Complexity and Simulation in Social Sciences

1
Complexity and Simulation in Social Sciences
Helder Coelho
Universidade de Lisboa
2
(No Transcript)
3
Quotations

• ?Probing the boundaries -- what complexity can
  and cannot be successfully applied to -- is one
  of the biggest intellectual tasks the scientific
  endeavor has faced, and we are still in the
  middle of it?.
• David E. Goldberg
• ?The unique thought that lives is the one that
  keeps to the temperature of its own destruction?.
  
• Edgar Morin
• Serendipity - ?the art of finding what we are not
  looking for by looking for what we are not
  finding?.
• Qu?au, 1986

4
On the Edge of Chaos

• Complexity, chaos and the origin of order are
  themes that come together, as well as dynamical
  systems, when we solve very difficult problems.
• Emergence of new forms and theories,
  non-linearity, evolution, autonomy,
  self-organization and adaptation to diversity are
  other themes that come afterwards.
• A key question is always Are there any possible
  mathematical descriptions for these topics? How
  can we understand them?

5
Contents

• 1. Introduction.
• 2. Would-be-worlds needs.
• 3. Examples.
• 4. Conclusions.

6

• 1. Introduction

7
Definition

• I will be brief. Not nearly
  as brief as Salvador Dali,
• who gave the worlds shortest speech.
• He said, ?I will be so brief I have already
  finished,?
• and he sat down.
• E. O. Wilson
• Complexity is the study of those systems made of
  myriad simple parts, baffling whole.

8
Focus

• Face big questions and gain insight.
• Disentangle a complex web of interrelationships.
• Collective dynamics of complex and adaptative
  systems consequences and preferred choice of
  actions are influenced by the activity of the
  others.
• Balance models (based on agents) with virtual
  scenarios (the theatre idea).
• Multidisciplinary ground understanding behaviour
  processes of economists, game theorists,
  political scientists and sociologists.

9
Methodology

• From REALITY
• (unknown processes between causes and effects)
• to FORMAL MODELS
• (interaction of agents between a specified
  behaviour and the emegent result of simulation)
• and to INTUITION
• (intuitions about behaviour generates forecasts)

10
Idea of Complexity

• When it is difficult to formulate its overall
  behaviour, even when
• Given almost complete information about its
  atomic components and their inter-relations.
• Different from size, ignorance, variety, minimum
  description length, and order.
• Different from computational complexity.
• Paradigm against reducionist approach.
• Need for measures (of processes).

11
Complexity/Simplicity

• Number and variety of components (entities,
  processes, agents), and intrincacy of interfaces
  between them.
• Number of conditional branches.
• Degree of nesting and type of forms (structures,
  functions and organizations).
• Dynamical properties of the (nonlinearly)
  interactions.
• System agents capable of generating emergent
  behavioural patterns, of deciding upon rules, and
  of supporting upon local data.

12
Complexity Sciences

• Need of epistemological shift from reducionism
  holistic approach.
• Superiority of abduction deductive and inductive
  reasonings are not able to jump to a higher
  level, far from observed facts.
• Simple and local components/interactions versus
  complicate and global behaviours.
• Main Features form, heterogeneity, variety,
  change.

13

• 2. Would-be-Worlds

14
Small Worlds vs. Games

• The question of agent rationality based upon
  search (utility, determinism, closed) or choice
  (values, non-determinism, open)?
• One, two or more dimensions (multidimensional
  world) to handle preference?
• Unknown evolution versus expected problem
  resolution.

15
Social Sciences vs. Economics

• Social Sciences Economics
• Society Economic interaction
• World of (social) actions Game/puzzle
• Interdependence Interaction
• Dependence, Value Utility
• Interference, Influence, Exchange Strategy
• Action Move
• Dependence theory Game theory

16
Theory

• Dependence theory the idea of complementary and
  interdependence.
• Game theory the idea of utility. Situations
  where instead of agents making decisions as
  reactions to exogenous prices, their decisions
  are strategic reactions to other agent actions.
• An agent is faced with a set of moves he can play
  and will form a strategy, a best response to his
  environment, which he play by.
• Strategies pure (play a particular move), mixed
  (random play).

