Multi-Agent Framework for Virtual Learning Spaces Leonid Sheremetov, Gustavo Núñez Agents Laboratory, Computer Science Research Center, National Technical University (CIC-IPN), Mexico Unidad Profesional “Adolfo López Mateos”, Zacatenco, Av. Juan de Dios Bátiz s/n, Esq. Av. Miguel Othón de Mendizabal, Col. Nueva Industrial Vallejo, CP 07738, México, D.F., México. Tel.: (+52) - 5- 729-60-00, ext. 57576. Fax: (+52) -5- 5862936 e-mail:
[email protected]
1
Multi-Agent Framework for Virtual Learning Spaces Leonid Sheremetov, Gustavo Núñez Abstract This article reports on the first results of the research work within the paradigm of Configurable Collaborative Learning, which we have named EVA (states for Virtual Learning Spaces in Spanish). The main purpose of this project is to develop models, architectures and multi-agent environment for collaborative learning and experimentation. Each space in EVA consists of a number of components, composed of a set of deliberative and auxiliary agents. The article focuses on theoretical and practical issues of personalized collaborative learning with artificial learning companions, personal learning assistants with activities planning, and experimentation activities with agents. A unified framework for the distributed heterogeneous learning environment is defined on the basis of distributed component and agent models. Prototypes of agents have been developed using Microsoft VC++, LALO, JAVA, and JATLite for Unix and Windows platforms. Key words: collaborative learning environment, interaction, multiagent system, architecture, virtual reality. Introduction Since their conception more then a quarter of a century ago, knowledge-based learning environments have offered significant potential for fundamentally changing the educational process (Chan, 1996; Youngblut, 1994). Nevertheless, despite of many expectations, few learning environments have made the difficult transition from the laboratory to the classroom, so the development of pedagogically sound tools has been the time challenge. Recent trends in complex software systems development in general and learning environments, in particular, involve collaborative work as embodied in Computer Supported Cooperative Work (CSCW) for human-computer interaction and Distributed Artificial Intelligence (DAI) for learning. These resulted from the proliferation of networks ranging from local sites to the Internet. Computer-Supported Cooperative Learning (CSCL) systems have been built on the lessons learned from the past approaches such as Computer-Based Training, Intelligent Tutoring Systems (ITS), and Interactive Learning Environments. Perhaps their most important significance is the ability to offer a collaborative environment that utilizes the experiences of other students, different teaching strategies, the group expertise recorded in knowledge base and the capabilities of software tools to arrive at a particular convergent solution in a group fashion. The notion of agents is the central part of contemporary CSCW and CSCL systems (Müller et al., 1997; Barros & Perkusich, 1996; Gordon & Hall, 1998). The investigation project, which we have named EVA (Espacios Virtuales de Aprendizaje in Spanish - Virtual Learning Spaces), applies the methodology and tools of Distance Learning, CSCW and ITS to obtain a new paradigm of the Configurable Collaborative Learning in Virtual Learning Environment (VLE) (Núñez et al., 1998). This project is dedicated to the research and development of pedagogic models and information technologies that provide spaces of knowledge, collaboration, consulting, and experimentation for supporting the learning activities of teams of students.
