An Introduction to Complexity in Social Science

1
An Introduction to Complexity in Social Science
Bernard PAVARD IRIT-CNRS Univ. P. Sabatier
2
The COSI Network

• Goal
• Understanding and modelling socio-cognitive
  processes in the context of real organisational
  systems.
• Aims
• To increase awareness of the use of complexity
  theory in social science.
• To promote the new culture of pluri-disciplinary
  research applied to concrete industrial problems
• To initiate an emergent European research
  movement in the domain of the simulation of
  social science.
• Main themes of the project
• Complexity
• Socio-organisational systems
• Modelling Simulation

3
The European COSI Team

• ARAMIIHS-GRIC Laboratory, Toulouse, France.
• Kings College, UK
• University of LIEGE Bruxelles, Belgium
• University of Birmingham (was De Montfort
  University), UK
• University of SIENA, Italy
• National Technical University of Athens NTUA,
  Greece
• University of Granada, Spain
• University of Lisbon, Potugal
• Insitituto Technologico Y De Estudios Superiores
  de Monterrey, Mexico
• Universidad Federal de Rio de Janeiro, Brazil

The non-European COSI Team
4
Partners have been chosen for their expertise in
different theoretical or methodological domains
COSI Network

• Nat. Technical Univ of Athens (N. Marmaras)
• Cognitive ergonomics
• Univ. of Granada (J. J. Merelo)
• Computer modelling simulation
• Univ. of Lisbon (H. Coelho)
• Agent modelling
• Inst. Tec. Mexico (R. Zorola)
• Cognitive simulation modelling
• Univ. of Rio de Janeiro (M. Vidal)
• Ergonomics

• ARAMIIHS-GRIC (B. Pavard)
• Cognitive engineering complexity modelling
• Kings College, London (C. Heath)
• Ethnomethodology
• Univ. Liege Bruxelles (F. Decortis P.
  Nardone)
• Cognitive psychology ergonomics
• Complexity modelling implementation
• Univ. of Birmingham (J. Rowe)
• Complex adaptive systems
• Univ. of Siena (A. Rizzo)
• Cultural psychology distributed cognition

5
COSI Work Tasks

• WT0. Management of the partner network.
• Development of Web site, General administration.
  Responsibility ARAMIIHS GRIC
• WT1. Assimilation of Complexity Paradigm
• Assessment of different approaches to complexity.
• Aided by simple examples, tutorials etc. on web
  site written papers.
• Contributors ALL Partners. Responsibility
  Birmingham
• WT2. Identification of specific work situations
  Field Studies
• Use industrial contacts
• Contributors Liege, Kings, Athens, Aramiihs,
  Sienna, Brazil. Responsibilty Liege
• WT3. Model Development
• Translate data from field studies into models for
  simulation.
• Contributors ALL (need close collaboration
  between Alife S.Science teams) Responsibilty
  Spain
• WT4. Simulation for Calibration Experimentation
• Use data from field studies for validation.
• Contributors All Responsibilty Athens
• WT5. Assessment of Complexity Approach in
  Designing Organisational Systems
• Output Symposium
• Contributors Aramiihs-GRIC, Liege, Sienna,
  Athens. Responsibilty Siena

6
1. Definition of Complexity

• A complex system is a system for which it is
  difficult, if not impossible to restrict its
  description to a limited number of parameters or
  characterising variables without losing its
  essential global functional properties.

