
An Emergence of Coordinated Vladimir Kvasnicka Jiri Pospichal Communication in Populations Department of Mathematics of Agents Slovak Technical University 812 37 Bratislava Slovakia [email protected] [email protected] Abstract The purpose of this article is to demonstrate that coordinated communication spontaneously emerges in a Keywords population composed of agents that are capable of specific genetic algorithms, coordinated communication, emergence, agent, cognitive activities. Internal states of agents are characterized Darwinian evolution, Baldwin by meaning vectors. Simple neural networks composed of effect, Dawkins’ memes one layer of hidden neurons perform cognitive activities of agents. An elementary communication act consists of the following: (a) two agents are selected, where one of them is declared the speaker and the other the listener; (b) the speaker codes a selected meaning vector onto a sequence of symbols and sends it to the listener as a message; and finally, (c) the listener decodes this message into a meaning vector and adapts his or her neural network such that the differences between speaker and listener meaning vectors are decreased. A Darwinian evolution enlarged by ideas from the Baldwin effect and Dawkins’ memes is simulated by a simple version of an evolutionary algorithm without crossover. The agent fitness is determined by success of the mutual pairwise communications. It is demonstrated that agents in the course of evolution gradually do a better job of decoding received messages (they are closer to meaning vectors of speakers) and all agents gradually start to use the same vocabulary for the common communication. Moreover, if agent meaning vectors contain regularities, then these regularities are manifested also in messages created by agent speakers, that is, similar parts of meaning vectors are coded by similar symbol substrings. This observation is considered a manifestation of the emergence of a grammar system in the common coordinated communication. 1 Introduction Human language [9, 10, 30] makes it possible to express a huge number of quite different meanings by token sequences composed of a small number of simple elements, and to interpret such sequences by the meanings that they contain. A standard meaning of the term “grammar” refers to the systematic regularities between meanings and their representation by token sequences in a language. These structural regularities of a language constitute a basis for expressions of novel meaning combinations. The hearer can accurately interpret the received sequences as involving those familiar structures and relations, even though their specific combinations may have never been used before. This means that a communication system endowed with a grammar can be used to express new unusual meanings related to specific situations. The ability to c 2000 Massachusetts Institute of Technology Artificial Life 5: 319–342 (1999) Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/106454699568809 by guest on 25 September 2021 V. Kvasnicka and J. Pospichal An Emergence of Coordinated Communication communicate by a system composed of structural regularities represents an important achievement of a species, for which coordinated social activities are vital to survive, and where an accurate communication between two individuals from the same species represents a definite selection advantage. When this benefit is recognized, it is natural to explain language communication as the result of Darwinian natural selection [21, 22, 28]. The creation of coordinated communication between agents can be interpreted as an emergence of a new phenomenon in a population of agents that are capable of specific pairwise communication accompanied by a learning process. Starting from the general ideas of Darwinian evolutionary theory, a source of emergence of coordinated communication in a population of agents may be found in the ability of coordinated communication to increase agent fitness; that is, there exists a selection pressure for a spontaneous emergence of coordinated communication. Individuals that are incapable of coordinated communication are strongly handicapped. Their evaluated fitness is substantially lower than that of those agents that are capable of coordinated commu- nication. Frequency of appearance of individuals that are not capable of coordinated communication decreases in the course of evolution. The purpose of this article is to study a hypothesis that coordinated communication together with grammar regularities are the result of an evolutionary process running in a population composed of individuals, or agents. These agents are endowed with an ability to perform specific cognitive activities. By the term coordinated communication we mean an exchange of messages between agents that is unified (with the same semantic contents) for the whole population. For elementary communication acts the term “coordinated” implies that both speaker and listener understand each message in the same way. This requirement is formally expressed by a sequence of elementary steps in the communication act: The speaker’s internal state (e.g., corresponding to an internal representation of the environment) is coded into a signal-message (token sequence) received by the listener, who decodes this message into a form of internal state: A selection of speaker’s internal state, ⇒ transformation of this internal state into a message represented by a token sequence, ⇒ the speaker sends the message to the listener, and ⇒ the received message is decoded by the listener into a form of internal state. From this simple communication scheme it immediately follows that agents should be capable of specific cognitive activities that consist of coding the internal states onto mes- sages represented by token sequences (direct cognitive activity) and also in decoding the received messages onto internal states (an inverse cognitive activity). Moreover, the listener compares internal states constructed from the received messages with original internal states of speakers. If these internal states, original and decoded, are different, then the listener modifies parameters of his or her cognitive device so that differences between speaker and listener internal states in the forthcoming interaction between agents decrease. We presume that the speaker’s internal states correspond to some external surrounding reality, which can also be determined by a listener. Therefore we postulate that a listener can find out in some other way the internal state of the speaker. The problem of emergence of coordinated communication, where cognitive activ- ities of agents are performed in many different ways, is very intensively studied in the current artificial life literature [3, 5–7, 16, 26, 29, 31–33, 35] as well as in that of evolutionary linguistics [21–24, 27]. The present article is based on Batali’s idea [5, 6] that neural networks represent agent cognitive devices. We substantially enlarged this 320 Artificial Life Volume 5, Number 4 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/106454699568809 by guest on 25 September 2021 V. Kvasnicka and J. Pospichal An Emergence of Coordinated Communication interesting idea so that the Baldwin effect [4, 8, 19, 25, 34] and Dawkins’ memes [11] are considered as well. Batali studied the emergence of language only in the span of one epoch of evolution, whereas in the present article the emergence spans many epochs, allowing reproduction of the best agents, inheritance and spread of memes throughout generations. Recently, a similar approach was studied by Steels [33], who explored the emergence of a common vocabulary enhanced by cultural transmission. However, his model included neither neural networks nor learning with the Baldwin effect. Even though his “cultural heritage” was transmitted by a similar process as the communication act itself, it may not be the best model in biological reality, where the meaning of a new word is understood in childhood mainly through situation evaluation enhanced by feelings. 2 Darwinian Evolution, Baldwin Effect, and Dawkins’ Memes Darwinian evolution is based on the principle of natural selection, according to which only the best-adapted individuals from a population survive. By reproduction of strong individuals offspring are produced that will be with high probability well adjusted for survival. In 1896 Baldwin [4] proposed a hypothesis: If learning helps survival, then the organisms best able to learn will have the most offspring, thus increasing the frequency of genes responsible for learning. If the environment remains relatively fixed, so that the most profitable information to learn remains constant, selection can lead to a genetic encoding of a trait that originally had to be learned. Baldwin called this mechanism “organic selection,” but it was later dubbed the “Baldwin effect.” As an extension of the Baldwin effect we consider Dawkins’ theory of memes [11]. Memes are ideas (messages) broadcasted throughout the whole population and are composed of useful information on how to increase fitness of chromosomes. Dawkins assumed that memes might be important for an acceleration of evolution. Many social traits (e.g., altruism) are discussed in the framework of population genetics [14, 15] only with some strong restrictions and assumptions that are not acceptable to all evolutionary biologists. An application of Dawkins’ memes represents a plausible possibility to explain and interpret many features of social behavior of
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages24 Page
-
File Size-