2
The conceptual architecture of EVA is structured into the four essential learning elements, called Virtual Learning Spaces. These spaces are: knowledge - all the necessary information to learn, collaboration - real and virtual companions that get together to learn, consultation - the teachers or assessors (also real and virtual), who give the right direction for learning and consult doubts, and experimentation - the practical work of the students in virtual environment to obtain practical knowledge and abilities. The others main characteristics of EVA are the following: 1. The "Multi-book" concept, a personalized electronic book, generated by concatenating of selected units of learning material (ULM) along the learning trajectory for each knowledge domain. 2. A multilevel knowledge representation model of the problem domain based on concept graph formalism. 3. The intensive use of agent-based tools • as virtual tutors, virtual students or learning companions, virtual personal assistants that help students to learn, • for mining information, managing, planning and scheduling learning activities, information retrieval and filtering, student assistance and tracking for evaluation of their intentions and performance, and to organize the workgroups and groups activities, and re-configure their work and • knowledge spaces. All the applications are developed for the domain of the Master and Doctorate education in Computer Science. In this paper, we consider the architecture of EVA multiagent system (MAS) and implementations of different agents belonging to different types considered in (Dillenbourg, 1997). The rest of the paper is organized as follows. In the following section, a task of planning of learning trajectories and scheduling of learning activities in loosely coupled knowledge domains by cooperative agents is considered. Next, we emphasize the role of interactions in an artificial learning community and describe the corresponding formal model of interactions. In the following sections, we consider the architectures of virtual learning community and virtual cooperative experimentation space with behaviour visualization with VRML. Finally, a unifying framework for distributed heterogeneous VLE is discussed. In conclusion we present the main results of the paper and outline directions of future work. Multiagent Planning in EVA A learner navigates the virtual learning spaces by routes (study plans) suggested in an automatic manner by EVA. So, the purpose of the planning system is to design a particular learning trajectory for each student in the knowledge space, formed by ULM, which is organized in knowledge domains. At the next stage, personalized books, called Multibooks, are armed by concatenating of selected ULM along the learning trajectory for each knowledge domain. In the same way, groups of students with similar interests are arranged. Finally, student learning activities must be scheduled to satisfy temporal constraints. This task is iteratively repeated based on evaluation results. Initial study plan and learning activities scheduling is generated on the bases (i) of student's initial knowledge in each area of Computer Science at the graduate level, and (ii) his interests in terms of sub-specialty or separated courses from the area, which defines student's final state in the knowledge space.
3
To perform planning and scheduling within the MAS paradigm, we have proposed to associate a planning agent with each domain model (fig. 1). These agents must be deliberative, because simple reactive agents, which cannot explicitly represent goals, they are trying to achieve, cannot successfully fulfill the task. That means that our agents must have goal and belief representation capabilities. When viewed from the perspective of the system goal, the global study plan appears as an AND-OR tree progressing from the system goal (at the root), down through goals and plans, to local plan fragments distributed among the agents. Of course, since this is a distributed environment, no single agent has a complete view of this tree. Insert Figure 1 here. An agent may not simply satisfy a local goal by choosing any plan fragment, but must coordinate its choice so that it is compatible with those of other agents. Pre-requisite relations are used by this agent to retrieve the information about the relationships of his plan fragments with other domains. Since local goals and decisions are interconnected, proposed multistage negotiation algorithm provides means by which an agent can acquire enough knowledge to reason about the impact of his local decisions and modify its behavior accordingly to construct a globally consistent decision. A special coordinator agent facilitates communication and performs managing activities. The algorithm consists of the following stages: 1. 2. 3. 4.
study plan generation, construction of the space of alternative plans, plan commitment, scheduling of committed plan alternatives.
Learning Activities Planning System with Multistage Negotiation is implemented using JATLite package (JATLite, 1998). It is composed of n planning agents and a coordinator agent, which inherit their methods from the RouterClient class. Agents are implemented as JAVA applets, so they also inherit methods from the Applet and Frame (to support graphic interface) classes of JDK 1.2 package. For implementation details see (Sheremetov & Núñez, 1999). Model of Interactions in Collaborative Learning Once having the plan of his learning activities, a student enters the learning spaces, where additional support by learning community agents is provided. The design of learning environments, virtual or not, aims to promote productive interactions. In this type of learning a student changes from being a passive information receiver to an active collaborator, interacting with the tutors and colleagues in the learning process. Learning does not only result from acquiring knowledge, solving problems or using tools, but also from interacting about these on-going activities with persons and agents. This primary concept has been widely accepted in the learning community. Vygotski argued that the process by means of which people learn, bases on the introspection of the patterns of social interactions (Vygotski, 1978) and, in more general terms, Piaget considered that children acquire conscience of the reality by means of their interactions with the environment (Piaget, 1954).