17
Sociability Models

• Top-down organizational structure.
• Bottom-up based upon utility based upon
  complementary.
• Weak Points of Models based upon structures and
  utility
• Lack of a dynamic point of view.
• Social actions are only strategic.
• Need of social desires and intentions (mental
  models).

18
Ontology

• Agent active entity, able to make moves and for
  a strategy.
• Society a collection of agents (agency).
• Organization (group, coalition) constraints
  (norms), applied upon the agents within some
  society, able to guarantee that each agent will
  want and will do what must be done in some
  instant of time.
• Interaction exchanges (influence, cooperation)
  of information among agents (perception,
  communication, action).

19
Social and Political Sciences

• Q How can an agent adapt to a changing and open
  environment?
• A With a social reasoning power upon the other
  ones and the received data, and able to
• Infer dependence relations by taking into account
  its plans and the other plans (qualitative and
  quantitative issues),
• Build dependence networks, and
• Identify dependence situations, relating the
  other agent goals.

20
Decision Taking

• Simple decision choice process of an option or
  action course among a large range of alternatives
  (preference criteria).
• Four steps Information gathering, likeliwood
  evaluation, deliberation, and choice.
• Complex decision uncertain environments, several
  decisions in sequence, large length of the action
  sequence, need of optimal policies, state methods
  replaced by heuristics, state decomposition into
  state variables, and Markov techniques.

21
Decision Models (Rationality)

• Optimization model optimal rationality by
  maximizing the utility function.
• Satisficing model sub-optimal rationality by
  served by the expected utility.
• BDI model means-end analysis for ajusting
  actions /decisions to desires/goal structures
  and in according to beliefs. Two cycles 1)
  deliberation (decide what to do) and 2) action
  selection (decide how to do). Weak choice
  process!
• BVG model agent motivations and preferences
  guided by values.

22
BVG Model

• Well adjusted to complex, dynamic and
  unforseeable environments.
• Support a better autonomy for agents with less
  resources.
• Able to tune filtering processes by importance
  calculus.

23
Utility under Attack

• For complex domains it is difficult for an agent
  to have beliefs for everything.
• Agents are not perfect and their choices are not
  consistent and transitive.
• Generally, agents make options according to
  pleasure, hapiness, satisfaction, and not always
  upon profit (maximum utility).

24
One vs. Multidimensions

• Q What is a rational decision-making agent?
• The one dimensional side
• A1 Some one that maximizes the expected utility
  (cost function) the world is simple!
• The multidimensional side
• A2 Some one that has a limited rationality and
  makes decisions with values (Simons aspiration
  levels of utility, or qualitative goals).

25
Game Theory

• A Nash equilibrium will be reached when each
  agents actions begets a reaction by all the
  other agents which, in turn, begets the same
  initial action. The best responses of all players
  are in accordance with each other.
• Games non-cooperative (or strategic, static and
  dynamic) and cooperative (or coalition).
• Games two person constant sum (minimax theorem),
  two person (non zero sum), n persons.

26
Types of Problems

• Those that require collections of intelligent,
  adaptative and heterogeneous agents.
• Intelligent they use rules to choose and decide
  the adequate actions at any time.
• Adaptative they are ready to change their rules
  if the old rules are not working so well anymore.
• Heterogeneous they dont use all the same rules.

27
Oligopolistic Market

• Oligopoly complex and hard to model!
• Consider a particular case duopoly (Strategic
  Game) (Caldas and Coelho, 1994).
• Duopolistic market 2 producers and n consumers.
• Competition versus monopoly.
• Sequence of price decisions price competition
  with strategies (cut-throat competition,
  co-operative/leader, co-operative/follower), and
  moves (keep the price, raise the price, lower the
  price, and retaliate by lowering the price).

28
Kinds of Complexity

• Environmental complexity.
• Complexity of agents models.
• Behavioural complexity.
• Capture behaviour and processes!

29
Trade-offs

• More competing criteria for model evaluation.
• Trade-offs involving this criteria.
• Complexity does not correspond to a lack of
  simplicity.
• When modelling is done by agents with resource
  limitations, the acceptable trade-offs between
  complexity, error and specificity can determine
  the efective relations between these.

30
Form-Meaning

• Form configuration (architecture).
• Meaning mapping of the input and resulting
  predictions in the model.
• Space of all relevant possibilities
• Models.
• A priori knowledge.
• Observations.
• Goals.
• Actions.

