Artificial Agent And Learning Models

Artificial Agent and Learning Models Emerging Systems in Learning Models Vito Francesco De Giuseppe*

Abstract The being Systems are characterized by the capability to interact with the environment that surrounds them and from this interaction take form the structure allowing to biologic systems to adapt to the environment. In this work are shown the first results of the testing activity on disclosure of emerging phenomenons in the establishment Cognitive Field of an artificial Agent, based on Kurt Lewin 's Field Theory and on Systems Theory, through the use of biological evolution simulation software. The first results seem to suggest how even during the simulation the emerging phenomenon can be comparable to that which is stressed in human behaviors during the first developmental levels. Keywords:Genetic Algorithm, Artificial Life, Field Theory, Systems Theory, Emerging Systems.

Knowledge Setting: The PIK Type Agents can be accounted as components that build learning. We can hit on Agents who set knowledge in accordance with bottom-up modality and on Agents who set it according to top-down ones. Observing the TOGA type (Gadomsky and others, 2003), has been considered a type in which a catalyst component which we shall define Preference, by which we suggest the internal condition of the learning agent. To be more precise, on the one hand the tension to the information, on the other hand a skill, that we'll indicate with the letter P, correlate the Information, that we'll indicate with the letter I, to the Knowledge, indicated with the letter K, in which will be implemented the Information I. The Information establish a new level of Knowledge K. The agent in this way outlined will be defined PIK and set knowledge according to the bottom-up modality. I can be connected only to with K. P can be connected with I and K. The vector that connect I with K moves only in this direction and not in reverse. P connect with I and with K with a one-way vector that moves from K to these two components and cannot move in the opposite direction. P is the only component that can connect with both. I cannot connect with any other I, as well K cannot be connect with any other K. An I can be connect with several K. If we try to account what we have said with a diagram, we obtain a topological defined figure (Fig.1).

1

K

P

I

P K

K

I

I

P K

P

I I

P

I

I P

K

K

P I Fig. 1

If we reduce the diagram to its basic elements, that is P, I and K, we obtain a base two-dimensional figure: a triangle. The vectors that connect the components create an organization taking a well-defined shape in the space. To unite the three factors is necessary that the vectors shape angles of different largeness, depending on the position in planar space (Fig.2). 2

K P

I Fig. 2

I

P

K I

P Fig. 3 If we increase the number of components, the number of triangles increase, but not in proportion to the components, since an only element P can't be connected with several components I and K. The shape will be more complex owing to the basic shape repetition, though the rules that organize the structure remain unchanged.

3

I

P

I P K

I

K

P I Fig 4

It can be shared among several knowledges. Even in everyday experience we see that the same information can be implemented in specific knowledges, also different, but sharing the same information (Fig.4). For example the informations concerning a human cell working belong to a Biologist's and a Medical man's knowledge. The Biologist and the Medical man are two different agents with different knowledges, but they share the same information that gives birth to two different knowledges from which spring different abilities and skills. P can share several I and K, too. Indeed, both the Biologist and the Medical man can have choices compared to the human cell working. Each K will remain specific, since the Biologist's and the Medical man's knowledge will remain specific. As a consequence no K will connect with a K, that will remain ..... and specific, as well each information is unique and, at the same time, can't be many informations. To stay in this example, the information concerning a human cell working can't be, at the same time, that concerning a car working, and vice versa. From this reasoning we can formulate the notion that K (Knowledge) is function of P (Preference) and I (Information): K= f (P, I)

A genetic model of PIK type What we have said up to now is the effort to propose in formal terms a type of knowledge organization that describes the relation between the informations, skill and the condition of the learning agent. Now we try to determine the Knowledge evolution when the agent is inserted in an environment where informations are organized on the bottom-up modality to rebuild the features of an Agent's Cognitive Field, 4

in the meaning of emerging phenomenons of interaction of acquired informations to analyze and, when it is possible, to detect a type about how informations can pattern a Cognitive Field, on the base of theoretical paradigms of Lewin's Field Theory and Emerging Systems.