4
Many aspects of the interaction problem have been formally and empirically studied in the works on collaborative discourse (Rich & Sider, 1997), as well as in the MAS community. However, although various authors have indicated the importance of developing a formal theory of the interaction, this problem is still a challenging issue (Agre, 1995). For example, Wegner (1995) considers the interaction problems as not algorithmic, and that is why proposes an extension of the Turing Machine - which he denominates an interactive machine - that allows to consider external actions or events during calculations. Nevertheless, this approach should be developed to become a formal theory of interaction. Bretier & Sadek (1997) proposed a formal theory of interaction, which is expressed in a first-order multimodal logic of mental attitudes (belief, uncertainty, and intention) and actions. Nevertheless, this proposal is rather a theory of communication than that of interaction, and though communication is a necessary condition of interaction, these concepts are not equivalent. In this section we focus on an outline of our logic formalization of interactions. Our emphasis lies in the role of interactions in an artificial learning community as a group of real and artificial learners, tutors and facilitators, working, supporting and learning from each other. Moreover, we shall bypass the problems of learner and artificial tutor interaction in one-on-one teaching model, which have broad coverage in the literature on ITS and ILE, for instance in (Dillenbourg et all., 1993; Barros & Perkusich, 1996). In our approach to the formalization of the interaction, we use a propositional modal logic based on the three-valued strong logic of Kleene with operators of knowledge (K) and belief (B) valued as true (v), false (f) or indefinite (u) and with an extended possible world semantics. We call it the Information Epistemic Logic (IEL) (Alvarado & Nuñez, 1997). In the IEL, the information that has each agent constitutes a lattice, whose nodes are the possible worlds and the accessibility relation between the worlds is an information order: a world w is more informed than another one w', if they have true and false statements and, at least, an indefinite statement in w', which is true or false in w. The indefinite statements (i.e. with value u) of an agent, computational or human, become true or false as a result of the process of interaction with the other agents and the environment. So, the logic that is to be used to model agents (human or artificial), must be necessarily non monotonic to give them a capacity to add and/or to modify their knowledge base. All true statements are known by an agent. In other words, if K (ϕ) is true, it means that ϕ is true in every most informed world. For each and every stage of learning, the indefinite statement (ϕ) can be believed (Β (ϕ)) or known (K (ϕ)), as soon as ϕ becomes true in some or every world, respectively. This is a monotonic process. On the other hand, if ϕ becomes undefined in some of these worlds by adding new information as a result of interaction, then K (ϕ) becomes undefined, as well. In this case , the process becomes non monotonic. In EVA, an interaction is a relationship between agents. Formally, if we designate by A a set of all the agents, an interaction of order n (n ≥ 2) is a n-arity relation over A, that is to say, is a subset of n times of the Cartesian product of A, A ×... × A = An . A set of all n
interactions of order n of A, is the power set of An, 2A , and a set of all the interactions of +∞
any order of A, is
U2 n =2
An
.