7
Complicated vs Complex system

• A car is a complicated system
• Making a good omelet is complex
• A jumbo is a complicated system
• The stock exchange is a complex system
• But three planets interacting altogether is a
  complex system

8
Few examples of complexity in social science
9
Complexity and reliabilityin ATC

• The paradox of reliability in complex work
  settings
• Control in cooperative work settings

10
The Air Traffic Control Desk
11
How ordinary errors are opportunistically
handled by controlers?

• Cr MON 598  Turn left 120 
• Co  Its amazing but when I turn left, I do not
  behave like that 
• Co forget to update the flight strip
• Co forget to move back the LTU130
• Co-référence error
• Co  à lAF, tu ne lui a pas donné 15 degrés?
• Cr Non
• Cr LTU 130 vous pouvez reprendre votre route
  sur Angers 

MON598
LTU130
12
The context is always changing and difficult to
handle analytically
From P. Salembier  Cognition(s) Située
distribuée, socialement partagée, etc, etc. 
13
Artefacts may drastically influence socio
cognitive mechanisms
14
Complex systems are open (no  operational
closure )

• It is often supposed that a system is supposed to
  change only inside its own frontier (this is the
  definition of operational closure)
• The frontier between a complex system and the
  environment is not always easy to identify
• Frontiers depend of the observer, actors and
  context

15
The limits of the classical (analytical) approach

• Over simplification
• Complexity of the real world cannot be
  represented in terms of a limited set of rules
• Interaction with the environment is too limited
• History of the system not enough taken into
  account

16
Approaches in social sciences to escape the
drawbacks of analytical approaches

• Theory of internal external representations
  (Zhang Norman, 1994)
• Ethnomethodology conversational analysis
  (Garfinkel, Sacks, Schegloff and Jefferson, Heath
  Luff, 1994)
• Distributed situated cognition (Suchman, 1990
  Hutchins, 1995)
• Complexity theory
• Chaos theory (H. Poincaré)
• Distributed and self organised systems
• Dynamic of cognition
• Action theory (Pierce, Theureau, 1992)
• Autonomy, autopoiese (Varéla, 1989)
• Activity theory (Léontiev, Vygotsky, Kuutti,
  Engeström)
• ²

17
Some Properties of Complex Systems

• Property One
• Non-determinism and non-tractability. A
  complex system is fundamentally
  non-deterministic non-tractable.
• Property Two
• Limited functional decomposability.
• Property Three
• Distributed nature of information and
  representation.
• Property Four
• Emergence and Self-organisation.

18
Property OneNon-determinism non tractability
A complex system is usually non-deterministic
non-tractable

• In this example, the Medic (Med) is telephoning
  an external agent (C).
• Due to the proximity relationship between the
  medic and all other agents, the conversation is
  opportunistically listened to by the agent O who
  then sends an ambulance (because she inferred the
  case discussed by the medic was urgent)

19
Property TwoLimited functional decomposibility

• Plasticity in the division of Labour in Social
  Insects
• Different activities are often performed
  simultaneously by specialised individuals, but
  the division is rarely rigid.
• Workers switch tasks to adjust to changing
  conditions maintaining the colonys variability
  and reproductive success.
• Factors which cause the change in role are due to
  internal colony perturbations or external
  challenges, e.g food availability, predation,
  climate change.

External Influences (e.g. food availability,
predation, climatic change, etc.)
20
Interaction between an operator and its
environment are sometimes complex and cannot
support functional decomposability
In this example, the agent in white (a doctor) is
dynamically controlling the emitting power of the
radio loudspeaker in order to selectively
broadcast information to other people in the room.
21
Property ThreeDistributed nature of information
and representation

• A system is said to be distributed when its
  resources are physically or virtually distributed
  on various sites.
• i. Repartition
• ii. Redundancy
• iii. Robustness

22
Different meanings of distributed systems

• Physically distributed resources
• Distributed cognition (over artefacts, agents)
• Ubiquitous distribution of information (neural
  networks) which bring robustness

23
Property FourEmergence and Self-organisation

• Emergence is the process of deriving some new and
  coherent structures, patterns and properties in a
  complex system.
• Emergent phenomena occur due to the pattern of
  interactions between the elements of the system
  over time.
• Emergent phenomena are observable at a
  macro-level, even though they are generated by
  micro-level elements.
• Explore emergence with this interactive essay
  from MIT http//llk.media.mit.edu/projects/emergen
  ce/index.html