Field Theory In the Field Theory, Lewin premise that what he defines Field is a set different from the simple addition of components that compose it and describes that the Field is divided in Areas and it is delimited by a Border. Areas and Borders are mobile and, as the whole Field, tend to a balance “almost firm”, in the sense that they are to the limit of a change or of a passage from a state to another. In the Field don't exist empty spaces. Areas recognize Agent's mnestic traces and their size set the weight in the Field. Both the Field and the Areas evolve on several dimensions and consequently show many border areas (Lewin, 1980). If we consider the Field so defined by Lewin, we think of an Agent that modulates learning in adaptive terms, according to requests tending to restructure the same Agent's Field.

Systems Theory By the Systems Theory elements composing a structure, an organization, create a configuration with specific characteristics and problems, dissimilar from those of the same components. In the General Systems Theory the emergency is the factor through which systems establish.

Agent With the term Agent we indicate a system gift with sensors, through which know the environment. The processing of the perceived datum has as consequence the implementation of applied activities through actuators. This includes all those activities that have the purpose of acquiring further knowledge. This is the definition of intelligent or rational Agent. If we think about a man, we define his eyes and ears as sensors and his limbs as actuators. Among the many classifications that can be done about the Agents, that increase with the addition of new types implementations, we find those concerning (Gadmosky, Fontana, 2003) • • •

aim and objective; avail; learning.

Specific Agents and Unspecific Agents Specific Agents are entities that take up orientated informations, i. e. take up just a sort of informations, elaborate just specific features and implement resulting behaviors of specific sort, according to characterizing modalities. This specificity is produced by the typology of the sensors of which they are composed, that can detect just a 5

determined input. The specificity doesn't concern only the datum quality, but the measure of the datums in entrance. In case of a human being, for example, eyes acquire the optical datum in a well-defined range of the electromagnetic spectrum. The optical field is outlined by the eyes position and can be focused only elements put in a part of its. For example, if we focus the objects on the background, the close-up objects put out of focus and vice versa. The elaboration capability is made by specific elements and unspecific ones. The firsts elaborate the stimulus and its physical features, the seconds those unspecific, that is common to the various inputs. First, It answers to the request of adaptation and to do it the Agent needs to acquire informations. The acquisition of informations build knowledge, meant as structure in which organize acquired informations. Abstract agents are (totipotenziali) , of simple structure, that can be assemble together with others and that are used in easy and continuous repetitions.

Artificial Life Learning: Experimentation In the endeavor to analyze Agents' activity with a bottom-up approach, an experimentation has been launched on the disclosure of emerging phenomenons in the formation of an artificial agent's Cognitive Field through the use of software of biological evolution. Informations and agents are considered equally, as elements composing an environment, the Cognitive Field, to value the possibility to detect a higher level of complexity than those of lower levels. The aim is not to create an algorithm that describes the behavior of one agent, rather using an instrument that underlines the results of all the agents' interactions in an environment (Annunziato, Liberto, Pannicelli, 2007). To build an evolutive type of bottom- up modality we used an artificial life simulator, precisely the software Avida, to simulate an Agent's evolution to see if in the knowledge setting is possible to find an emerging system. With the use of the software we tent to implement the PIK type in another type, that through a genetic algorithm make the system evolve.

Experimental hypothesis An analogy has been assumed between the amino acids that aggregate to form a DNA molecule and the informations that aggregate to form the Cognitive Field. Both the DNA and the Cognitive Field are constituted, develop and evolve on incentive of the adaptive requests gave by the environment. If amino acids interact to form a complex structure, we can assume that simple informations interact to pattern complex forms. If bonds among amino acids reproduce types of bonds among informations, the simulation of amino acids bonds will reflect that of the informations. Cognitive Field evolution can be simulated through a system that simulate biological evolution of a genotype. So, we thought to use an Artificial Life Simulator that, trough genetic algorithms, simulate Cognitive Field' s evolution of an Artificial Agent during the adaptation process to the environment in which it stays. The information I has been considered as a cellular organism. It responds to the closeness of its peers, coupling to these when the features enable their conjugation. Systems produced by cellulars conjugations become knowledge K. Ks could originate two daughter cells I. We thought to build an agent resting on development genotypical equations. The informations meet few interaction rules. Basically the relations we had indicated as vector connections in the PIK type, in this type they were replaced by colorations. Informations were linked through colors. Each color reflected a different typology of information. The connexion of several colored elements gave birth to a cell mother, that divided in two daughter cells. During the division a small part of the information could 6

