificial intelligence

9

Artificial life

• Theoretical biology• Artifactual (man-made), not unreal• Bottom up, not top down• Synthesis, not analysis• Leverages emergence• “The ‘artificial’ in Artificial Life refers to the

component parts, not the emergent processes. If the component parts are implemented correctly, the processes they support are genuine—every bit as genuine as the natural processes they imitate.” (Langton)

Page 10: Artificial life and artificial intelligence

10

Boids

• Craig Reynolds (1987) work on flocking behavior

• Virtual birds with basic flight capability• 3 rules

– (i) collision avoidance – avoid collisions with nearby flock-mates

– (ii) velocity matching – attempt to match velocity with nearby flock-mates.

– (iii) flock centering – attempt to stay close to nearby flock-mates

• Each boid is a basic unit that “sees” only its nearby flock-mates and “flies” according to the 3 rules.

Page 11: Artificial life and artificial intelligence

11

Boids (cont.)

Page 12: Artificial life and artificial intelligence

12

Boids (cont.)

• Result: boids flocked and flew as a cohesive group. When obstacles appeared in their way they spontaneously split into 2 subgroups, without central guidance, and rejoined after clearing obstruction.

• Illustrate basic principles of Alife systems – Large number of simple elemental units– Units interacting with nearby neighbors with no

central controller– High-level emergent phenomena from low level

interactions

Page 13: Artificial life and artificial intelligence

13

Novelty

Page 14: Artificial life and artificial intelligence

14

Genetic algorithms

• Genetic algorithms (GA’s) are an optimization technique

• GA’s are based on Darwin’s theory of evolution

• Genetic algorithms combine search algorithms with the genetics of nature.  

• Invented by John Holland in mid 70’s

Charles Darwin 1809-1882

John Holland

Page 15: Artificial life and artificial intelligence

15

Biological terminology

• Cell nucleus contain the genetic information

• Genetic information is stored in the chromosomes

• The chromosome is divided in parts: genes

• The different settings of the genes for one property is called: allele

• Every gene has an unique position on the chromosome: locus

Page 16: Artificial life and artificial intelligence

16

Reproduction

• During sexual reproduction, crossover occurs

• This recombination of chromosomes become the new individual

• Hence genetic information is shared between the parents in order to create new offspring

• During reproduction “errors” occur: mutation

Page 17: Artificial life and artificial intelligence

17

Natural selection

• The Origin of Species: “Preservation of favourable variations and rejection of unfavourable variations.”

• There are more individuals born than can survive, so there is a continuous struggle for life.

• Individuals with an advantage have a greater chance to survive: survival of the fittest.

Page 18: Artificial life and artificial intelligence

18

How does it works?

Page 19: Artificial life and artificial intelligence

19

Why genetic algorithms?

• If nature can do it, why can’t computers?

• Power of evolution to solve optimization problems

• Particularly well suited for hard problems where little is known about the underlying search space

Page 20: Artificial life and artificial intelligence

20

Selection

• Main idea: better individuals get higher chance

• Roulette-wheel sampling gives more fit individuals better chance:

A C

1/6 = 17%

3/6 = 50%

B

2/6 = 33%

f(A) = 3

f(B) = 1

f(C) = 2

Page 21: Artificial life and artificial intelligence

21

1-point crossover

• Sexual reproduction is performed by mixing genes from the parents

• With a certain probability, we cross over some couples.

• Choose a random point on the two parents• Split parents at this crossover point• Create children by exchanging tails

Page 22: Artificial life and artificial intelligence

22

Mutation

• Occurs at the isolated gene level within a chromosome

• Can result in changes, both positive and negative, in the fitness of an individual

• When positive, can help increase individual’s chance of being selected to be a parent of next generation

Page 23: Artificial life and artificial intelligence

23

A standard genetic algorithm

1. Start with a randomly generated population of n chromosomes (genotypes)

2. Calculate the fitness developed by each chromosomes in the population

3. Repeat the following steps until n offspring have been created:a) Select a pair of parent chromosomes from the current population,

the probability of selection being an increasing function of fitness

b) Crossover the pair, with probability pc (crossover rate), at one or two randomly chosen points, to produce two new offspring

c) Mutate the offspring at each locus with probability pm (mutation rate), and place the resulting chromosomes in the new population

4. Replace the current population with the new population

5. Go to step 2 (next generation)

Page 24: Artificial life and artificial intelligence

24

Iterated Prisoner’s Dilemma

• Cohen, Riolo, and Axelrod. 1999. “The Emergence of Social Organization in the Prisoner's Dilemma” (SFI Working Paper 99-01-002)http://www.santafe.edu/research/publications/wpabstract/199901002