24
2. History of Complex Systems

• Henri Poincaré showed a new conceptual
  difficulty how a completely causal system could
  have indeterminate behaviour (1889)
• Non-linear systems chaos
• Game of life (Gardner - 1970)
• Neural networks (1974)
• Distributed self-organised systems

25
2.1 Henri Poincaré (1854-1912)

• Is the solar system stable forever?
• The 3-Body Problem (1889)
• Others Lorentz (1917), Hénon (1931), May
  (1936), Feigenbaum (1945), Wisdom (1970)
• The N-Body Problem
• http//members.fortunecity.com/kokhuitan/nbody.ht
  ml
• Chaos and Henri Poincaré
• http//zebu.uoregon.edu/js/21st_century_science/
  readings/Parker_Chap3.html
• Try the 3 body problem
• http//astro.u-strasbg.fr/koppen/body/ThreeBodyH
  elp.html

26
2.2 Non-linear systems and chaos

• Non-linearity any system
• in which input is not
• proportional to output

27
How to represent interaction between populations?
(In Turbulent Mirror)
28
Phase space Limit cycles Attractors
(In Turbulent Mirror)
29
The way to chaos
30
The notion of bifurcation
31
From stable to strange attractor
32
Stability as a period of calm during chaos
33
Chaos

• Chaos theory attempts to explain the fact that
  complex and unpredictable results can and will
  occur in systems that are sensitive to their
  initial conditions. A small change in the initial
  conditions can drastically change the long-term
  behaviour of a system
• The dynamics of the system cannot be replicated
• Stability can be seen as a window of calm
  between periods of chaos
• Deterministic systems can have chaotic,
  unpredictable behaviour

34
Conclusion

• NL chaos theory has been an historical step
• It allows us to break the dominance of the
  analytical paradigm
• It forces us to analyze systems in terms of
  structural stability instead of input-output
  transfer functions

35
2.3 Neural networks (1974)

• They will bring the notions of
• Distributed processing
• Robustness
• Self organisation
• Automatic problem solving
• Associative memory (instead of content
  addressable memory)

36
Brief history of neural networks

• 1958 Rosenblatt perceptron
• 1961-1969 Minsky Pappert alliance against the
  Rosenblatt perceptron (XOR argument)
• 1974 Backpropagation network (P. Werbos Y
  LeCun) A supervised NN can do logical
  operations
• 1972-1982 Teuvo Kohonen network (self
  organisation without supervision)
• 1982 Hopfield network Spin glass An
  unsupervised network can self organize and
  compute optimal solution when faced to a new
  problem
• 1998 Weakly connected neural networks are
  equivalent to Hopfield networks and constitute
  oscillatory neuro computers with dynamic
  connectivity (computational embryogenesis, new
  perspectives in understanding evolution)

37
Neural network Backward propagation networks
38
Neural network Character recognition with a NN
The problem with the BP NN it needs to be
externally driven!
39
NETTalk Sejnowski (1988)
Conversion Graphème - Phonème Couche d entrée
29x5 unités, fenêtre de 7 caractères Sortie 55
unités Une couche intermédiaire 8à unités 309
neurones - 18629 connexions Apprentissage sur 100
mots 12 heures d apprentissage sur station de
travail RIDGE 95 de succès sur les
prototypes 90 de succès sur les mots nouveaux
40
Hopfield networks

• Each basin correspond to a memory
• Hopfield networks like human memories retrieve by
  similarity not by address like with the computer
  metaphor
• An Hopfield network is an associative memory
• Every image is distributed everywhere
• The system is robust

41
2.4 Game of Life (Gardner - 1970) Emergence
and Self-organisation

• Simple things interacting in simple ways can
  yield surprising forms.
• Game of Life rules At each step, life persists
  in any location where it is also present in 2 or
  3 of the 8 neighbouring locations, otherwise it
  disappears (from loneliness or overcrowding).
  Life is born in any empty location for which
  there is life in 3 of 8 neighbouring locations.
• http//www.bitstorm.org/gameoflife/

