Agent Societies and Artificial Intelligence

Very recent articles have addressed the problem of the interaction between populations of virtual robots. Vijayakumar and Davis ("Metacognition in a "society of mind"") investigate the concept of mind as a control system using the "Society of Mind" idea from Marvin Minsky. They develop Metacognition in a model based on the differentiation between metacognitive strategies. Metacontrol is a part of metacognition. The authors explore Metacognition mechanisms in developing optimal agents for the fungus world testbed. The fungus world environment allows the behaviours of a robot to be monitored, measured and compared. It is generated a swarm intelligence of how the group of agents work together to achieve a common goal. Vijayakumar and Davis implement an architecture with six layers including reflexive, reactive, deliberative, learning, metacontrol and metacognition layers. The authors design an experiment with different types of agents (random, reflexive, reactive...) All agents move in the environment, changing direction in case of obstacles. To compare the results of each agent, the following data were collected: life expectancy, fungus consumption, resource collection and metabolism. Total performance was denoted by the combination of resource collected and life expectancy. Experiments were conducted for each type of agent. In Simulation 1, deliberative agents collected 50% of resource and left 64% of energy. In Simulation 2, the results of deliberative agents were compared to Metacognition agents, collecting metacognitive agents 82% of resource and lefting 71% of energy. The result concludes that Metacognition agents are better than other cognition and deliberative agents. Thus, Metacognition is a very powerful tool for control and self-reflection and intelligent and optimal agents can be viewed as collective behaviours as a "Society of Mind". Really to develop a Metacognition concept in Artificial Intelligence seems necessary for building self configurable computational models and true intelligence.

MathPsych 2009 (Amsterdam)

The Annual Convention of the Society for Mathematical Psychology ("MathPsych 2009") was held in Amsterdam on August 1-4, 2009. We stress the highlights for the purpose of our blog, Social Cognition.

On August 2, Joe Johnson (Miami University) illustrated, via a simple mathematical model, the ability for predicting decision behavior based solely on perceptual data. Johnson uses an evidence accumulation model with a ratio choice rule to predict athletes´intuitive, initial choice in a realistic game situation. Markus Raab (German Sport University in Cologne) investigates the preference for intuitive decisions in contrast to deliberative ones. He applies a mathematical choice model based in team handball attack situations. Bayesian models also are applied and so Tom Lodewyckx (University of Leuven) and others design a Bayesian state-space model for affectivity. They use Markov chain Monte Carlo methods to estimate the model parameters. This framework is used to high resolution psychophysiological and behavioral data obtained during adolescent-parent interactions expressing dynamical emotions.

Vanpaemel (Umiversity of Leuven) and Michael Lee (University of California, Irvine) advocate the advantages of hierarchical Bayesian modeling in providing one way of specifying theoretically-based priors for competing models of category learning. In a similar line, the group of Rich Shiffrin (Indiana University) spoke about information integration in perceptual decision making. According to the authors, researchers studying judgment and decision making have shown that people employ sub-optimal strategies when integrating information fron multiple sources. But another group of researchers has had success using Bayesian optimal models to explain information integration in fields such as perception, memory and categorization. Shiffrin and colaborators design a decision making experiment to test the range of this difference.

Amy Perfors and Daniel Navarro (University of Adelaide) consider the situation in which a reasoner must induce the rule that explains an observed sequence of data, but the hypothesis space of possible rules is not explicitly enumerated or identified. They present mathematical optimality results showing that as long the hypotheses tend to be sparse (that is, tend to be true only for a small proportion of entities in the world), then confirmation bias is a near-optimal strategy. The authors propose, in a very interesting manner, to chose queries that one knows will lead to an affirmative response for at least some hypotheses (hypotheses being considered). This positive-test strategy is closely related to the confirmation bias.

The Meeting finished on August 4 hoping to achieve the same successful objectives for coming events.

Public Good Games and Social Cognition

Public good games provide an interesting testbed for the study of social dilemmas. Social dilemmas are inherent in decision making. Despite the diversity of dilemmas, all share the same structure: each individual benefits from behaving selfishly but a group gets greater rewards if its members cooperate.

In a typical public good game, players may divide their initial endowments into both a private account and a group account. The frequent strategy for a rational player is defection but most participants in public good experiments do cooperate to some extent. Factors that enhance cooperation are the communication among players or the payoff structure of the game. Nowadays it is thought that behavioral investments could be an important determinant of cooperation in social dilemmas. Participants´ endowments are distributed asymmetrically among a group to discover whether "rich" individuals will contribute more than poor individuals. The conclusion is that the differences in the level of contributions may be reduced if positions are assigned on the basis of merit rather than chance. Although asymmetries in participants´ behavioral investments (for instance, effort investments) affect fairness judgments and group identification, members are often not aware of such differences. But the origin of the endowments affects contribution levels in the public good game (Muehlbacher and Kirchler, 2009). Subjects who earned their endowments through a greater amount of effort were less cooperative than those individuals who had earned the money with ease.