undergo a light change. It's been also assumed a limit to the information life, to simulate the memory decay phenomenon. When an agent doesn't use an information for a long time that information decays till it disappears and it's not used anymore. In short, it dies. We considered an information as an organism that dies owing to the lack of food or cause of the impossibility to reply to the evolutionary requests. At the bottom if we consider the adaptation as a the application of strategies that depart from informations, we can see that if an information is no more suitable for the target, it is no more used and disappears from the cognitive repertory of the agent. Subsequently, it is replaced by another information. This retraces the evolutionary scheme of an organism, when its working it's not suitably equivalent to environmental stimuluses. In terms of adaptation it dies for the inability of its strategies, so an organism more operative takes its place. Using the same scheme, we thought to such a body acting in an environment, an agent who changes depending on the number of organisms and their whole dynamic.

Experimental design The experiment has provided a Cognitive Field simulation lasting thirty minutes. Was used a software simulation of biological evolution, with whom to observe how the informations interacting between them,once acquired, structure the vision field. During the observation the Cognitive Field was championship, at different times of growth and evolution. Was used a bottom-up approach to observe emerging phenomena.

Methodology The methodology has provided an elaboration of thirty minutes total using software of biological evolution simulation. Data were initially structured and were given the rules of interaction data. Were then created the rules of interaction data for the elaboration process. The software had process the data with batch-processing. Evolution sampling was performed in three stages through screening drawing: t0, upon; t1, after ten minutes later; t2, after twenty minutes later; t3 at the end of the process at the thirtieth minutes following the elaboration arrest.

Tools Was used a computer Notebook Fujitsu-Siemens Amilo A with mobile processor AMD Athlon Xp tm 1800 and a memory RAM of 217,2 MB. The Operating System used was Ubuntu 7.10 “Gusty Gibbon”-x86 version- and the simulation software used was Avida Evolution Simulator. The Avida Evolution Simulator is a self-organizing system. Used on digital platforms in research on Artificial Life, is a digital world that changes and evolves. It is used in researches and experiments on the dynamics and evolution for the study of biological theories that cannot be tested in real biological systems. In Avida organisms are composed of a “body” which contains a set of components. These components are: • The memory, that is a sequence of instructions that may be performed, copied, changed, etc. • A Pointer of Education (instruction pointer: IP) which indicates the next point in which education may be enforced. • Three records that can be used by organisms to store the informations that are handled. The content isn't default and is a 32-bit. • Two Stacks (stacks of data in which the extraction of the data follows a reverse order compared to that of release) that are used to store the data obtained by organisms during processing. Theoretically, their capacity has no limit, but for convenience you prefer not exceed more than 10. • A buffer of inputs and an output, used to receive information and return the results made by the bodies. • A Reading Headline (Read-Head), a Writing Headline (Write-Head), and a Flow Headline (Flow7

Head), used as specific positions in Memory of CPU.

Analysis of data t0: Initial Moment. t0 is when the screen captured at the star of the simulation program. The black space is the environment in which the staff is immersed. The red dot indicates that the information that the agent is in possession corresponds to its Cognitive Field. Information and Cognitive Field match. The agent has a unique information: I AM HERE. The indication side, the red square represents the information, white place of residence, which is however canceled and doesn't structure in Cognitive Field (Fig.5).

Fig. 5 t1: after 10 minutes of elaboration. In t1 the Screen captured after ten minutes of simulation processing. The white space that represents the Cognitive Field has become broader and informations are diverse and are present in different points of the Field, but all around outdoors areas. The agent has more informations and a structured Cognitive Camp. In side indications, the squares that corresponds to informations have increased in number and they diversify into different colors, namely differ in characteristics (Fig.6).

Fig.6 8

t2 : after 20 minutes of elaboration Screen captured after 20 minutes of simulation processing. The white space, the Cognitive Field, has grown relatively little. The field has a greater number of elements (white squares). It can be argued that is now more divided. The points of information aggregation have decreased and occur only on outdoor areas. This stretch is more prominent than at the time t1. The information is steadied, by type and number (Fig.7).

