Artificial Life

rucksackbulgeAI and Robotics

Dec 1, 2013 (3 years and 7 months ago)

76 views

Artificial Life

Artificial Intelligence

Department of Industrial Engineering
and Management

Cheng Shiu University

Outline


Introduction


Artificial Life


Simulation of Individuals


Simulation of population


Iterated Prisoners


Dilemma


N
-
Persons Iterated Prisoners


Dilemma


Norm Competition

Introduction


Christoph Adami suggested :



Life is a property of an ensemble of units that share
information coded in a physical substrate and which, in the
presence of noise, manages to keep its entropy significantly
lower than the maximal entropy of the ensemble, on
timescales exceeding the

natural


timescale of decay of the
(information
-
bearing) substrate by many orders of
magnitude.



To solve computational problem by biological skill


To solve biological problems by computational skill

Artificial Life (ALife)


To this point, we

ve used nature as the inspiration for
algorithms


Genetic algorithms


evolution


Ant colony algorithms


ant colonies


Particle swarm optimisation


flocking/swarming
behaviours


And we will look at artificial immune systems, based
on the human immune system


Artificial life is somewhat different


Computer systems simulating life



Artificial life is about better
understanding what it is to be alive.



Biology is primarily reductionist


an
explanation of a behaviour or
phenomenon at one level can be explained
by further investigation at the level below
(see left)



This is a reasonable top
-
down approach.



Artificial life takes a bottom
-
up approach.



Organism

Organs

Tissues

Cells

Organelles

Molecules


Study into Alife is conducted primarily at 3 levels


Wetware



using bits from biology (e.g. RNA, DNA) to
investigate evolution


Software

(what we have been/will be dealing with)


simulating biological systems


Hardware


for instance, robotics.



And with 2 distinct philosophies


Strong ALife



life is not just restricted to a carbon
-
based
chemical process. Life can be

created


in silico
.


Weak ALife



computer simulations are just that,
simulations and investigations of life



In fact, all the techniques we

ve seen so far can be
considered Artificial Life in so much as:


Genetic algorithms

are simulating or actually doing
evolution


Ant colony algorithms

are simulating the real behaviour of
ants


Particle swarm algorithms

are simulating the real behaviour
of flocks


What if we consider strong Alife?


Actual evolution, ants and flocks?


Almost certainly not, but what about a

life


Turing
Test?


We will be looking today at a software
-
based
technique


cellular automata.


One of the original Alife techniques, cellular
automata embodies the bottom
-
up approach


It is involved with the emergent behaviour of
collections of simple elements


Similar to the

emergent


behaviour seen in swarm
intelligence


These automata are mainly used for the simulation of
biological systems, although they can be used for
optimisation

Artificial Life


Emulation, simulation and construction of living
systems


Rules simple, Interaction complicated


Life seems to be a property of a collection of
components but not a property of the components
themselves
-

component interaction


Set components of lives evolved independently to
make the whole entropy decrease
-
>evolve a new life
style
-

competition leads to convergence


Simulation of units


Simulation of Population

Artificial Life V.S. GA Approach


No fitness function


Agents (Individuals) are part of environment


Agent compete to obtain the fitness value


Agents co
-
evolved by each other


Dynamic search space

up to Competition


Initialise population randomly, evolve it to
decrease entropy.


Simulation of Units


In the emulation of single living agents, the function
of just one or a few units.


Put together of units and simulate their connection to
form a life style which could work well.


Co
-
evolution and learning


Interaction among units


Interaction between individual and environment

Simulation of Population


Individuals:


simple rules

but
complicated interactions


Survival means wining the competition


Strategies encoded by simple rules


By encoded strategies, agents complete against others
and obtain the fitness.


Higher fitness means better strategy


Populations:


Initialised as a random one.


Evolved to be Converged finally, i.e., no agent can take
advantage of others.

Evolution of Cooperation


Evolution from survival competition

Selfish creature


While cooperation leads to better survival chances

.


Selfish evolution
->
emergence of cooperation


Selfish and rational evolution describe human
behaviors


Rational evolution leads to irrational behaviors


Simulate Human behaviors to see if cooperation
emerges.

The Iterated Prisoners


Dilemma


Cooperate

Defect


R


T

Cooperate

R


S



S


P

Defect

T


P



Payoff Scheme


T>R>P>S



R>(S+T)/2

The N
-
Player Iterated Prisoners


Dilemma






The payoff matrix must satisfy


D
i

> C
i

for


D
i+1

> D
i

and C
i+1

> C
i

for


C
i+1

> ( D
i+1

+ C
i
)/2 for


1
0



n
i
2
0



n
i
2
0



n
i
Number of cooperators
among the remaining n
-
1 players


0

1

2


n
-
1

C

C
0

C
1

C
2

……..

C
n
-
1



Player
A

D

D
0

D
1

D
2

……...