5
In any particular domain, we are generally interested not in all possible interactions, but only in a subset of them. For example, in EVA the relationships that we are interested in, are those established for a group in terms of interaction categories (see table 1). In general, let I2,...,In be the groups of interactions of order 2,... ,n respectively, which are of interest for a certain domain. For each k=2,... ,n and for each α∈ A, are defined the following sets:
Ik− 1(α) ={(α1,...,αi− 1,α,αi+ 1,...,αk ) 1≤i ≤k, (α1,...,αn ) ∈ Ik }
I k− 1 (α ) ⊆ Ak Intuitively, I k− 1 (α ) is identified with the interactions of order k that α has with the agents of A. Finally, the set of all interactions that α has with all the agents of A, I(α), is defined as follows: I (α ) =
∞
UI
−1 k (α )
k =2
In EVA, two agents α,β∈ A are equivalent from the point of view of an interaction with other agents if, and only, I(α)=I(β). Note that this relationship is the equivalence in A. Intuitively, the equivalence class of an agent under this relationship, defines all the agents of A that are undistinguished with α, from the point of view of his interactions with the other agents. This formalizes the fact that agents can be characterized in terms of their interactions. EVA Virtual Learning Community All the agent's interactions are described in terms of interaction categories (Table 1) (Anzieur & Martin, 1971). Table 1. Interaction categories Answers
Interaction Categories
Questions
Communication
query for information, repetition, confirmation query for opinion, evaluation, analysis, to express feelings query for suggestion, instruction, possible actions
information, confirmation, explanation, clarification opinion, analysis, manifest of feelings and desires
Interaction Categories
Positive
Negative
Decision
confirm, accept easily and understand try to diminish tension, smile, show satisfaction, laugh and is happy prove solidness, help, encourage his partners
discard, reject passively
Evaluation
Influence
Tension
Integration
hints and indications, respecting the independence of the others
declare tension, ask for help, retire, show unsatisfaction, displeasure discordance, disapprove passively, reject help
6
The agents who take part in the process of learning/education of EVA are human (students and advisers) and computational (planning, companion, monitor, etc.). All of them are completely characterized by the set of their interactions with the others. The common knowledge of agents who constitute a learning group in EVA, is formed by the intersection of knowledge that has each agent of a group. Our prototype of learning community incorporates a monitor agent and a learning companions agent (fig. 2). Agents have mental states represented in terms of beliefs, knowledge, commitments, with their behaviour specified by rules. These agents have been implemented as JATLite agents with rule-based inferencing capabilities, programmed in Jess (Friedman-Hill, 1998). The most attractive feature of Jess is that originally inspired by the CLIPS expert system shell, Jess has grown into a complete, distinct Java-influenced environment of its own. So, Jess can easily be integrated into Java-oriented tools and environments for intelligent agents programming (Watson, 1997). Insert here figure 2 Collaborative activity starts with the selection of the goals of pedagogical curriculum. It is based on an concept graph model, where concepts with associated teaching goals and learning activities (including the group ones) are represented as a network with prerequisite and precedence relations. It also depends on the group profile and its performance history. In EVA, the following group discussion attributes are used: initiate ideas about a topic, follow the existing focus of a conversation by responding to previous contributions, justify opinions, question other group members, and keep the conversation on task. To provide communication support for specific discussion skills, structured conversation tools are used, such as HyperNews and structured chats, making possible the classification of human communications to be adsorbed by agents. Each group has an associated Monitor Agent. He maintains the shared knowledge model of a group and compares it with a group problem model from the knowledge base that contains the objectives, concepts, activities, etc. that characterize a group. His behavior is guided by a set of domain independent conversation rules, which refers to the interactions between the group members. During the group session, the monitor agent maintains the current problem state (shared group knowledge space) and the history record of all contributions for each participant. The problem state in terms of shared beliefs and knowledge is used to change the interaction mode, choosing one of monitoring techniques from his rule base. This result in changing behavior of learning companion from group leader (strong companion) through a week companion to a passive observer. This monitoring is implemented in a very simple way: each learner communication creates as a side-effect an instance of the class 'interaction-mode'. This class possesses two variables 'more-guidance' and 'less-guidance'. Student's questions like 'information request' from associated ULM or 'asking for instructions' increments the weight of 'more-guidance'. Student's answers and other communication actions, like 'confirmation', 'provide analysis', and etc. increment 'less-guidance' variable. Companion Agents can differ in expertise level and role strategy, using 3 different pedagogical styles from supportive to non-supportive ones that can change their role in the group. Strong companions can adsorb the role of a group leaders, developing and