• In The Evolution of Cooperation, Robert Axelrod (1984) created a computer tournament of IPD– cooperation sometimes emerges– Tit For Tat a particularly effective strategy

Page 25: Artificial life and artificial intelligence

25

Prisoner’s Dilemma Game

Column: C D

C 3,3 0,5Row:

D 5,0 1,1

Page 26: Artificial life and artificial intelligence

26

One-Step Memory Strategies

C

D

C

D

p

q

Memory:C

D

Strategy = (i, p, q)i = prob. of

cooperating at t = 0

p = prob. of cooperating if opponent cooperated

q = prob. of cooperating if opponent defected

t-1 t

Page 27: Artificial life and artificial intelligence

27

The Four Strategies

Name i p q

ALLC 1 1 1

TFT 1 1 0

ATFT 0 0 1

ALJD 0 0 0

Page 28: Artificial life and artificial intelligence

28

TFT meets ALLD

t0 1 2 3 4

Row(TFT)

Column(ALLD)

i=1

i=0

p=1; q=0

0

5

1 1

1 1 1

1

= 8

= 3+ + +

+ + +

D

C

D

C

D

D

D

D

D

D

p=0; q=0

CumulatedPayoff

Page 29: Artificial life and artificial intelligence

29

Payoffs for 4x4 strategies

Own strategy

Other‘s strategyALLC TFT ATFT ALLD

pay/move

sum

pay/move

sum

pay/move

sum

pay/move

sum

ALLC 333312

333312

00000

00000

TFT 333312

333312

01539

01113

ATFT 555520

51039

13138

10001

ALLD 555520

51118

155516

11114

Page 30: Artificial life and artificial intelligence

30

Genetic encoding for IPD model

• For each of the possible historical cases – encode a move.

C D D C D C D D D C C D D C C C D C

DD D C D C D D C C C C C C C D C C

Player 1

Player 2

History

Sliding window (memory depth =

3)

Future

C

D

Page 31: Artificial life and artificial intelligence

31

Genetic encoding for IPD (cont.)

Initialphantom

move

Second phantom

move

Two-rounds memory

D C D C D C D C D C D C D C D C DC C

C

C

C

C

D

0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 01 1

D

D

D

D

D

D

Memory depth = 2

Page 32: Artificial life and artificial intelligence

32

GeneIPD

• Do cooperation emerge in a spatial world?

• Do defection emerge in a soup?• Is this always the case?• How does the crossover and

mutation rates affect the simulation?

Page 33: Artificial life and artificial intelligence
Page 34: Artificial life and artificial intelligence

34

Definition of life

• What is life? Is “artificial” life possible?• Webster dictionary

– 1a. The quality that distinguishes a vital and functional being from a dead body or purely chemical matter.

– 1b. The state of a material complex or individual characterized by the capacity to perform certain functional activities including metabolism, growth and reproduction

Page 35: Artificial life and artificial intelligence

35

Life versus Intelligence

• Life is…Islands of information that

persist and reproduce.

• Intelligence is…The use of information to persist

and reproduce.

Page 36: Artificial life and artificial intelligence

36

Alternative Crossover Operators

• Performance with 1-point crossover depends on the order that variables occur in the representation– More likely to keep together genes that are

near each other– Can never keep together genes from opposite

ends of string– This is known as Positional Bias– Can be exploited if we know about the

structure of our problem, but this is not usually the case

Page 37: Artificial life and artificial intelligence

37

n-point crossover

• Choose n random crossover points• Split along those points• Glue parts, alternating between parents• Generalization of 1 point (still some

positional bias)

Page 38: Artificial life and artificial intelligence

38

Uniform crossover

• Assign 'heads' to one parent, 'tails' to the other• Flip a coin for each gene of the first child• Make an inverse copy of the gene for the

second child• Inheritance is independent of position

Page 39: Artificial life and artificial intelligence

39

Genetic encoding for IPD (cont.)

• 1 move remembered:If CC (case 1), then CIf CD (case 2), then DIf DC (case 3), then CIf DD (case 4), then D

• Can be encoded as the string “CDCD” or “1010”

• For two possible moves remembered ->16 (4 x 4) possibilities (16 bits string)

• For three possible moves remembered ->64 (4 x 4 x 4) possibilities (64 bits string)

Tit-For-Tat

Page 40: Artificial life and artificial intelligence

40

Genetic encoding for IPD (cont.)

• But this is not enough, as the strategy requires results from previous games

• Extra genes to encode hypothetical moves:1 move memory: 4 + 1 = 5 bits2 moves memory: 16 + 3 =19 bits3 moves memory: 64 + 7 = 71 bits

• Number of potential strategies for 3 moves history is in the order of 1021