Volume 10, Number 3/4 1999
Lora Aroyo and Piet Kommers
University of Twente, P.O. Box 217 7500 AE, Enschede, The Netherlandskommers@edte.utwente.nl
Before the more detailed exposition of underlying mechanisms and architectures for an agent-based learning support system, we would like to provide you with the overall outline of the aspects that will be addressed in this special issue:
1. Historical developments of intelligent tutoring and support systems
2. Agent paradigms and agent-based user support systems
3. Tendencies in agent development and application
Agent technology appears to be a promising approach to address the challenges of modern day educational environments, influenced enormously by advanced information and Internet technologies. It has seen a great recognition in quite a lot of educational and training computer-based activities. The existing world of education is currently changing rapidly in respect to all new technologies and methods coming up on the market. This change is taking place as well in technological as in instructional methods used in traditional and on-line education. Intelligent agents appeared to contribute rather important advantages for the scientific and educational computing. They have a major influence in different application fields of educational systems. They provide new educational paradigms, support theories, and happen to be rather helpful entities for both students and teachers in their computer-aided learning-teaching process. Their application in the educational field is mostly as personal assistants, user guides, alternative help systems, dynamic distributed system architectures, human-system mediators, and so forth.
All these changes imply that on the one hand, increasingly complex and dynamic educational infrastructures need to be managed more efficiently and, on the other hand, new types of educational services and mechanisms need to be developed and provided. It is in particular that such services need to satisfy a diverse range of requirements, for example, personalization, adaptation, support for user mobility, support for users while coping with new types of technologies, effectiveness, information support, and so forth. Agents appear to support in a more efficient way those requirements in comparison to other already existing technologies. Besides the ability to process autonomy and cooperation among themselves, agents possess the capabilities for issues such as security, both online and offline service providing, and so forth.
It is the aim of this special issue to introduce to a wider audience, the state of the art in the agent technologies with respect to educational environments and computer-based systems. Of interest are also development environments, languages and tools, going along with agent-based systems that demonstrate the power, and the effectiveness of agent technologies applied in the field of education.
Lately an important discussion is going on with respect to the development and the design of intelligent support systems to facilitate online and web-based study environments. The main problem is to provide easy and flexible access to various data as residing in multiple, unstructured, heterogeneous information resources, and to integrate those data into semantically coherent information. Virtual environments are rather complex environments and as such, they invite one to apply new approaches for the support and facilitation of ongoing processes. The same is also true for virtual study environments. The solution is the analogy and the metaphoric way of thinking about virtual environments. This new approach introduces autonomous, goal-driven and sensitive objects as a main software entity. Following this metaphoric thinking people often refer to those objects or pieces of software as “agents” or “intelligent agents,” where the agent metaphor enables their understanding, controlling, and monitoring complex environments (Travers, 1996).
When a new paradigm is introduced, the most important issues are definitions with major characteristics, taxonomies, operationalising concepts, functionality, and application. Intelligent agents have been characterised in a large range of definitions. An extensive overview of them is not the deliberate purpose of this article. A good overview can be found in (Franklin & Graesser, 1996) and (Jennings & Wooldridge, 1995). In particular, there is no real agreement on what an agent is. Agents’ abilities vary significantly, depending on its roles, capabilities, and environments. In order to describe these abilities, different notions of agents have been introduced. Intelligent agents are introduced by most of the researchers (Wooldridge & Jennings 1998; Jonker, 1998) with four major concepts defining their behaviour: autonomy, responsiveness or reactiveness, pro-activeness and social ability. There is also a strong notion on the characteristics of agents, which refer to adaptiveness, pro-activity and intentionality (Denett, 1987; Meyer, 1997; Rao, 1996). There are also various taxonomies created for agents. According to Nwana (1996) there are seven categories of agents—collaborative, interface, mobile, information, reactive, hybrid, and smart agents. Franklin and Graesser (1996) identify a taxonomy, which is enhanced with categories of Nwana and restructured. Some new categories are included, such as “life-like” agents. In this context, intelligent agents have been associated with a variety of functions, for example, personal assistants, information managers, information seekers, planning agents, co-ordination agents or collaborative schedules, user representatives, and so forth.