7
explaining problem solution, while the weak one is used to enhance student's role in a group and stimulate his learning abilities by enforcing him to teach his companion. The conversational categories are used to classify interactions and generate learning companion's messages and are mapped into a set of communicative actions, expressed in Knowledge Query and manipulation Language (KQML) (Finin, 1997). Emotions are simply visualized by means of companion's appearance. In his individual learning activities, students also receive support from assistant agents, the details of their implementation are beyond the scope of this article. These agents are: Internet search and filtering agent: automatically searches for additional teaching material on the Internet, not yet located in the Knowledge Space, according to student's learning trajectory and through natural language searching tool Clasitex (Guzmán, 1998). Evaluator agent: verifies periodically the learning advances, tries to find causes of misunderstanding, communicates with the advisor agent to re-organize the information to study. Personal assistant: classifies their e-mails, reminds a student about events, conferences, videoconferences, courses, tasks, tests, etc. Injector of agents: places demons in remote places, using the available network, works under a variety of network types, communication types, and protocols. It supposes a hostile or disordered environment in the host, therefore needing to send several agents to become familiar with the host's environment (Martínez, 1998). At the current stage of experiments, the prototypes of search/filtering and injector agents have been implemented, while scheduling and evaluator agents are being developed. Agents for Virtual Laboratories The experimentation is one of the essential parts of learning activities. Even though the virtual learning environments has been studied from a pedagogical perspective in the last years, few researchers has formulated clear methodologies how to create Virtual Laboratories in the WWW, for teams whose members are geographically separated (Das, 1994; Satava, 1994; Rodríguez et al., 1998; Forbus & Kenneth, 1996), even less, based on Virtual Reality (VR) scenarios (Lemays et al., 1996). The virtual space for experimentation in EVA contemplate in an integrated way, the visualization of experiment scenarios and agent behaviour, assistance in performing the experiments that enables configuration, performing and further explanation of results with virtual learning companions, and collaborative work in the distributed environment, based on VR interfaces. Hence, we propose the name Virtual Space of Cooperative Experimentation, as a successor o extension of a Virtual Laboratory concept. In the architecture of the virtual multiagent laboratory, the following elements are distinguished (fig. 3): Insert Figure 3 here. • Behaviour Agents: Agent 1, Agent2, ... , Agentn. These agents form the multiagent system and their visualization must be generated in virtual environment. • Agent Name Server (ANS). It plays acquaintances functions and maps logical names of agents to physical addresses of the platforms, where they are executed to establish peer-to-peer communication. ANS is a part of LALO environment (Gauvin, 1995).
8
• Facilitator Agents. These agents are in charge of receiving the requests to update data and to generate the visual representations of the Behavior Agents in virtual environment. • Interface Agents. The function of these agents is to generate a user's interface (a Web page) and to allow the user to control the Behavior Agents through KQML performatives. In this way, it is possible for example, to know agent's commitments or to modify them to make certain tasks. • Experimentation companion. This agent pertains to the learning companion type, discussed above. His particular tasks in the experiment environment is to follow the user in communicating with Behaviour Agents, verifying KQML messages, suggesting tasks and making proposals on how to resolve them. The implementation scheme is based on the integration of four programming languages: (1) LALO like agent programming language, (2) VRML for the generation of the virtual environment and the graphical representation of these agents allowing to generate 3D representations of dynamical scenes through a browser, (3) C++, resulting