INTRODUCTION
Governments and institutions are facing the new demands of a rapidly changing society. Among many significant trends, some facts should be considered (Silverstein, 2006): (1) the increment of number and type of students; and (2) the limitations imposed by educational costs and course schedules. About the former, the need of a continuous update of knowledge and competences in an evolving work environment requires life-long learning solutions. An increasing number of young adults are returning to classrooms in order to finish their graduate degrees or attend postgraduate programs to achieve an specialization on a certain domain. About the later, due to the emergence of new types of students, budget constraints and schedule conflicts appear. Workers and immigrants, for instance, are relevant groups for which educational costs and job incompatible schedules could be the key factor to register into a course or to give up a program after investing time and effort on it. In order to solve the needs derived from this social context, new educational approaches should be proposed: (1) to improve and extend the online learning courses, which would reduce student costs and allows to cover the educational needs of a higher number of students, and (2) to automate learning processes, then reducing teacher costs and providing a more personalized educational experience anytime, anywhere.
As a result of this context, in the last decade an increasing interest on applying computer technologies in the field of Education has been observed. On this regard, the paradigms of the Artificial Intelligence (AI) field are attracting an special attention to solve the issues derived from the introduction of computers as supporting resources of different learning strategies. In this paper we review the state-of-art of the application of Artificial Intelligence techniques in the field of Education, focusing on (1) the most popular educational tools based on AI, and (2) the most relevant AI techniques applied on the development of intelligent educational systems.
EXAMPLES OF EDUCATIONAL TOOLS BASED ON AI
The field of Artificial Intelligence can contribute with interesting solutions to the needs of the educational domain (Kennedy, 2002). In what follows, the type of systems that can be built based on AI techniques are outlined.
Intelligent Tutoring Systems
The Intelligent Tutoring Systems are applications that provide personalized/adaptive learning without the intervention of human teachers (VanLehn, 2006). They are constituted by three main components: (1) knowledge of the educational contents, (2) knowledge of the student, and (3) knowledge of the learning procedures and methodologies. These systems promise to radically transform our vision of online learning. As opposed to the hypertext-based e-learning applications, which provide the students with a certain number of opportunities to search for the correct answer before showing it, the intelligent tutoring systems perform like coaches not only after the introduction of the response, but also offering suggestions when the students doubt or are blocked during the process of solving the problem. In this way, the assistance guide the learning process rather than merely saying what is correct or what is wrong.
There exist numerous examples of intelligent tutoring systems, some of them developed at universities as research projects while others created with business goals. Among the first ones, the Andes systems (Van-Lehn, Lynch, Schulze, Shapiro, Shelby, Taylor, Treacy, Weinstein & Wintersgill, 2005), developed under the guidance of Kurt VanLehn of the University of Pittsburg, is a popular example. The system is in charge of guiding the students while they try to solve different sets of problems and exercises. When the student ask for help in the middle of an activity, the system either provides hints in order to step further towards the solution or points out what was wrong in some earlier step. Andes was successfully evaluated during 5 years in the Naval Academy of the United States and can be downloaded for free. Another relevant system is Cognitive Tutor (Koedinger, