evolution of discrimination (experimental ethics) chung tue “duck” nguyen bse (hon.) project...
TRANSCRIPT
Evolution of Discrimination(Experimental Ethics)
Chung Tue “DuCK” NguyenBSE (Hon.) ProjectSupervisors: Ann Nicholson, Kevin Korb
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
2
Outline Introduction & Motivation & Main goal Related concepts Experimental Design
The world The agents
Observations & Results Conclusion References
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
3
Introduction & Motivation Discrimination:
The ability to recognise differences. The tendency to offer different
treatement on the basis of those differences.
Positive: telling useful from harmful. Negative: unfair treatement: racism,
sexism, etc. Does discrimination evolve?
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
4
Main goal
The goal of this study is:
To observe the evolution of discriminatory tendencies in intelligent species through social interactions.
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
5
Related concepts Evolutionary Theory
Natural selection – survival of the fittest Evolutionary Psychology
Evolution of behaviours/mental traits Baldwin Effect
Roles of learning in evolution Ethocentrism
Ingroup hyper-evaluation Outgroup negative streotype
Iterated Prisoner’s Dilemma (IPD) Alife Simulation
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
6
Experimental Design
Using simulations to study discrimination
Standard Alife setup: The world The agents
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
7
The world
A wrap-around board of n-by-m locations
Each location can contain an infinite number of agents
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
8
The agent
Chromosome: mutable genotype
Behaviour: flexible phenotype
Discrimination is modelled using chromosome and behaviour
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
9
Chromosome affects behaviour at birth.
Behaviour affects the tendency of possible actions.
Chromosome is modified through mutation.
Behaviour is modified through learning. Social learning (mimic) Individual learning (learning through
pain/pleasure)
Chromosome vs Behaviour
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
10
Chromosome, Behaviour and the Environment
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
11
Modelling discrimination Chromosome:
“discrimination gene” Behaviour:
“discriminatory behaviour”
Discrimination <-> Preference Preferential genotype Preferential behaviour Measurable
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
12
Preferences How positive/negative an agent “feels” about agents in
his own group/the other group
Represenation: two numbers, each between 0 and 1
Two types of preferences: Ingroup preference: towards the ingroup Outgroup preference: towards the outgroup E.g.: Ingroup preference = 0.7, outgroup preference = 0.4
Preferential behaviours <-> discriminatory actions
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
13
Preference test Test for preference (ingroup/outgroup):
Choose randomly: x in [0,1] Compare x with p x <= p : like x > p : dislike
x =Test(preference)
Like Dislike
successfailure
p0 1
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
14
Main genotypes Based on the values of preferences
(chromosome), there are 5 main genotypes, each may evolve differently.
Out-group-pref.
>0.5 <0.5 =0.5
In-group-pref.
>0.5 Do-gooder Supremacist Supremacist
<0.5 Betrayer Hater Betrayer
=0.5 Do-gooder Hater Moderate
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
15
The agent (cont.)
Other traits: Race (can also be: religion, ideology, etc.) Health Age Location Fitness = #children + 2 x #grandchildren
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
16
A typical cycle During 1 cycle of the simulation,
every agent can perform all actions, in the following order: Move Learn socially (mimic the best) Mate + reproduce Interact + learn through pain and
pleasure Note: Agents don’t eat, health is affected
directly through interaction
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
17
Preferences in action - Moving Agents can move in four directions
Agents choose their location based on how much they like the majority of agents in a particular location.
Agents considers their current location as well as four cells around them.
Considerations are done in random order
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
18
Moving example
In_pref = 0.8Out_pref = 0.3Majority = RedTest(Out_pref) = 0.4Result: 0.4 > 0.3 => Failed
In_pref = 0.8Out_pref = 0.3Majority = BlueTest(In_pref) = 0.6Result: 0.6 > 0.8 => Success
Note: much more likely to move in with its own kind
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
19
Preference in action – Social Learning (mimicking)
Agents learn from other agents in the same location
Agents learn from the race they like first Agents learn from the agents the see as
most successful (the idol) Success = fitness x (maxAge – age) Mimicking:
Depends on social learning success rate Change preferences towards the idol’s preferences
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
20
Preference in action – Mating
There is no gender Inter-racial mating Mating is consensual
Consent is given on the basis of racial preference
I.e. both partners must pass the preference test
Mating hapens locally
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
21
Reproduction When mating is successful, a new agent is
created in the same location as its parents New agent has a 50-50 chance of getting:
Either parent’s race Either parent’s ingroup preference
(chromosome), or mutate Either parent’s outgroup preference
(chromosome), or mutate A small chance of mutation:
Change the value to any x in [0,1]
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
22
Reproduction example
Chromosome: 0.3/0.5Behaviour:0.6/0.9
Chromosome: 0.6/0.4Behaviour:0.3/0.5
Chromosome: 0.6/0.4Behaviour:0.3/0.7
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
23
Preference in action – Interaction & Indiv. Learning
Every agent interacts with another agent in the same location, randomly chosen.