code of the compilation of LALO source programs, (4) Java like interface between VRML and C++. The communication between the Behavior, Interface and Facilitate Agents, since all of them are programmed finally in C++ and under the schemes generated by LALO, is transparent through KQML performatives. The Facilitator Agent incorporate additional code that allows to decode this performatives and transfer this information to a VRML program to update the visualization of behavior agents. In order to get the information in VRML, we use Java script. The mechanism of sockets plays a very important role in this schemata. In order to establish communication between Facilitator Agent (programmed in C++) and Java script (which communicates with the VRML program) we have defined two mechanisms: the CVRSocket and JVRSocket, denominated in general way VRSockets. They allow bidirectional communication between the Java script program and the Facilitator Agent and the compatibility of information transferred between C++ and Java by means of a special protocol. To assign behavior (described through a program in C++ or Java) to virtual world components programmed in VRML, a tool called EasyVRML has been developed (Quintero et al., 1998). We have developed several applications for the experimentation space of "Distributed Intelligent Systems" course, which allow the students to develop their agents, execute them in the multi-agent environment, visualize their behaviour and communicate with them through KQML. Examples include virtual bank system, virtual travel agency, and virtual manufacturing system. A Unifying Framework for Distributed Heterogeneous Components in EVA Since there exist a great number of education technologies, pedagogical models and styles, as well as software technologies to be reflected in the architecture of potential learning environment, we need to be concerned about whether any software architecture has a possibility of outliving its designers and of providing a suitable foundation for unanticipated additions of significant new features. Current consensus on these issues is
9
probably that the most robust unit of reusability is a framework making use of agents and objects or components (Bradshow et al., 1997; Grimes & Potel, 1995). In general, EVA agents are sorted into functional levels: presentation services, application services, management services, and data services (fig. 4). Agents providing presentation services are designed to hide the differences between viewers of different data types. From the point of view of the other agents, this means that there is a core viewer service protocol that is shared among all viewers. A viewer of a new kind of data need only implement an agent that converts this minimal viewer service protocol to the specific call formats that the viewer application expects. Insert Figure 4 here. The application services level currently contains any agents supporting application services, discussed in the previous sections of the article. Agent management services level is implemented according to the FIPA proposal on the basis of Microsoft DCOM model and includes the following components (FIPA, 1998; DCOM, 1996): 1) The Agent Communication Channel Router, which routes messages either to agents of the same platform or agents of different platforms. 2) The Agent Management System, which coordinates the creation, erasing, suspension, resuming and authentication of agents in the platform. 3) The Directory Facilitator, which holds a directory of services that offer each agent. This services are necessary to provide active communication among agents. Agents in EVA have to know about the capabilities of other agents. For example, when the group monitor detects the student's repetitive erroneous actions, it communicates with the learning companion to change his behaviour pattern to become more helpful. Finally, data services include data locators which encapsulate search and indexing functions, data accessors which retrieve data from heterogeneous data sources, and data monitors which feed information to clients based on user-configurable policies. Figure 5 shows the conceptual view of the overall EVA architecture, integrating principles of distributed object technology for agent communication and client-server architecture of the WEB. Specific client applications are built from various components that are integrated via an open presentation layer bus, such as Netscape's LiveConnect. The purpose of the bus is to allow HTML and client-side components (EVA agents, Java, JavaScript, plug-ins and ActiveX components, ORBs) to share a common object and messaging model, enabling seamless integration of tools, services, and user-interface elements. Insert Figure 5 here. Generic agent services built on the foundation of existing distributed object services allow software components the option of using a common agent-to-agent interlingua (currently, KQML) to communicate and coordinate their actions at the knowledge-level. In this case, EVA agents use an agent-to-agent (A2A) protocol that runs on top of standard lower-level protocols such as sockets and ORPC (for example, EVA planning and experimentation MAS). In addition to standard client-server connection protocols such as HTTP, RMI, and JDBC, a connection to a server-side DCOM component bus is provided and generic agent
