How should we study the embodied, environmentally embedded mind? The problem becomes acute once we realize that nature's solutions will often confound our guiding images and flout the neat demarcations (of body, brain, and world) that structure our thinking. The biological brain is, it seems, both constrained and empowered in important and sometimes non-intuitive ways. It is constrained by the nature of the evolutionary process—a process that must build new solutions and adaptive strategies on the basis of existing hardware and cognitive resources. And it is empowered, by the availability of a real-world arena that allows us to exploit other agents, to actively seek useful inputs, to transform our computational tasks, and to offload acquired knowledge into the world. A major debate in contemporary cognitive science thus concerns the correct choice of exploratory apparatus for modeling such multi-faceted processes. Recent work in robotics, developmental psychology and cognitive neuroscience has been taken (See e.g. Thelen & Smith (1994), Kelso (1995), Beer (1995)) to suggest the need for a dynamical systems perspective in addition to, or (more radically) in place of, the traditional apparatus of computational and representational explanation. In this paper I sketch some examples of embodied, environmentally embedded problem solving and argue for a hybrid explanatory apparatus incorporating both dynamical and more traditional (computational and representational) elements.
Behavior of cognitive game players who build the opponent's internal model is studied. Internal models are constructed by the recurrent neural network, and the iterated prisoner's dilemma game is performed. The complicated transients of actions are observed before the stable mutually defecting equilibrium is reached. These chaotic dynamics reflect the dynamical and high-dimensional rugged landscape of the internal model space. A possible world analysis reveals the other deep problem in the game thoery, i.e. uncertainty of games. Differences in a payoff matrix show that different matrix gives different possible worlds in behind. Some possible worlds can sustain a mutually cooperative state with mutually believing that the other player is playing Tit for Tat.
In this study, an artificial market approach, which is a new agent-based approach to investigate foreign exchange markets, is proposed. Using this approach, three emergent phenomena of markets were explained. First, in order to investigate the learning patterns of actual dealers, we held an interview with a dealer. Second, based on the field data acquired, we constructed a multiagent model of a market using genetic algorithms. Finally, the emergent phenomena of markets were analyzed using the simulation results of the model. The results showed that the interaction between the agents' forecasts and the relationship of demand and supply caused the phase transition of forecast variety. The three emergent phenomena were explained by the phase transition. This approach, therefore, integrates the fieldwork and the multiagent model, and provide quantitative explanation of the micro-macro relation in markets.
The authors propose a dynamical model that represents how people searches Japanese Chinese characters (Kanji) in memory space. We conducted two experiments exploring processes in which forty-five university students searched eight target characters by figural cues. The output pattern, the duration and the protocol data indicated that the process of search were divided into three phases: (i) direct search for two target characters, (ii) indirect search for three with wrong characters, (iii) impasse and insight for three difficult characters. A dynamical model of memory searching process was constructed with the results. The output pattern and the duration in the direct search phase are dependent on three constraints-relaxation of default conditions (object, relation and goal). The impasse and insight phase were simulated by a chaotic neural network. The transient dynamics were leaded by evaluation function. The implementation of the model could simulate a typical searching process of the participants.
Dynamics of structure in relationships among words via communication is studied by a constructive approach. Our study is based on the dynamical view of language, which is that meanings of words are dynamically created through activities of sense-making by individual language users. We simulate “conversation” in which agents having word relation matrices as their internal structure speak and listen to sentences. The sense-making activity is modeled by situating words in a web of relationships among words according to the usage of the words. We observe that agents develop cluster structure in word relationships, which is endowed with both stability and adaptability. It is considered that the cluster structure shares some features with prototype category. The structure changes dynamically by appearing new usages of language. The coexisting commonality and individuality of cluster structure is observed in an ensemble of individuals.
According to a study of Situated Cognition, learning for individuals is not valid until they join into practice and acquire their own roles under the social environment. We call such a capability “sociality”, a capability of finding its own role or niche in the social environment through interactions with their restricted neighbors. Our main purpose in this paper is to clarify an emergent mechanism of such “sociality” from the viewpoint of a multiagent study. In this paper, we emphasize that the emergence of “sociality” seems to depend on the dual capabilities of an individual's referencing; self-referential and social-referential abilities. In addition, we present a learning model of an agent having such dual capabilities as a Bi-Referential Model, in which each referencing capability is implemented by an evolutionary computation method of classifier system. Finally we present simulation results obtained by the proposed Bi-Referential Model and discuss the relation between the emergent process of “sociality” and the changes of resources that are commonly available to the agents.
Recently, mindreading ability is the topic of intensive investigations, fired by the “Theory of Mind” research by Premack & Woodruff. Few researchers, however, tried to define “mindreading” algorithmically. We mindread other people everyday, and think we know what mindreading really is—e.g. to reproduce in ourselves the “thought” existing in other's mind. Unfortunately, this definition cannot be applied for machine systems. Even for human beings or animals we cannot examine the thought in other's mind directly, and there is no method to verify the coincidence. We propose a definition of “mindreading” independent of the inner representation (thought) of mindreaders and mindreads. Next, Multi-player Prisoner's Dilemma Game (MPD) is proposed as a task where mindreading is expected to be effective for the survival of the players. Computer simulation shows “(by our definition) mindreading programs” are actually acquired through evolution in the MPD society. Lastly the validity of our definition of mindreading, the implication of our definition, and condition for the emergence of mindreading are discussed.
A parallel distributed processing (PDP) model of resolution of ambiguous Japanese nouns and verbs shown with contexts is presented. As a result of phonological restriction, Japanese possesses a large number of homophones. In understanding Japanese, the problem of lexical ambiguity resolution is especially important. In implementing the PDP model, the pronunciation, spelling, syntactic information, and semantic information of ambiguous words are represented as a distributed pattern of activation levels over a large set of simple processing units in a fully interconnected network. After the network is trained using the delta rule, the recall of learned patterns is evaluated by presenting just part of the lexical and contextual information. The similarity of recalled and learned patterns is used to indicate lexical decision times or the frequency of recall in empirical results. Consistent with several experimental data, the PDP model can successfully simulate the effect of context-dependent and context-independent frequencies of ambiguous nouns and verbs in the time-course of activation, as well as accessing an alternative meaning of ambiguous nouns in associative recall.