Chaos Theory and Social Cognition

We will review the last contributions of the theory of nonlinear dynamical systems to the field of social cognition, specially

Vallacher and Nowak (2009).

Social relations evolve and change in the absence of external influences. Psychological systems display intrinsic dynamics. Three basic types of attractors have been identified: fixed-point attractors, periodic attractors and deterministic chaos. A fixed-point attractor describes the case in which the the state of the system converges to a stable value. It is similar to the notion of homeostasis. It corresponds, for example, to a desired goal. Multiple fixed-point attractors express that people can have different (perhaps contradictory) goals and patterns of social behavior.

Some systems display oscillatory behavior. A temporal pattern showing this tendency is a periodic attractor. Social judgement often oscillates between positive and negative assessments.

Chaos represents a possibility in many social phenomena. Social psychology presents many nonlinear phenomena, such as complex interactions among variables or inverted-U relations.

Social interdependence is very important to game theoretic approaches in Psychology, for instance, the prisoner´s dilemma game. So, Nowak et al. (1990) used cellular automata to model the dynamics of social influence. Each individual had three properties: an opinion on a topic, a degree of persuasive strength, ans a position in a social space. In each round of the computer simulations, one individual was chosen and influence was computed for each opinion in the group. The updating rule was the following: the individual changed the opinion to match the prevailing opinion if the resultant strength for this opinion position was greater than the strength of the individual´s current position. The minority opinion survived by forming clusters of like-minded people.

The dynamical account of social influence describes how the state of a single individual depends on the state of other individuals. However, individuals are best conceptualized as displaying patterns of change rather as a set of states. Social influence can be approached as the coordination over time of individual dynamics. Individuals in a relationship are represented as separate systems capable of displaying rich dynamics. A recent model of synchronization has been developed by Nowak, Vallacher and Zochowski (2002). Coupled logistic maps are used in this model. The behavior of each individual not only depends on the preceding state but also on the preceding state of the other person.

Dynamical social psychology has generated a deep source of formalisms but social reality can not be confused with physical reality. Individuals are not interchangeable in the way that atoms are. People live in a symbolic world and do not respond in a reactive way to the objective features if the environment. For it, human dynamics contain some degree of randomness, and human behavior is often unpredictable and perhaps chaotic.

Clint Sprott (Fractal image)

Peter Gollwitzer and social cognition

I would like to pay homage to Peter Gollwitzer, presenting some of his outstanding contributions to social cognition.

Peter Gollwitzer represents the great German tradition in Psychology: from Wilhelm Wundt to Kurt Lewin, from Asch to Heinz Heckhausen, Germany always has brought excellent researchers to the study of the volitional processes. Gollwitzer is renamed expecially by his contributions to intentional behavior studying implementation intentions. Intention implementations are currently at the center of research in the world of the Psychology. By means of implementation intentions, bringing a goal pursuit to a successful end is facilitated. They are effectives in different areas such as health domain or academic performance. So, the likelihood of performing a breast revision was enhanced by forming implementation intentions. Besides, the effects of implementation intentions on reducing dietary fat intake were relevant. Bad habits are controled by implementation intentions and Schweiger Gallo has verified the effectiveness of forming implementation intentions on the suppression of undesired emotional responses (for instance, fear of spiders). But perhaps more important is the study of implementation intentions effects on critical populations, such as addicts in withdrawal, schizophrenic patients or subjects with frontal brain lesions. Another clinical population who profits from implementation intentions is ADHD children: in a study, the ADHD-children who formed an implementation intention showed the same response inhibition performance as children without any psychological disorders.

But perhaps the two pending tasks is to study the design of Artificial Intelligent systems incorporating implementation intentions and the underlying mechanisms of implementation due to neurophysiological research. The first one has became to be analysed by the author of this blog in his article about implementation intentions and artificial agents. The second one is being studied by means of the electrocortical correlates of emotion regulation by ignore-implementation intentions. It seems that these intentions produce their effects through cortical control that sets in the information processing system blocking the emergence of negative emotions. Electrocortical correlates offer the possibility of determining at what point different kinds of implementation intentions (suppression or inhibition) exert their effects after the critical stimuli are encountered.

Forming implementation intentions reveals to be an effective self-regulatory instrument in domains including social psychology and clinical psychology. Gollwitzer continues the tradition.