In this paper we take a closer look at agents from three different points of view: architectural, functional, and structural.
From an architectural point of view agents are acting as mediators between users performing specific tasks and a specific software application. They are often decomposed into functional modules, that refer to perception, modelling, planning, coordination, and task or plan execution. The strong point is that the agents provide the ability to execute and coordinate independently and autonomously sub-tasks, which lead to completing the major user task.
Recent trends have made it clear that software complexity will continue to increase dramatically in the coming decades. The dynamic and distributed nature of both data and applications require that software not merely respond to the request for information but also anticipate, adapt, and actively seek ways to support users. In this context users expect dedicated assistance from the applications they are using. In response to these requirements, software agents play an important role in the human-computer interaction and in the coordination of the internal processes of the system.
From a functional point of view we see agents in the context of the different roles they could be applied to. Conceptual structures play an important role in the functional operationalisation of the agent. They could be applied for knowledge representation and its manipulation mechanisms, navigation, and problem definition. In this context ontology plays an important role.
The term ontology has been recently adopted by the artificial intelligence community to refer to a set of concepts or terms that can be used to describe some area of knowledge or build a representation of it (Swartout & Tate, 1999). Ontologies bring quite a fundamental change in the way systems are constructed. Also the combination with agents creates a real future-oriented approach in coping with the complexity of the new information age. They are referred as vocabularies, content theories, world descriptions, shared and reusable knowledge representations, and task definitions. Agent-based intelligent systems in this respect can provide a complete new understanding of the computer and its functionality.
The structural point of view concerns the intersection and combination of theories, implementation, and the usage of intelligent agents in the specific application area of education. The introduction of agent technologies sketches the history of this new paradigm. Different points of reference are used to present a clear and structured picture of the main issues involved, concepts and considerations for the design and realisation of agent-based educational software.
This volume contains selected papers on intelligent agents applied in the field of computer-based education and learning. The idea was to provide an overview and analysis of the ideas presented by the numerous research projects. The articles in this journal issue are mainly oriented towards presenting multi-agent systems for collaborative support. They represent the evolution of intelligent tutoring systems influenced by the agent approach. There are 11 papers grouped around several topics:
Agent-based tutoring systems: Intelligent web-based educational system, multi-agent tutoring system, collaborative intelligent tutors. This section is dedicated to tutoring systems that are enhanced by using agent technology for user-oriented education and training support.
Instructional design and learning theories: Modelling the intelligence in instructional processes and students learning. The selection of articles focuses on instructional design issues and learning theories realised by means of agent-based architectures.
Agent-based development tools and environments. This section provides examples and perspective on development tools and environments for the realisation of agent-based educational software. It gives an overview of some existing well-known development applications and presents a real time object-oriented agent development system used to build an embedded distributed rational agent that can manage the acquisition and presentation of multi-media information in a distance learning component of a course.
The following sections focus on different types of agent-based personal assistants and user support tools. They present examples of single user and collaborative learning support, navigation tools for learning materials, interface agents, virtual laboratories, and simulation environments.
This special issue covers topics and presents papers that illustrates important points in the evolution of Intelligent Tutoring Systems with the existence of the agent paradigm. Most of the papers present both recent prototyping work and already implemented systems, tested in real learning environments. A variety of tools have been explored and used for the construction of the systems, like IBM Agent Builder Environment used by Trang Dang, Hamada Ghenniwa, and Mohamed Kamel for the construction of an instructional assistant interface agent within the Practical Algebra Tutor (PAT) system. Here we also find an application of Java Abstract Window Toolkit (AWT) components for building interaction dialogues of the interface agent and Knowledge Interchange Format (KIF) is used to build the agent’s knowledge base.
Many of the papers included in this special issue are dealing with multi-agent systems. The paper of Konstantin Solomonos and Nikolaos Avouris presents a multi-agent system where the communication language among numerous tutoring and broker agents is based on the well-known KQML. In this paper we also see a servlet-based development of the agent society.