D
n
-
1


The payoff matrix for a single player in NIPD


The N
-
Player Iterated Prisoners


Dilemma






The payoff matrix must satisfy


D
i

> C
i

for


D
i+1

> D
i

and C
i+1

> C
i

for


C
i+1

> ( D
i+1

+ C
i
)/2 for


1
0



n
i
2
0



n
i
2
0



n
i
Number of cooperators
among the remaining n
-
1 players


0

1

2


n
-
1

C

C
0

C
1

C
2

……..

C
n
-
1



Player
A

D

D
0

D
1

D
2

……...

D
n
-
1


The payoff matrix for a single player in NIPD


Experimental Results on NIPD

1. Evolved by the GA
-
Roulette Wheel


13/25 achieve the mutual cooperation


0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
1
62
123
184
245
306
367
428
489
550
611
672
733
794
855
916
977
Generation
Amount of competition outcomes
0 cooperator
1 cooperator
2 cooperators
3 cooperators
4 cooperators
5 cooperators
6 cooperators
7 cooperators
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
1
69
137
205
273
341
409
477
545
613
681
749
817
885
953
Generation
Amount of competition outcome
0 cooperator
1 cooperator
2 cooperators
3 cooperators
4 cooperators
5 cooperators
6 cooperators
7 cooperators
The Market Competition Model


There are N players in the market.


High barrier of entrance for players outside.


Each player chooses high price/low price.


High price makes some customers move to
competitors with low price.


Only information for a player is his prices and
his competitors


at the previous rounds.


The Market Competition Model


The parameters:


PH: High price, the price that company chooses when it
cooperates with other companies.


PL: Low price, the price that company decided when it
defects on other companies.


a
: Market share lost, The percentage of customers
belonging to this company who will move to other low
-
price companies in next round when it adopts the high price.


b
: External effect function. It encapsulates overall market
growth or decline.


C: Cost of single product or service, Set as constant.

Player

s Profits in The Market


The calculation of profit:


Profit=(price
-
cost) X customer_number



When the player cooperates, price=PH, and when
companies defect , price=PL.


A cooperator loses
a

of customers.


Defectors share equally the customers who move out from
the cooperators.


The customers remain loyal when all the vendors choose
the same price.

Experimental Results on The
Market Competition

1. Pre
-
Test

3 Players' Game when (PH-C)/(PL-C)=5 and
a
=0.2
14
0
5
10
15
20
25
0
200
400
600
800
1000
Generation
Number of cooperation
0
100
200
300
400
500
600
700
800
900
1000
0
50
100
150
200
250
300
Generation
Number of outcome
0cooperator
1cooperator
2cooperator
3cooperator
4cooperator
Traffic Model


Two options for players:
Car

and
Bus
.


Player intends to increase his own utility.


Player

s strategy is based on utility at the last round.


More cars on the road decreases the utilities of all players.


Utility for players follows profit scheme(Linear):


Car traveller:
f(
t
)
=
b
-

a *
t


Bus traveller:
g(
t
)
=
d
-

c *
t


t
:

Number of Car Driver


Where
a

0 and
c

0.


-

b>d


-

a


c


50
100
80
60
67
40
50
60
70
80
90
100
110
0
10
20
30
40
50
60
70
80
90
100
Percentage of Car Drivers in population
Payoff
Payoff of Car Driver
Payoff of Bus Passenger

Strategies Represented in Chromosomes



Length of chromosome: (
p
+1) or (2p+1)



1:Cooperation/Bus chosen


0:Defection/Car adopted



The max. payoff as
B

and min. payoff as
S
.




S
+(
B
-
S
)/p * (
I
+1)


Interval

I

>
S
+ (
B
-
S
)/p * (
I
)



Profit at last round falls into an interval, player will adopt
corresponding transportation.


B

S

1

1

1

1

0

0

1

0

1

0

1

0

0

1

0

0

0

1

0

1

1



Nash Equilibrium point

66.7
40
50
60
70
80
90
100
0
10
20
30
40
50
60
70
80
90
100
Percentage of Car Drivers in population
Payoff
Payoff to Car Driver
Payoff to Bus Passenger
50
40
50
60
70
80
90
100
0
10
20
30
40
50
60
70
80
90
100
Percentage of Car Drivers in population
Payoff
Payoff to Car Driver
Payoff to Bus Passenger
33.3
40
50
60
70
80
90
100
0
10
20
30
40
50
60
70
80
90
100
Percentage of Car Drivers in population
Payoff
Payoff to Car Driver
Payoff to Bus Passenger
25
40
50
60
70
80
90
100
0
10
20
30
40
50
60
70
80
90
100
Percentage of Car Drivers in population
Payoff
Payoff to Car Driver
Payoff to Bus Passenger

A game in F1 of Experiment 3

70%
82%
81%
80%
81%
80%
81%
80%
81%
80%
0%
20%
40%
60%
80%
100%
Round 1
Round 2
Round 3
Round 4
Round 5
Round 6
Round 7
Round 8
Round 9
Round
10
% of Car Drivers
% of Car Drivers
0
1
2
3
4
5
6
7
8
9
10
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
Player
Number Choosing Car