Top Banner
March 11, 2003 Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.
47

March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

Dec 13, 2015

Download

Documents

Blaze Webb
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003 Hridesh Rajan1

Humans – Actual, Imagined & Implied

From Swarm Intelligence by Kennedy Et. Al.

Page 2: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan2

Before We Begin…

This chapter is a survey of psychological and sociological efforts relating it to swarm behaviors in some ways.

In “my opinion” the author did not relate well bits and pieces of the loosely bound sections of this chapter to the main theme of the book.

I have almost “NO” background in psychology and sociology.

Page 3: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan3

Roadmap

Behavioral psychology to cognitive psychology. Simulating social influence. Culture.

Page 4: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan4

Behavioral Psychology

BEHAVIORAL PSYCHOLOGY is the subset of psychology that focuses on studying and modifying observable behavior by means of systematic manipulation of environmental factors.

From: www.killology.com/article_behavioral.htm

Page 5: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan5

Behaviorist’s Doctrine

Classical conditioning

- Organisms passively

react to events in the

environment.

- Organism pushed by

the stimuli.

[Pavlov]

Operant conditioning

- Organisms act on the

environment to obtain

a reinforcement.

- Organism pulled

towards a stimulus.

[ Watson, Hull and Skinner]

Page 6: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan6

From Behavioral To Cognitive

Page 7: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan7

Cognitive Psychology

HumanStimuli Response

There is more to human behavior than just stimuli and responses.

Page 8: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan8

How Are Stimuli and Response Related?

How to explain why small rewards had greater effect?

How to explain variation in problem solving time?

How to explain social learning?

Page 9: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan9

Vicarious Reinforcement

Learn a task without actually doing it. Key is to watch someone else do it. “Bobo doll” experiment. [Bandura 1962].

Page 10: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan10

Further on the Cognitive Way

Gestalt process. Lewin’s field theory. Sociocognitive efforts in parallel to behavioral

and cognitive psychologists.

Page 11: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan11

Gestalt’s Prägnanz /Good Form

Tendency to organize perceptions into coherent wholes.

Permits partitioning the environment into recognizable objects.

These objects can now be processed to generate responses, whereas it was not possible to process the environment.

Page 12: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan12

Lewin’s Field Theory

Life space consists of individuals. Individuals denoted by bounded regions. They act upon others and are acted upon. A person can be divided into number of

separate but interconnected regions. An environment can be divided into number of

separate regions.

Page 13: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan13

Lewin’s Field Theory

A person can be part of another person’s environment (society).

A person can move through the life space, which is called locomotion.

Regions that are interconnected influence each other and this can cause locomotion to achieve equilibrium.

Page 14: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan14

Social Influence

[Sherif 1936] experiment showed that the individual behavior tend to drift towards the norm of the group.

[Asch 1965] showed a similar result of peer pressure with human confederates.

[Crtutchfield 1955] and [Deutsch and Gerard 1955] used automated confederates.

Page 15: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan15

Sociocognition

Thinking is a social activity. Coordinated cognitive activities evoke

intersubjectivity(shared understanding).

Page 16: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan16

Social Aspects to Memory

Transactive memory: using people you know well as references to encode, store and retrieve memories. [ Dan Wegner at UVA]

Content: past social actions and experiences. Symbolic communication.

Page 17: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan17

Introducing Simulations as Evaluation strategy

Page 18: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan18

Simulating Social Influence

Inability to distinguish between a phenomenon and a simulation of the phenomenon.

Making edible sculpture of food. Simulating mind. Imitative social behavior. Induced compliance paradigm [ you are free to

do what ever you wish, but if I were you …].

Page 19: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan19

Prisoner’s Dilemma[AxelRod 84]

Two player competition or co-operation. Both co-operate -> high payoff. Both compete -> low payoff. One compete other co-operate, competing

players payoff is high whereas co-operating players payoff is abysmally low.

[This was the first computer experiment widely accepted by social psychologists].

Page 20: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan20

Solutions

TIT-FOR-TAT (Winner) TIT-FOR-TWO-TATS. LOOKAHEAD DOWNING (Simulates behavior seen in human

subjects in the situation).

Page 21: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan21

Hutchin’s Network

Each person is represented as a parallel constraint satisfaction network.

Positive interconnection of a node of such network to corresponding node of other network signifies belief communication.

Hutchin’s example had two globally optimal solution.

Page 22: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan22

Hutchin’s Network

Positive LinkNegative Link

Page 23: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan23

Hutchin’s Network Results

When the nodes are highly connected, it results in sub-optimal pattern. (Striking similarity to social influence results.).

When the nodes are isolated, nothing special. When the nodes are moderately connected

optimal pattern is reached.

Page 24: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan24

Inferences

Moderate ignorance not only permits cognitive consistencies, but agreement among members of a group.

“..We are always in a negative state of knowledge, ignorance.”

Page 25: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan25

Culture

Page 26: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan26

Coordination Games

Prisoner’s dilemma revisited. Game of chicken Battle of the sexes El Farol problem

Page 27: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan27

Game of Chicken

Choices

• Swerve

• Stay on the Road 0

06

18

12

12

6

18

b2

b1

a1 a2

Page 28: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan28

El Farol Problem [Arthur 1994]

N people decide independently to go to the bar. Optimal situation: present <= 0.6 * population. Choices are unaffected by previous visits; there is no

collusion or prior communication among the agents. Only information available is the numbers who came in past

weeks. There is no deductively rational solution--no "correct"

expectational model.