Colm O’Riordan and Josephine Griffith present interesting work in the field of web-based educational systems focusing on agent-based support for information management and user modelling. The system adopts the agent-based approach to provide pro-active learning by maintaining a user model, and by supporting peer learning via multi-agent collaboration. The emphasis of this paper is on the question of how to take full advantage of the possibilities provided by World Wide Web by applying agents. This on the other hand, raises questions of how agents can help in overcoming some disadvantages of this medium as well.
Mahmoud M. El-Khouly, Behrouz H. Far, and Zenya Koono take the perspective of the current intelligent computer aided instruction systems and present the advantages of applying agents in them to overcome some shortages, like language dependence and not precise extraction methods for student’s knowledge. The authors present intelligent agent-based tutorial systems consisting of two personal assistants for teachers and students (PAA-T and PAA-S).
Jaakko Kurhila and Erkki Sutinen present in their paper, a learning environment, in which an agent provides the user with navigation support through multi-dimensional learning space. It takes the perspective of comparing traditional tutoring systems with the proposed type of learning environments. The system Ahmed is an adaptive hypermedia system targeted for children with deficiencies in mental programming. An interesting new instructional concept is introduced—learning seeds, referring to educational space items closely related but higher in value to CAI frames.
The paper of Leonid Sheremetov and Gustavo Nunez presents the usage of multi-agent frameworks in virtual learning spaces. Quite interesting and innovative is the conceptual architecture of EVA structured into different learning spaces and together with the presentation of unifying framework for distributed heterogeneous components in EVA. It reports research on a Configurable Collaborative Learning. It focuses on theoretical and practical issues of artificial learning companions, personalised collaborative assistants, planning, experimenting, and virtual laboratories.
A valuable contribution to the special issue is the paper of Leonard Wesley, Simon Shim, Robet Booth, and Shreemathi Atreya. They take the perspective of an agent development system and give a consistent overview of the existing technologies and developments in this respect, for example, IBM Agent Building Environment, Standford’s Java Agent Template, the Java Expert System Shell, and many others. The authors present the ROADS system - a real time object-oriented agent development system used to build an embedded distributed rational agent that can manage the acquisition and presentation of multi-media information in a distance-learning component of a course. The ROADS system suggests how agents can provide better and purposeful Distance Learning Environments.
Other papers deal with theoretical issues in respect to instruction and how agents could support it. In their paper, Enrique Espinosa and Fernando Ramos address general didactic theories and theories of instructional system design. They present a modal-logic-driven agent system and report experimental research of user tests done with the prototype agent system, which makes the paper rather relevant and an interesting contribution to the special issue.
The article of Janice Whatley, Geof Staniford, Martin Beer, and Phil Scown reviews and synthesises the most important problems encountered when students perform work on group projects. It presents a multi-agent architecture for student support in their on-line group work. The help provided in collaborative computer aided environments is concentrated mainly in support of maintenance roles, such as planning the work, task allocation, and monitoring progress, as well as recognising problems as they arise. Agents are applied in order to solve problemssuch as, the lack of skills, the lack of time, and the absence of human expertise.
Finally, in the last articles of this special issue we bring together the latest empirical evidence on the cognitive and creative impact of agent-support for concept mapping, in particular for problem solving in advanced learning settings. These article attempts to provide the reader with a potential experimental model that should be practiced for a longer time, aiming at ever more clearly pinpointing the net added value of the propagated tools and methods. We hope that this special issue also stimulates you to pursue the many still unsought metaphors and paradigms for the growing need in autonomous and still effective learning environments.
Konstantinos Solomos and Nikolaos Avouris
Learning Technology Group Electrical and Computer Engineering Department, University of Patras GR-265 00 Rio-Patras, GreeceN.Avouris@ee.upatras.gr
This paper describes an open distributed multi-agent tutoring system (MATS) and discusses issues related to learning in such open environments. MATS is a prototype that models a “one student—many teachers” learning situation. Each MATS agent represents a tutor, capable of teaching a distinct subject. All MATS tutors are also capable of collaborating with each other for solving learning difficulties that their students may have. In order to build this prototype, the following parts of the architecture had to be defined: an adequate agent architecture and multi-agent platform, a knowledge interchange language suitable for learning tasks, and a general ontology of learning environments as a foundation for knowledge sharing. MATS can be used for supporting collaboration of heterogeneous learning objects and for this reason is an interesting paradigm of learning in the rapidly expanding, open distributed world of knowledge surrounding us. The challenges that the learners face when participating in such environments are also discussed in the last part of the paper where the learners’ roles in the MATS context are described.