10
template with multi-level communication architecture, making use of KQML and ActiveX controls has been developed (Sheremetov & Smirnov, 1999). This enables developers to selectively expose their interfaces, providing a standard way for system components to provide and access required services and data from each other. This not only ensures interoperability among our end-user tools and reusable components, but also allows us to take advantage of third-party DCOM and CORBA services that can be used and customized as needed (Durfee et al., 1997). Conclusions Making intensive use of agent technology in EVA, we have developed models, architectures and multi-agent environment for personalized collaborative learning and experimentation, such as (i) three-valued non-monotonic logic (IEL) with extended possible world semantics, to model the process of collaborative learning, (ii) multiagent generic open environment, based on agent and object technologies, (iii) learning companions, (iv) personal learning assistants for planning of learning trajectories and search of additional materials, and (v) multiagent experimentation space. Our experiments to-date have demonstrated the high flexibility for the EVA MAS system as well as the interoperability with non-agent software modules. We have also detected some difficulties while working with the proposed architecture. For example, all the agents and protocols must exist (be active or able to be initialized automatically) for a service to be provided. Also, working with different databases, we have detected some problems when accessing Access with JDBC drivers on WindowsNT platform. Nevertheless, our expectation is that as the set of agents grows, the development will be easier. The work on virtual learning community, composed of learning companions and personal learning assistants is in process. We are also working on animation assignation to learning companions and tutors (especially in the experimentation space) to obtain their strong visual presence in a 3D learning environment. We have developed a number of prototypes but a lot of work is still to be done to convert them into a real VLE. Acknowledgments Support for this work has been provided by the CONACyT within the REDII Program and IPN, Mexico. The EVA project is the effort of many people. We would particularly like to thank our colleagues from Agents and Distributed Systems Laboratories of CICIPN for their determination and contribution to the system development, especially Jesús Martínez and Matías Alvarado. We also thank the students of Doctorate and Master Programs for contributing ideas and talent to EVA MAS programming, especially L. Balladares, R. Quintero, A. Armenta, B. Parra and G. González. References Agre, P.E. (1995) Computational research on interaction and agency. Artificial Intelligence 72 , 1-52 Alvarado, M. & Nuñez, G. (1997) Formalization of Knowledge and Belief Based on Kleen's Strong Logic. In Proc. of the 4th Workshop on Logic, Language, Information and Computation, Oxford Univ. Press, 1997. Anzieur, D. & Martin, J.-Y. (1971) "Los métodos". La dinámica de los grupos pequeños. Editorial Kapelusz, S.A.Buenos Aires, Argentina, 65-89.
11
Barros Costa, E., & Perkusich, A.(1996). Modeling the Cooperative Interaction in a Teaching/Learning Situation. In Intelligent Tutoring Systems: 168-176. Bradshaw, J. M., Dutfield, S., Benoit, P., & Woolley, J. D. (1997). KAoS: Toward an industrial-strength generic agent architecture. In J. M. Bradshaw (Ed.), Software Agents. Cambridge, MA: AAAI/MIT Press: 375-418. Bretier, P., & Sadek, D. (1997) A rational agent as the kernel of a cooperative spoken dialogue system; implementing a logical theory of interaction. Chan, T.W. (1996).Learning companion Systems, Social Learning Systems, and Intelligent Virtual Classroom, In Proc. of the World Congress on AI in Education, 1996, Vol. 7, No. 2, 125-159. Das H. (1994) Telemanipulator and telepresence technologies, SPIE 2351.
Distributed Component Object Model protocol DCOM/1.0 (1996), Network Working Group. Dillenbourg, P., Hilario, M., Mendelsohn, P., Schneider, D., & Borcic, B. (1993), Intelligent Learning Environments, Report on the project No. 4023-2701 "Les systèmes explorateurs intelligents", University of Geneva Switzerland. Dillenbourg, P., Jermann, P., Schneider, D., Traum, D. & Buiu, C. (1997), The design of MOO Agents: Implementations from an Empirical CSCW Study, In B. Du Boulay and R. Mizoguchi (Eds.) Artificial Intelligence in Education, IOS Press, 15-22. Durfee, E.H., Kiskis, D.L., & Birmingham, W.P. (1997) The agent architecture of the University of Michigan digital library, In Huhns, M.N. & Singh, M.P. (Eds.) Readings in agents, Morgan Kaufmann Publ., Inc., 98-108.