Page 29: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan29

El Farol Problem

Each of the agent can form k hypotheses of the form [f(present1,…,presentn) => present0].

Each agent decides to go or stay according to the currently most accurate predictor in his set.

Once decisions are made, each agent learns the new attendance figure, and updates the accuracies of his monitored predictors.

Page 30: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan30

El Farol Problem

From: http://www.santafe.edu/arthur/Papers/El_Farol.html

Page 31: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan31

Sugarscape[Epstein & Axtell]

Artificial society emulator.- Seeded with population, an environment and rules.

- Can be used to test whether certain phenomenon of economics are necessary outcomes of dynamic principles.

- What parameters affect the pattern of observed behavior?

Page 32: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan32

An Epidemiological Model in Sugarscape

Germs are coded as bit strings of length 5 example: 10110.

Immune system of an agent is coded as bit string of length 50.

If disease agent bit string is a sub-string of immune system bit string agent is immune to that disease.

Page 33: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan33

An Epidemiological Model in Sugarscape

Agents are allowed to propagate genetically, evolving immune system.

Fitness function: how well does the immune system protects agents from diseases in the population.

Page 34: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan34

An Epidemiological Model in Sugarscape

Disease are spread from agent to agent when they interacted.

An interesting observation was that the immune system could evolve that was shorter than the sum of lengths of antigen it guarded against.

Page 35: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan35

ACE [Tesfatsion]

Key observation:

“ In the real world people choose whom to talk with, whom to interact with, whom to do business with, ..”

Objective of the project is to understand how coordination arises in decentralized systems.

Page 36: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan36

Competing-Norms Model [Picker 97]

One kind of behavior might remain prevalent even if a superior behavior is available.[e.g. people doing dangerous things, ignoring threats, starving themselves in the name of beauty.]

It is not easy to enforce laws that contradict popular ways of doing things.[ American drug prohibition]

Page 37: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan37

Mutable Prisoner’s Dilemma

An individual agent is represented as a cell in CA and it plays repeated games with the the members of its payoff neighborhood.

An individual agent also belongs to an information neighborhood, with whom it gathers feedback about their strategies and success.

Choice of strategy is randomly assigned for the first round.

Page 38: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan38

Results

Populations almost always converge to a unanimity.

Usually they converge on the best strategy, but when the relative benefit fall below threshold or when the initial best population was too low, the population converged on the inferior choice.

Page 39: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan39

Inference

By just communicating in the local neighborhood it is possible to optimize even a very complex decision function.

Page 40: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan40

Ringelmann Effect [Ringlemann 1913]

Number of People

Tot

al S

tren

gth

Page 41: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan41

Dynamic Social Impact Theory [Latane 1981]

Probability of any individual helping someone in need decreases as number of people present increases. [ With John Darley]

Group influence is proportional to the strength, immediacy, and the number of group members.

Polarization: Individuals in a population resemble their neighbors, whereas regions of population differ from each other.

Page 42: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan42

Dynamic Social Impact Theory

Consolidation- the diversity of opinion reduces as individuals are exposed to a preponderance of majority arguments.

Clustering- people become more similar to their neighbors in social space.

Correlation- attitudes that were originally independent tend to become associated.

Continued diversity - clustering protects minority views from complete consolidation.

Page 43: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan43

Evolutionary Culture Model [Boyd and Richerson 1985]

Some part of the human behavior is determined by genetics, on the other hand, our genes predisposes us to behave socially in a way that results in culture.

Much of our behavior is acquired by imitation, through a process called cultural transmission.

Page 44: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan44

Evolutionary Culture Model

When is individual learning more adaptive?

Environment is relatively homogeneous and stable over time so that generalization is possible.

When is social learning more adaptive?

Environment is diverse and each time an individual samples it, different results are obtained, so individual can only attain a comprehensive view by learning from other’s experience.

Page 45: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan45

Types and Aspects of Cultural Transmission

Basic facts: - Genotype is the genetic coding of an organism. - Phenotype is the expression of the genotype. - Phenotype develops through the interaction of the

genes with the environment. - Some phenotypes are more variable than others. - Some phenotypes depend more on the environment for

expression; e.g. phenotype freckles do not appear unless the person spends some time in sun.

Page 46: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan46

Boyd and Richerson’s Observation

Human phenotype expression of behavior depends on two type of learning – learning derived from cultural norms that the person is exposed to and the learning acquired through individual experience.

Upon evolution, individual’s adaptations - and their subsequent probability of survival and reproduction – depended jointly on their individual experience and on what they learned from society.

Further tendency to learn more in one way or the other was also genetically evolved.

Page 47: March 11, 2003Hridesh Rajan 1 Humans – Actual, Imagined & Implied From Swarm Intelligence by Kennedy Et. Al.

March 11, 2003Hridesh Rajan47

Without a Culture

“A man who has been alone since birth will have no verbal behavior, will not be aware of himself as a person, will possess no techniques of self management, and with respect to the world around him will have only those meager skills which can be acquired in one short lifetime from nonsocial contingencies.. To be for oneself is to be almost nothing.”

– Skinner 1971.