Colm O’Riordan and Josephine Griffith
Information Technology Centre National University of Ireland Galway, Irelandcolmor@it.nuigalway.ie
Given the increasingly widespread use of computers, it is not surprising that both students and course-providers are focusing more attention on using computers as an educational aid. A popular area of research has been the delivery of course material via the Internet. The majority of such courses focus predominantly upon content and do not exploit the full capabilities of the available technology. By utilising the potential of current technology, one can provide an intelligent, personalised, and adaptive web-based education system. In our system, we adopt an agent-based approach to provide for pro-active personalised learning by maintaining a user model, and to support peer-peer learning via multi-agent collaboration. Our system also incorporates intelligent information filtering agents to facilitate the automatic gathering of related course material. We attempt to accurately represent users’ ability and progress and subsequently use this representation to guide automatic modification of content. This paper describes the system architecture—user modelling agents, information filtering agents, and the multi-agent interaction—and also summarises implementation details and initial results.
Mahmoud M. El-Khouly, Behrouz H. Far, and Zenya Koono
Saitama University, Faculty of Engineering, Information and Computer Sciences Department, FAR Lab T 338, Urawa-Shi, Shimo-Okubo, 255 Saitama-ken, Japan{elkhouly,far,koono}@cit.ics.saitama-u.ac.jp
This paper presents a new vision for intelligent computer aided instuction (ICAI) in the presence of agent technology. An agent-based computer tutorial system consists of two sub-agents; (a) personal assistant agent for teachers (PAA-T), and (b) personal assistant agent for students (PAA-S). PAA-T that allows the teachers to cope with the knowledge base of a computer language under investigation, and add or modify the commands’ structure that will be taught. Also, this agent can generate a new tutoring dialog for a new computer programming language by consulting previous tutoring dialogs for another computer programming language. PAA-S contains a student model and a tutoring module. In the student model, the system can accept free-format answers from the student, and check it against the language structure. Tutoring text has been separated from the tutorial module, such that the student’s mother tongue can be used. The system is suitable for any computer procedural language (e.g., FORTRAN, PASCAL, etc.). The system has been tested in some schools, and the feedback has been taken into consideration. Using these kinds of agents allows us to expand their features to include communication with other agents and to exchange teacher’s experiences as well as tutoring dialogs.
Jaakko Kurhila
Department of Computer Science P.O.Box 26 (Teollisuuskatu 23) FIN-00014 University of Helsinki, Finlandkurhila@cs.helsinki.fi
Erkki Sutinen
Department of Computer Sciences, Purdue University 1398 Computer Science Building West Lafayette, IN 47907-1398, USAWe present a learning environment where an agent supports a learner through a multi-dimensional learning space by finding challenging pieces of learning material and problem solving tasks to the learner. The individualized path through the space helps the learner to construct his or her own model of the topic in the learning space. Our learning environment differs from traditional intelligent tutoring systems. The aim is to leave the responsibility of the pedagogics of the material to human experts creating the learning material. The learning environment is implemented to the context of special education. The intended users have motor impairments, but also difficulties in organizing a given assignment, keeping up their attention, or navigating towards a goal. Similar problems occur frequently within other user groups but with different learning spaces, like those of project planning or mastery of life; the scheme presented can also be applied to these areas.
Leonid Sheremetov and Gustavo Núñez
Agents Laboratory, Computer Science Research Center National Technical University (CIC-IPN), Mexicosher@cic.ipn.mx
This article reports on the first results of the research work within the paradigm of Configurable Collaborative Learning, which we have named EVA (stands for Virtual Learning Spaces in Spanish). The main purpose of this project is to develop models, architectures, and a 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.