Finin, T., Labrou, Ya., & Mayfield J. (1997) KQML as an agent communication language, in J. Bradshaw, ed., Software Agents, AAAI/MIT Press, 1997. FIPA7412, (1998). Draft 1.0, document of the Foundation for Intelligent Physical Agents (FIPA). Forbus, D. & Kenneth, (1996). Articulate Virtual Laboratories for Science and Engineering Education; http://www.qrg.ils.nwu.edu/projects/NSF/avl.htm Friedman-Hill, E. J. (1998), Jess, The Java Expert System Shell, Version 4.3, Technical report SAND988206 Sandia National Laboratories, Livermore, CA, http://herzberg.ca.sandia.gov/jess. Gauvin, D. (1995).Un environnement de programmation orienté agent. Dans Troisièmes journées francophones sur l'intelligence artificielle distribuée et les systèmes multiagents, France. Gilmore, D. & Self, J. (1988), The application of machine learning to ITS. In Self, J. (Ed.) Artificial Intelligence and Human Learning, Intelligent Computer-Aided Instruction, Chapman and Hall, 179-196. Gordon, A. & Hall, L. (1998).Collaboration with Agents in a Virtual World. In Proc. of the Workshop on Current Trends and Artificial Intelligence in Education, 4 World Congress on Expert Systems, Mexico, 1998. Grimes, J. & Potel, M. (1995). Software is headed toward object-oriented components. IEEE Computer(August), 24-25. Guzmán A. (1998). Finding the main themes in a Spanish document. Journal Expert Systems with Applications, Vol. 14, No. 1/2, Jan/Feb. 139-148. JATLite Beta Complete Documentation (1998), Stanford University, http://java.stanford.edu Lemays, L., Murdock, K., & Couch J. (1996). 3D Graphics and VRML 2, Sams Net. Martínez G.(1998) Automatic installer of systems: mobile agents with blackboard architecture. Master of Science thesis, Electrical Engineering Department (Computer Section), Centro de Investigación y Estudios Avanzados del I.P.N. Mexico City. (In Spanish). Müller, J., Wooldridge, M. J. and Jennings, N. R., eds.(1997), Intelligent Agents III: Proc. of the 3rd Int. Workshop on Agent Theories, Architectures, and Languages, Lecture Notes in Artificial Intelligence, Vol. 1193, (Springer-Verlag, Berlin, 1997) Nuñez, G., Sheremetov, L., Martínez, J., Guzmán, A., & Albornoz, A. (1998). The Eva Teleteaching Project - The Concept And The First Experience In The Development Of Virtual Learning Spaces. In
12
Proceedings of the 15th IFIP World Computer Congress "The Global Information Society on the Way to the Next Millennium", Vienna and Budapest, 31 August - 4 September, Part II, 769-778. Piaget, J. (1954) The construction of Reality in the Child, translated by Margaret Cook. Basic Books, New York. Quintero, R., Armenta, A., Balladares, L. & Nuñez G. (1998) EasyVRML Herramienta CASE para la asignación de comportamiento a componentes de mundo VRML. XI Congreso Nacional ANIEI, Mexico, 1998 (in Spanish) Rich, Ch. & Sider, C.L. (1997). COLLAGEN: when agents collaborate with people. In Proc. of the 1st Int. Conference on Autonomous Agents, American Association for Artificial Intelligence. Rodrigues de Queiroz, L. et al. (1998). A Robotics and Computer Vision Virtual Laboratory, IEEE´98 In Proceedings of AMC´98, Coimbra, June29-July 1, 694-699. Sheremetov, L. & Núñez, G. (1999). Multi-Stage Cooperation Algorithm and Tools for Agent-Based Planning and Scheduling in Virtual Learning Environment, The 1st International Workshop of Central and Eastern Europe on Multi-agent Systems (CEEMAS'99), 30th May-3rd June, St. Petersburg, Russia. (to appear) Sheremetov, L.B. & Smirnov, A.V. (1999). Component Integration Framework For Manufacturing Systems Re-Engineering Based In Agents And Objects. International Journal of Robotics and Autonomous Systems, Elsevier, North-Holland, (to appear). Vygotsky, L.S. (1978). Mind in Society: The development of higher psychological process. In Cole, M. , John-Steiner, V. , Scribner, S. and Souberman , E. (eds.) Harvard University Press, Cambridge, MA. Originally published in Russian in 1934. Watson, M. (1997). Intelligent Java Applications for the Internet and Intranets, Morgan Kaufmann Publishers, Inc. Wegner, P.(1996). The Paradigm Shift from Algorithms to Interaction. Brow University. Youngblut, Ch. (1994) Government - Sponsored Research and Development Efforts in the Area of Intelligent Tutoring Systems. Institute for Defense Analyses, USA.