31
Modelling

• Traditional mathematics do not help, because they
  are based upon physics where agents adopt the
  same rules.
• Complex and adaptive systems are different large
  spectrum of rules and changing rules require new
  mathematical structures.

32
Method

• Adopt simulation to study the problems.
• Face emergent phenomena knowledge and
  understanding are not able to give forecast data.
• Do experiments at large.
• Think about the best ways to think.
• How can we develop a mathematics that accounts
  for radically novel evolutionary adaptative
  events?

33
Because

• Complex behaviour comes from the operation of
  simple underlying rules.
• But, sometimes deducing behaviour from rules is
  not possible, there is no practical way to study
  the network of causality in detail.
• Therefore, we need clever ways to short-circuit
  the gaps, to establish general principles that
  make many of the detailed deductions obsolete.
  Even alowing short cuts, the spectre of
  computational intractability still haunts the
  darkness between rules and consequences.

34
Insight

• ?Simulation is a third way of doing science,
• in contrast to both induction and deduction.
• Like deduction, it starts with a set of explicit
  assumptions.
• But unlike deduction, it does not prove theorems.
• Instead, a simulation generates data that can be
  analized inductively.
• Unlike typical induction, the simulated data
  comes from a rigorously specified set of rules
  rather than direct measurement of the real world.
• While induction can be used to find patterns in
  data,
• and deduction can be used to find consequences of
  assumptions,
• simulation modeling can be used to aid
  intuition.?
• Robert Axelrod, 1997.

35
Agent as a Population of Competing Mental Models
Environment
Actions
Observations
36
Agent Mental Model

• Agents are specified by their mental states and
  their behaviour is governed by mental laws.
• Agent goals are fixed by their mental states
  (desires).
• Agents means to attain their goals are based upon
  their own mental state beliefs (knowledge plus
  strategies).
• Control to evaluate the agentsgoals fullfilment
  is done with the mental state expectations and
  several attributes of the mental state
  intentions.

37
Agent Mental Model

• The attribute satisfaction is a function able to
  measure the agent proximity to satisfy one goal
  (intention).
• The attribute uncertainty measures the degree of
  being sure to attain the goal (some intention
  according to a strategy).
• In case of conflicting intentions the agent
  waits a while or requests some one to help him
  (joint intention).
• Complexity locus interactions of mental states.

38
Validation

• We look to see if the models behaviour is at
  least qualitatively similar to the known
  behaviour of the real system.
• We create various scenarios and put real people
  who have spent their professional lives working
  with the system in front of the computer and ask
  them to antecipate what will happen when the
  simulation is turned on. They are unable to
  predict surprises, but they are able to trace out
  the causal connections generating the
  hard-to-predict behaviors.

39
Technological Map
Social aspects of agent systems
Social Science
Agent Based Computing
Agent based social simulation
Multi-agent based Simulation (MABS)
Computer Simulation
Social simulation
40
Social Simulation

• Statistical techniques for the description of
  social processes.
• Adequated to management and social sciences.
• Need for regression analysis and event forecast.
• Not applicable non stable distribution.

41
Multi-Agent Based Simulation

• Covers descriptive and formal processes.
• The process is validated by comparing its
  behaviour and the interaction of the agents with
  the behaviour of social entities.
• Applicable emergency of social phenomena.

42
Value of Simulation

• 1) What question(s) you want the simulation to be
  able to address
• This determines the level of resolution of the
  simulation, who the agents are, what kind of
  interactions take place, and so forth.
• 2) To make sure you have an expert advice
  available from people who understand the system
• Giving the agent realistic rules of action to
  choose among,
• Setting the way the agents decisions interact
  with each other to form the collective behaviours.

43
Toolbox Sampler

• Intelligent agents.
• Genetic algorithms.
• Neural networks.
• Cellular automata.
• Ant algorithms.
• Fuzzy systems.