Leonard P. Wesley, Simon S.Y. Shim, and Shreemathi D. Atreya
Computer, Information and Systems Engineering Department College of Engineering, San Jose State University One Washington Square San Jose, CA 95138lwesley@intellexus.com, sishim@email.sjsu.edu, atreyas@earthlink.net
Robert P. Booth
Motorola, Design/Ver Tools, Core & System Technology 6300 Bridgepoint Pkwy, Bldg #3 Austin, TX 78730rbooth@ibmoto.com
We describe work that is intended to extend the state-of-the-art in intelligent agent development environments (ADEs) and is intended to advance the use of automated agents to support distance learning (DL). A real-time object-oriented agent development system (ROADS) has been developed and used to build an embedded distributed rational agent that manages the acquisition and presentation of multi-media information in a distance learning component of an undergraduate data structures course. Some ROADS innovations include the use of a theory of objects as a foundation on which: (a) different agent development languages can be defined and used within a single ADE; (b) computational models of beliefs desires and intentions are readily implemented; (c) possible world semantics are readily supported; (d) reactivity and responsiveness are under direct dynamic control of the agent applications, and (e) a formal foundation to model cooperative multi-agent applications is provided. Some DL innovations include novel and effective ways to (a) model a user’s knowledge about course material; (b) manage the acquisition and presentation of distributed multi-media course material, and (c) profile user DL preferences and habits. Results and examples are presented that suggest the design of ROADS and ROADS agents begin to bridge existing ADE and DL technical gaps.
Enrique Espinosa
Instituto Tecnológico y de Estudios Superiores de Monterrey ITESM-Campus Ciudad de México, Departamento de Computación Calle del Puente 222 México D.F. 14380 Méxicoeespinos@campus.ccm.itesm.mx
Fernando Ramos
ITESM-Campus Morelos Programa de Graduados en Informática Paseo de la Reforma 182-A Lomas de Cuernavaca. Cuernavaca, Mor. México 62589framos@campus.mor.itesm.mx
Instructional Design, the pedagogical technique typically used to design computer based education software, including intelligent tutoring systems, relies on a set of correctness metrics called Instructional Integrity. Hereby, curricula should describe, in unambiguous terms, the structure of the information any student must acquire. A time interval is expected to have elapsed upon each transition among these states. It is assumed that knowledge must have been transferred to the student at the end of each interval. This is clearly consistent with Instructivism, since knowledge is quantifiable and incrementally administered. In reality, however, learning will occur holistically, in time. As a result, agent measuring and tutoring of the learning process at the end of predetermined causal-time intervals results in Temporal Holes where important events might be taking place. We deal with these holes by modeling the student’s learning process rather than the knowledge to be learned. We attempt this by using Agent-Prone Modal-Temporal Logic Specifications on a modified Interface Model for a Datastructures Tutor not designed as an Instructional Graph. We demonstrate prototype software, and provide test examples within three well-known instructional methods: Didactic, Inquiry, and Discovery. We conclude by presenting conclusions and further work strands.
Janice Whatley
Business and Information Technology Stockport College of Further and Higher Education Wellington Road South, Stockport, SK1 3UQ, UKJanice.Whatley@cs.stockport.ac.uk
Geof Staniford
Computing and Mathematical Sciences Liverpool John Moores University Rodney St, Liverpool, L1 1IS, UKMartin Beer
Department of Computer Science University of Liverpool PO Box 147, Liverpool, L69 3BX, UKPhil Scown
Department of Business Information Technology Manchester Metropolitan University Aytoun Street, Manchester, M1 3GH, UKAgent technology is being applied where information overload is a problem and where intelligent help for the user is required; one such example is online higher education. Intelligent help is traditionally supplied by knowledge-based systems, but these usually work for a single domain and require the user to initiate them. Agent based help watches the user at work and perform operations autonomously when the user needs them. We suggest using agent technology to provide help for students working on group projects online. Successful group work requires that the maintenance roles of the group are given attention as well as the task roles. Each student working on the project will have an agent, operating in the background, watching progress, measuring it against the plan, and taking remedial action when necessary. These agents will interact with the other student agents to ensure that the project is completed satisfactorily. This paper describes our initial investigations into the sorts of problems encountered when students work on group projects, and describes how the means to recognize and prevent problems will be incorporated into the design for student group support agents. A method is described that explicitly involves users in the design process.