Joint Action and Artificial Intelligence

In this article, we examine how the research about joint action (Sebanz, Knoblich, Bekkering...), can be intertwisted to the ideas about distributed coordination in uncertain multiagent systems (Maheswaran, Szekely, Rogers, Sánchez...)."Joint action can be regarded as any form of social interaction whereby two or more individuals coordinate their actions in space and time to bring about a change in the environment" (Sebanz, Bekkering, Knoblich, 2006, p. 70). According to the authors, successful joint action depends on the abilities to share representations, to predict actions and to integrate predicted effects of own and other´s actions. A mechanism for sharing representations of objects and events is to direct one´s attention to where an interaction partner is attending. However, a more direct mechanism is provided by action observation; studies about mirror neurons show that during observation of an action, a corresponding representation in the observer´s action system is activated. On the other hand, an efficient means to predict other´s actions that is not based on action observation is knowing what another´s task is. A series of recent studies has shown that individuals form shared representations of tasks quasiautomatically, even when it is more effective to ignore one another. But how individuals adjust their actions to those of another person in time and space can not be explained just by the assumption that representations are shared; action coordination is achieved by integrating the "what" and "when" of others´actions in one´s own action planning. This affects the perception of object affordances, and permits joint anticipatory action control. Very recently, Maheswaran, Rogers and Sanchez (2007) introduce the idea of distributed coordination in multiagent systems. I believe that the research by Sebanz and others in humans can be applied successfully to the area of distributed Artificial Intelligence. Centralized systems can generate fully coordinated policies but put a very high computational load on a single agent. Given a team of agents, every agent in the team has a set of activities that it can perform. Each activity has probabilistic outcomes. Only the agent has current knowledge of its policy at all times. The team reward is a function of the qualities of all activities, and the agents´ objective is to maximize this reward at some terminal time. One way this function can be composed is with a tree where the activities are leaf nodes. Each non-leaf node is associated with an "ancestral" operator which takes the qualities of its children as input. The output of the root node is the team reward function. An agent´s subjective view of the reward function can be defined. It is considered the case where each agent see all ancestral nodes of activities they own, and any nodes and links that connect to its activities and their ancestral nodes via directional operators. The authors introduce distributed coordination between artificial agents, that is, something like the joint action in humans studied by Sebanz and others in humans. In an immediate future, a fertile cross will happen between these two roads.

Winter School 2009: Social Structures in Communication Networks

On 5-10 January 2009 and chaired by Jorge Louça, has been celebrated at the University of Lisboa, the Winter School 2009: "Social Structures in Communication Networks". On Monday, Professor Louça presented the program and scientific goals in the context of the International Doctoral Program in Complexity Sciences. The "Universidade" is accomplishing a great effort following the way pioneered by Santa Fe Institute or the Complex Science Society at Paris.

Professor Araújo analysed whether biological-inspired network models contribute to the understanding of aggregate economic behaviors. Besides, and following the ideas by John von Neumann, she explained the redundant nature rather than efficient nature of the network structures, contributing to clarify the idea of efficiency versus reliability in social environments. In the evening, Professors Louça and Rodrigues introduced tools for the study of social networks like NWB, PAJEK, GUESS or UCINET.

On Tuesday, Professor Symons introduced in his second presentation, ideas including references to the analysis of emergent properties. Symons offers an alternative to the macro-properties model of emergence. Emergence is not restricted to macro-phenomena, but can appear at the intersection of networks. Objects or agents can be involved in distinguishable systems or networks simultaneously. An agent may participate in social, economic, political and other networks at the same time.

On Wednesday, Professors Louça and Rodrigues showed multi-agent based social simulation by means of tools like NETLOGO, MASON or BREVE, while on Thursday, both Professors outlined community detection in social networks, employing algorithms like Girvan-Newmann. In the evening, Professor Palla exposed the statistical properties of community evolution in complex networks. On Friday, Professor Marinheiro spoke about network architectures and organization. Network maps have been used to mitigate the problem of the understanding of the network structures. Professor Lopes explained that in the networked multimedia realm, "more is different".

Finally, on Saturday, 10th, it was programmed a keynote talk by James Sterbenz (University of Kansas), concerning adaptive computer network architectures.

I am very grateful to the University of Lisboa by the organization of this School in which young students and prestigious Professors, have shared very interesting viewpoints about the formal and social analysis of networks, mixing complex systems and dynamics of networks.

"The brain is a network of neurons; organizations are people networks; the global economy is a network of national economies, which are networks of markets...How do such networks matter?"

The link to the Winter School is:

http://idpcc.dcti.iscte.pt/ws2009/ws2009_home.html