44
Tools

• Swarm
• OAA
• Aglets
• Xdepint
• SimCog
• SEM (Agent editor, Agent code generator, Agent
  prober)

45

• 3. Examples

46
Exercises

• RoboCup Rescue (Trigo, 2002)
• Centipede Game (Antunes et al, 1999)
• A diversion of N-person repeated Prisoners
  Dilema (Caldas and Coelho, 1999)
• The Balance of Motives choice and institution
  (Caldas and Coelho, 1999)
• Power of Multitude (Coelho and Caldas, 2000)
• Wine Selection (Antunes et al, 2000)
• Multiple Partner Coalitions (David et al, 2000)
• Dynamics of innovation (Schilperord,2002)

47
RoboRescue

• Disaster rescue most serious social issues
  involving chaotic processes, very large numbers
  of heterogeneous agents in hostile environment,
  teamwork, uncertainty of information, real-time
  decision, logistic planning, and emergent
  collaboration.
• Aspects of disaster fire, housing and building
  damage, disruption of roads, electricity, water
  supply, gas, and other infrastructures, movement
  of refugees, status of victims, and hospital
  operations.

48
Needs

• Complex disaster information.
• Chaotic processes.
• Emergent cooperation in a changing hostile
  environment.
• Search-and-rescue strategies.
• Simulation involves 1000 to 10000 agents and
  events.
• Simulators situation, building and housing
  damage, ire, life-line damage, victim modelling,
  refugee behaviour, and data-collection and
  visualization.

49
Main Features

• Rescue Soccer Chess
• Number of Agents 100 or more 11 per team --
• Agents in the Team Heterogeneous Homogeneous --
• Logistics Major issue No No
• Long-Term Planning Major issue Less
  emphasized Involved
• Emergent Collaboration Major issue No No
• Hostility Environment Opponent players Opponent
• Real Time Second-Minutes Miliseconds No
• Information Access Very bad Reasonably
  good Perfect
• Representation Hybrid Nonsymbolic Symbolic
• Control Distributed, Distributed Central
• Semicentral

50
Centipede Game

• Imagine two players, A and B, are allowed to play
  a game. In the first move, A can choose to end
  the game with payoff zero for himself and zero
  for B. Alternatively, A can continue the game and
  let B play. B ends the game by either choosing
  payoff ?1 for A and 3 for himself, or 2 for both
  A and himself.
• The rational decision for A is to immediately end
  the game and receive zero, because the worst case
  alternative is to receive ?1 in the next move.
  Anyway, if A fails to choose this, the rational
  decision for B would clearly be to receive 3,
  leaving A with ?1. However, empirical experiments
  show that people tend to cooperate to reach the
  (2, 2) payoff.

51
N-person repeated Prisoners Dilema

• Imagine a collective action problem. There is a
  situation of anonymous (or system mediated)
  interaction with strong external effects, where
  a) the aggregated outcomes are determined by the
  actions of a large number of agents, b) the
  aggregated outcomes determine the
  agentsindividual rewards, c) the agents are
  unable to communicate, and d) the individual
  rewards may not be determined by the individual
  contribution of the collective payoff.
• This situation can be translated into an economic
  experiment.

52
N-person repeated Prisoners Dilema

• A set of individuals, kept in isolation from each
  other, must post a contribution (from 0 to a
  pre-defined maximum) in an envelope, announcing
  the amount contained in it. The posted
  contributions are collected, summed up by the
  experimenter and invested, giving rise to a
  collective payoff that must be apportioned among
  the individuals. The apportioned rule instituted
  (known to the agents) stipulates that the share
  of the collective payoff must be proportional to
  the announced contributions (not to the posted
  contributions). The posted contributions and the
  corresponding announced contributions are
  subsequently made public (but not attributed to
  individuals). Individual returns on investment
  are put by the experimenter into the
  corresponding envelopes and the envelopes are
  claimed by their owners.

53
The Balance of Motives

• A set of individuals, kept in isolation from each
  other, must post a contribution in an envelope,
  announcing the amount contained in it the posted
  contributions are collected, summed up by the
  experimenter and invested, giving rise to a
  collective payoff that must be apportioned among
  the individuals the rule instituted stipulates
  that the share of the collective payoff must be
  proportional to the announced contributions the
  posted and announced contributions are made
  public individual returns on investment are put
  by the experimenter into the envelopes which are
  claimed by their owners.