Interaction is essentially an IPD* game: Each agent does a preference test on the other If success: the agent cooperates If failure: the agent defects Agent’s health is updated accordingly
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
24
Interaction trade-offs
Cooperate Defect
Cooperate 3 | Pleasure -3 | Pain
Defect 4 | Pleasure -2 | Pain
A typical IPD is characterised with the following order of trade-offs:1. Defect – Cooperate2. Cooperate – Cooperate3. Defect – Defect4. Cooperate – Defect
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
25
Learning through pleasure and pain
Agents change their preference according to the result of their interaction with another agent.
If Pleasure: increase preference If Pain: decrease preference Preference test also applies:
Only increase preference if success Only decrease preference if failure
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
26
Experiments
Difference between genotypes Relationship between fitness and
preference Relationship between ingroup and
outgroup preference Population development and
geographical distribution of the races
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
27
Experimental modes
Separated communities: Two races start in isolation
Mixed communities: Two race mixed together from the start
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
28
Running the simulation
Run 30 simulations in each experimental mode
Each simulation is 1000-cycle long Starting population is 7000 World size: 50x50 All agents start off as “Moderate”:
In_pref = Out_pref = 0.5
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
29
Results - Overview The genotypes did evolve Order of fitness (most -> least):
Supremacist Do-gooder Hater Betrayer
Ethnocentric behaviours Most affected by the Supremacist and
Do-gooder
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
30
Results - Overview Separated communities different from
mixed communities: Separated:
Less variations in genotype More variations in races Complete domination + extinction
Mixed: More variations in genotype Less variations in races Domination changes hands
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
31
Results - Experiment 1
Difference between genotypesMixed communitiesSeparated communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
32
Results - Experiment 1Mixed communitiesSeparated communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
33
Results – Experiment 2 Relationship between fitness and preference
Correlations:Inpref-Fitness = 0.887, p = 0.01Outpref-Fitness = -0.519, p = 0.01
Correlations:Inpref-Fitness = 0.920, p = 0.01Outpref-Fitness = -0.660, p = 0.01
Mixed communitiesSeparated communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
34
Results – Experiment 3 Relationship between ingroup and outgroup
preference
Correlations:Inpref-Outpref = -0.750, p = 0.01
Correlations:Inpref-Outpref = -0.951, p = 0.01
Mixed communitiesSeparated communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
35
Results – Experiment 3
Correlation:Inpref-Outpref = 0.736, p = 0.01
Correlations:Inpref-Outpref = -0.774, p = 0.01
Separated communities Mixed communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
36
Results – Experiment 4 Population development and geographical
distribution of the races
Mixed communitiesSeparated communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
37
Results – Experiment 4
Mixed communitiesSeparated communities
Clear domination: 70%Extinction: 13.3%
Clear domination: 46.6%Extinction: 0%
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
38
Results – Experiment 4
Separated communities
Mixed communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
39
Results – Experiment 4
Separated communities
Mixed communities
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
40
Results – Experiment 4
More geographical distribution if time permits…
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
41
Conclusion Discrimination can be demonstrated to be
evolutionary A successful discriminatory strategy includes a
reasonable level of cooperation and defection Ingroup cooperation has more impact on fitness than
outgroup defection: don’t have to love thine neighbours but don’t mess up with thine family
Ethnocentric behaviours may have been evolutionary Mixed communities allow more variation and balanced
competition. Think: multiculturalism, immigration…
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
42
Futher work More + longer simulation
Race domination will stabilised or lead to extinction eventually?
Increase separation space/time to observe development before and after contact (in separated communities mode)
More efficient simulation Find a better way to observe geographical
distribution of genotypes in 3D. Thesis
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
43
References
Refer to References section in the Thesis
25/10/2006 Evolution of Discrimination - Chung Tue "DuCK" Nguyen
44
Questions?
Thank you.