Fig.7 t3: after 30 minutes of elaboration. Complete. Screen captured after 30 minutes of simulation processing. The white space, the Cognitive Field, has continued to grow with constant times, but more slowly than from t0 to t1, managing however to cover almost the entire environment. The Field has an internal Natural increase. Informations are diversified, as well as points of aggregation that do not occur only on outdoor areas (Fig.8).

Fig.8

9

Growth percentage of the Cognitive Field The growth rate of the Cognitive Field doesn't grow significantly between t2 and t3. Number of information between t1 and t3 The number of informations in the field half the time between t1 and t3. The time of growth of the Cognitive Field between t2 and t3 the timing of growth of the Cognitive Field doesn't grow significantly between the time t2 and t3.

Results of experimentation The first experiment seems to confirm some points • An Agent acquires and process more informations in an initial moment; • This seems to meet the need of the Agent to increase its Cognitive Field and this is appropriate for greater control in relation to a wider surface environment. This seems to be a typical adaptation characteristic; • At first, the Middle of the Field stabilizes and informations have easier aggregation in remote areas, then the Field presents new aggregations also towards more central points, in conjunction with a diversification of informations; • The processing time is not altered after an initial acceleration and assumes a constant; • The informations decrease over time: the Agent saturates the need to acquire new informations; • The Field gradually becomes more divided; • Initially the number of aggregations of informations decrease towards a steady, and then resubmit an increase of aggregations and a change in aggregations positions.

Conclusions At a first reading of experimental data, can be drawn only a few conclusions. The emerging phenomenon in the simulation, could allow to formulate a hypothesis that seems traceable to two situations: -The Child that learns; -The Teenager. In the situation of the Child that learns, initially, the Cognitive Field of the Artificial Agent is structured in a manner and processes typical of the child who learns, through a heightened sensitivity to collect new informations that produces a massive aggregation of informations, creation of new Field Regions and diversification in the type of information. After an initial acceleration times stabilize. In the situation of the teenager, reached a threshold limit, when the whole environment becomes part of the Cognitive Field, occurs once a mode of behavior characterized by the need to collect more new informations to meet the new instances that the environment proposes, in line with the changes occurred within the system. At this stage there is an increase of aggregations of informations and a further capacity to structure more Field Regions and a further diversification in the type of informations. The second situation appears to be similar to that present in the adolescent period, where restructuring the Cognitive Field to remodulate on new adaptive instances. The constitution of the Cognitive Field model is presenting emerging phenomena not dissimilar from those detectable in a colony of ants, the second one that seems to be a swarm logic. Beyond the first results described, experimentation illustrated must be considered as the beginning of a draft study.

10

* Psy.D, Psychotherapist.

11

Bibliography M. Annunziato, C. Liberto, A. Pannicelli. Modellazione ad agenti dei flussi passeggeri in una stazione di trasporto metropolitano, 2007. http://laral.istc.cnr.it/wiva3/atti/wiva3/presentazioni/BagnoliLioSguanci.pdf A. M. Gadomski, F. Fontana, Agenti Cognitivi Semi-Intelligenti per Imprese Virtuali di e-Learning: Applicazioni di Metodologia TOGA e Tecnologie Web Multimediali. Notizie AIIA – Periodico dell’Associazione Italiana per l’Intelligenza Artificiale, Anno XVI, N.1, pagine 66-72, 2003. S. Johnson: La nuova scienza dei sistemi emergenti. Dalle colonie di insetti al cervello umano, dalle città ai videogame e all'economia, dai movimenti di protesta ai network. Garzanti, 2004.

Webgraphy http://devolab.cse.msu.edu/software/ http://it.wikipedia.org/wiki/Agente_intelligente http://it.wikipedia.org/wiki/Algoritmo_genetico http://it.wikipedia.org/wiki/DNA http://it.wikipedia.org/wiki/Vita_artificiale http://laral.istc.cnr.it/wiva3/atti/wiva3/presentazioni/BagnoÂliLioSguanci.pdf http://sodaplay.com/ http://www.airs.it/AIRS/Associazione/associazione.htm http://www.alife.org/links.html http://www.his.atr.jp/~ray/tierra/ http://www.mitpressjournals.org/loi/artl?cookieSet=1 http://www.nobleape.com/sim/ http://www.psicopolis.com/Kurt/ http://www.stauffercom.com/evolve4/

12