54
Power of Multitude

• A population with a fixed number of economic
  agents is engaged in the production of a good,
  the price of which is computed in each market
  period by a mechanism represented by a linear
  demand function that sets a single price for all
  transactions which is negatively dependent on the
  aggregate supply. The amount produced and
  supplied to the market in each period of time is
  equal to the productive capacity of each agent.
  This capability is initially randomly generated.
  In each market period it is updated in the
  following simple and reactive manner (capacity
  rule) if in the market period the payoff is
  positive, then increase the capacity
  proportionally to payoff, else decrease it
  proportionally.

55
Power of Multitude

• Time erodes the productive capacity so that in
  every market period a fixed percentage of the
  capacity is lost. The unit costs of production
  for each agent depend on the aggregate supply of
  the group they belong to. External effects are
  present so that there are economies up to a
  certain amount of group production and
  diseconomies beyond this threshold.
• The dependence relationship is the following
  each agent is dependent of one and only one
  agent. One agent may depend on himself. If agent
  A depends on agent B, in each market period A
  will pay a percentage of its payoff (if it is
  positive) as a tribute to B. This tribute is
  added to B?s payoff increasing its?s capacity,
  and only the net payoff (payoff minus tribute) of
  A will increase A?s capacity.

56
Power of Multitude

• If we represent the population and the dependence
  relationship on a graph where the nodes stand for
  agents and the arcs for the dependence
  relationship, we can define a group as a maximal
  set of nodes linked by a path in this graph. So,
  dependence has a cost (tribute) and a potential
  benefit (lower unit costs). The dependence
  relation generates the groups of some network
  (hierarchies, rings, single-agent groups, and a
  combination of a ring and a hierarchy).

57
Wine Selection

• Some consumer wants to purchase a bottle of wine,
  and so goes to a wine merchant. He has some
  knowledge about wine, but naturally does not know
  every wine that is available to him. He will
  evaluate candidate wines using a fixed set of
  five dimensions V1 is the knowledge and trust he
  has of the producer V2 is the quality of the
  region in the given vintage V3 is the price of
  the wine V4 is the ideal consumption time
  window and V5 is the quality of the specific
  wine.

58
Wine Selection

• Domains for these dimensions will be D1 D2
  0,,,,,, D3 N, D4 NN, and D5
  0..20. Now imagine our buyer has the goal of
  giving a formal dinner this evening, and wants to
  impress his guests with a nice bottle of wine.
  So, he specifies as minimum standards (4, 4,
  1000, (0,5), 16), meaning that the wine should
  come from a known and trustworthy producer, price
  is no matter, should be ready to drink, and a
  quality of at least 16 points. Notice that
  different things could be tried for instance,
  his guests might be impressed by a wine
  fulfilling this specification (1, 5, 75, (0,5),
  18), meaning that a very expensive wine from an
  unknown producer turns out to be very good. In
  any case, there could be lacking information at
  the time the choice must be performed.

59
Multiple Partner Coalitions

• A company has a set of projects (goals) and
  different alternatives configurations (plans) of
  packages (actions) to build its software
  products. Each project holds a certain importance
  and a company may be willing to set up a
  strategically agreement with the others, instead
  of building service packages from scratch. Each
  project is associated with a given importance and
  each package with its cost. Build the dependence
  network of that company.

60
Dynamics of Innovation

• Innovation (new ideas, technologies, products,
  theories?) emerges from the complex interactive
  learning process, involving the recombination of
  different types of knowledge (codified, tacit?),
  information processing, and experiences.
• Territorial environment is a key factor
• (a) physical proximity facilitates the
  interaction
• (b) cultural identity facilitates communication
• (c) proximity facilitates spread of tacit
  knowledge
• (d) sustained local interaction supports trust
  and cooperation, and reduces uncertainty.

61
Dynamics of Innovation

• Articulation between the local environment and
  the global context the renewal of the local
  knowledge base is made possible by the
  interaction of the local nodes within the
  comprehensive global network. The topology of
  the global network and the place of each node
  within it, is crucial for the innovative
  opportunities of each node.
• Two spaces have to be taken into consideration
  the geographical space (proximity) and the
  topological space (form of the global network).

62

• 4. Conclusions

63

• Distributed modelling and simulation.
• Socioinformatics.
• Design of institutions and organizations.
• Transition from static to dynamic descriptions.
• Local links (interactions) versus global
  (phenomena).
• Qualitative what-if questions.
• Serendipity.