Trang Dang
Clarica 227 King Street South Waterloo, Ontario,N2J 4C5A CanadaHamada Ghenniwa
Department pf Electrical and Computer Engineering The University of Western Ontario London, Ontario N6A 5B9 CanadaMohamed Kamel
Systems Design Engineering Department, University of Waterloo Waterloo, Ontario, N2L 3G1 Canadamkamel@uwaterloo.ca
http://pami.uwaterloo.ca/kamel.html
In this era and beyond, the constant in many business environments is change. There are continuous changes in business climate, laws, markets, competition, and clients’ desires. Therefore, for an organization to survive and thrive in this era, ”life-long” training and learning for its employees becomes a necessity. Computer-based tutoring systems are playing an important role in supporting life-long learning environments. However, traditional computer-based tutoring systems lack the support for many aspects of effective training and learning, such as personalization and collaboration. To deal with these issues, a computer-based tutoring environment should be able to accommodate the constant changes of the learners’ knowledge level and needs, as well as to the knowledge domain. The goal is to create a collaborative learning environment between the learner and the tutoring software, and to proactively assist him/her in achieving the learning objectives. This paper proposes an interface agent that is autonomous, goal-driven, dynamic, and collaborative. The interface agent acts as a mediator between the learner and the tutoring system. The interface agent interacts with the learner and the tutoring systems to provide a collaborative environment. It monitors the learner’s habits and weaknesses to adapt its didactic directions. This approach also facilitates the reusability of different tutoring domains. To learners, the interface agent offers a flexible and customizable common interface that they can use for different subjects. A prototype of the interface agent was implemented using the IBM Agent Builder Environment Toolkit (ABE). It acts as a mediator between the learner and Practical Algebra Tutor (PAT) intelligent tutoring system for teaching algebra. In this implementation we demonstrated how an interface agent could be used to increase the flexibility and improve the effectiveness of traditional computer-based learning systems.
Lora Aroyo
Svetoslav Stoyanov
Piet Kommers
University of Twente Faculty of Educational Science & Technology P.O. Box 217, 7500 AE Enschede, The Netherlands{aroyo, stoyanov, kommers}@edte.utwente.nl
This paper provides results from research work done in respect to the application of agent technology within educational settings. It focuses on problem solving, information handling issues and idea generation. It is based on two research system examples: Solution, Mapping, Intelligent, Learning, Environment (SMILE) Creator and Agent based Information Management System (AIMS). Both systems present approaches for solutions to some of the basic problems within the context of computer based education and training, for example, adaptive learner support, problem solving, navigation, information structuring, presentation, and retrieval. They apply the concept-mapping approach as a basic mechanism for agents’ behaviour, information presentation technique, and instructional support. The agent-oriented approach is also applied in AIMS for the overall system architecture and design, where agents are the basic system modules. This paper reflects on the general approach for agents’ educational application presented in the papers of this special issue. It builds up a theoretical prototype of the general movement in the field and summarises its added value for learning technology.
Svetoslav Stoyanov and Piet Kommers
University of Twente Faculty of Educational Science and Technology Division of Educational Instrumentation P.O. Box 217, 7500 AE Enschede, The Netherlands{stoyanov, kommers} @edte.utwente.nl
This article presents an experimental verification of a hypothetical construct explaining the basic mechanism behind the behavior of an intelligent agent implemented in the Solution, Mapping, Intelligent, Learning Environment (SMILE) performance supported system. The SMILE agent, (called “facilitator”) supports a user in learning and applying a new concept mapping method for solving ill-structured problems. This article emphasizes the facilitator’s master performer model of behavior. The model reflects upon the SMILE Maker as a problem-solving tool and especially upon the SMILE concept mapping method.
The SMILE concept mapping method is based upon the four characteristics of: expressiveness, extension, externalization, and entireness (4E) hypothetical construct. The facilitator as a master performer of the SMILE concept mapping method reacts to the user’s behavior accordingly to the extent that these characteristics are accomplished.