42
CA properties and analogies

• Like NN CA self organize
• Like NN, CA generate spatial order through
  strictly local interaction
• CA can compute because they can form the
  mathematical equivalent of very efficient
  Hopfield nets
• CA realize parallel computation
• During embryonic morphogenesis, neurons and glia
  cells acts an an excitable media (like CA) out of
  which self organize the higher levels

43
Distributed activity among cells
Distributed activity
Axons looking for target
Death of a cell
44
2.4 Distributed Self-organised systems

• A distributed system is made of a collection of
  entities where the decision is totally or
  partially taken by these entities (e.g. an ant
  colony).
• No global view.
• Intelligent global behaviour and functionality
  emerges from local interaction.
• Structural flexibility, fast reaction to external
  environment changes, robustness.
• Complex social reorganisation, evolving functions.

45
BOIDS
Separation steer to
avoid crowding local
flockmates
Alignment steer towards the average heading of
local flockmates
Cohesion steer to move toward the average
position of local flockmates
46
Social interactions and distributed systems how
to catch meaningful behaviors?

• Step 1 activity analysis

47
Step 2 Formalize regulation mechanisms

• Ex Overhearing

48
Step 3 Write an agent based model
49
Step 4 Assess the model on scenario basis
50
Step 6 Run the model in new organisations
51
Why socio cognitive systems are complex?

• Impossible to understand some functional aspects
  of social systems using a classical analytical
  approach.
• Because the context is always changing and cannot
  be analytically modelled
• Artefacts often play a cognitive role (they are
  not passive)
• Socio cognitive processes are sometimes emergent
  (crowd, politics, economy, etc.)
• They are open systems (no operational closure)
• A complex systems approach can allows us to fill
  in the gaps in our understanding of social
  systems.
• Cognition is distributed, situated and socially
  shared

52
4. Why use the Complex Systems Approach to study
Socio Technical Systems?

• Over-simplification of models leads to non
  applicable results in real situations
• Emergence (and learning) cannot be explained.
   Everything  is contained in the initial model.
• Initial conditions must be perfectly defined in
  order to be able to predict behaviour

53
Summary the main concepts brought by complexity
theories

• Emergence
• Functional robustness
• Self organisation

Emergence of socio cognitive processes
Feedback at individual level
Environment
Local rules of interaction
54
Complexity and Anthropology
The Long House Valley Project
Micro economic model Village formation f
(rules of exchange between households,
minimisation of the cost of subsistence and of
exchange)
55
The ANASAZI cultural model
This model explain how the Anasazi culture
spreads its population between AD 900 and 1300
Quantitive reconstruction of annual fluctuations
of production potential in the valley f
(rainfall, sunshine, nature of the various soil,
etc.
geomorphology and climatology.
Anthropology (New Guinea)
Model of social adaptation of the population
migration, increase in the population
Archeology geoarcheology, palaeoethnobotany
History of the population, of its agricultural
production and its phases of migration
56
Evolution of the number of sitesbetween AD 900
and 1300
57
Approaches in social sciences to escape the
drawbacks of analytical approaches

• Theory of internal external representations
  (Zhang Norman, 1994)
• Ethnomethodology conversational analysis
  (Garfinkel, Sacks, Schegloff and Jefferson, Heath
  Luff, 1994)
• Distributed situated cognition (Suchman, 1990
  Hutchins, 1995)
• Complexity theory
• Chaos theory (H. Poincaré)
• Distributed and self organised systems
• Cours daction (Pierce, Theureau, 1992)
• Autonomy, autopoiese (Varéla, 1989)
• Activity theory (Léontiev, Vygotsky, Kuutti,
  Engeström)
• ²