Humans Actual Imagined Implied From Swarm Intelligence by

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

Before We Begin… l l l 2 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. Hridesh Rajan March 11, 2003

Roadmap l l l 3 Behavioral psychology to cognitive psychology. Simulating social influence. Culture. Hridesh Rajan March 11, 2003

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 4 Hridesh Rajan March 11, 2003

Behaviorist’s Doctrine Classical conditioning - Organisms passively react to events in the environment. - Organism pushed by the stimuli. [Pavlov] l 5 Hridesh Rajan Operant conditioning - Organisms act on the environment to obtain a reinforcement. - Organism pulled towards a stimulus. [ Watson, Hull and Skinner] l March 11, 2003

From Behavioral To Cognitive 6 Hridesh Rajan March 11, 2003

Cognitive Psychology Stimuli Human Response There is more to human behavior than just stimuli and responses. 7 Hridesh Rajan March 11, 2003

How Are Stimuli and Response Related? l l l 8 How to explain why small rewards had greater effect? How to explain variation in problem solving time? How to explain social learning? Hridesh Rajan March 11, 2003

Vicarious Reinforcement l l l 9 Learn a task without actually doing it. Key is to watch someone else do it. “Bobo doll” experiment. [Bandura 1962]. Hridesh Rajan March 11, 2003

Further on the Cognitive Way l l l 10 Gestalt process. Lewin’s field theory. Sociocognitive efforts in parallel to behavioral and cognitive psychologists. Hridesh Rajan March 11, 2003

Gestalt’s Prägnanz /Good Form l l l 11 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. Hridesh Rajan March 11, 2003

Lewin’s Field Theory l l l 12 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. Hridesh Rajan March 11, 2003

Lewin’s Field Theory l l l 13 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. Hridesh Rajan March 11, 2003

Social Influence l l l 14 [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. Hridesh Rajan March 11, 2003

Sociocognition l l 15 Thinking is a social activity. Coordinated cognitive activities evoke intersubjectivity(shared understanding). Hridesh Rajan March 11, 2003

Social Aspects to Memory l l l 16 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. Hridesh Rajan March 11, 2003

Introducing Simulations as Evaluation strategy 17 Hridesh Rajan March 11, 2003

Simulating Social Influence l Ø Ø Ø l 18 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 …]. Hridesh Rajan March 11, 2003

Prisoner’s Dilemma[Axel. Rod 84] Two player competition or co-operation. l Both co-operate -> high payoff. l Both compete -> low payoff. l 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]. l 19 Hridesh Rajan March 11, 2003

Solutions l l 20 TIT-FOR-TAT (Winner) TIT-FOR-TWO-TATS. LOOKAHEAD DOWNING (Simulates behavior seen in human subjects in the situation). Hridesh Rajan March 11, 2003

Hutchin’s Network l l l 21 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. Hridesh Rajan March 11, 2003

Hutchin’s Network Positive Link 22 Negative Link Hridesh Rajan March 11, 2003

Hutchin’s Network Results l l l 23 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. Hridesh Rajan March 11, 2003

Inferences l Moderate ignorance not only permits cognitive consistencies, but agreement among members of a group. “. . We are always in a negative state of knowledge, ignorance. ” 24 Hridesh Rajan March 11, 2003

Culture 25 Hridesh Rajan March 11, 2003

Coordination Games l l 26 Prisoner’s dilemma revisited. Game of chicken Battle of the sexes El Farol problem Hridesh Rajan March 11, 2003

Game of Chicken a 1 b 1 Choices 27 • Swerve • Stay on the Road b 2 Hridesh Rajan a 2 12 12 18 6 6 18 0 0 March 11, 2003

El Farol Problem [Arthur 1994] l l l 28 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. Hridesh Rajan March 11, 2003

El Farol Problem l l l 29 Each of the agent can form k hypotheses of the form [f(present 1, …, presentn) => present 0]. 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. Hridesh Rajan March 11, 2003

El Farol Problem From: http: //www. santafe. edu/arthur/Papers/El_Farol. html 30 Hridesh Rajan March 11, 2003

Sugarscape[Epstein & Axtell] l 31 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? Hridesh Rajan March 11, 2003

An Epidemiological Model in Sugarscape l l l 32 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. Hridesh Rajan March 11, 2003

An Epidemiological Model in Sugarscape l l 33 Agents are allowed to propagate genetically, evolving immune system. Fitness function: how well does the immune system protects agents from diseases in the population. Hridesh Rajan March 11, 2003

An Epidemiological Model in Sugarscape l l 34 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. Hridesh Rajan March 11, 2003

ACE [Tesfatsion] Key observation: “ In the real world people choose whom to talk with, whom to interact with, whom to do business with, . . ” l Objective of the project is to understand how coordination arises in decentralized systems. l 35 Hridesh Rajan March 11, 2003

Competing-Norms Model [Picker 97] l l 36 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] Hridesh Rajan March 11, 2003

Mutable Prisoner’s Dilemma l l l 37 An individual agent is represented as a cell in CA and it plays repeated games with 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. Hridesh Rajan March 11, 2003

Results l l 38 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. Hridesh Rajan March 11, 2003

Inference By just communicating in the local neighborhood it is possible to optimize even a very complex decision function. 39 Hridesh Rajan March 11, 2003

Total Strength Ringelmann Effect [Ringlemann 1913] Number of People 40 Hridesh Rajan March 11, 2003

Dynamic Social Impact Theory [Latane 1981] l l l 41 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. Hridesh Rajan March 11, 2003

Dynamic Social Impact Theory l l 42 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. Hridesh Rajan March 11, 2003

Evolutionary Culture Model [Boyd and Richerson 1985] l l 43 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. Hridesh Rajan March 11, 2003

Evolutionary Culture Model l l 44 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. Hridesh Rajan March 11, 2003

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. l 45 Hridesh Rajan March 11, 2003

Boyd and Richerson’s Observation l l l 46 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. Hridesh Rajan March 11, 2003

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. 47 Hridesh Rajan March 11, 2003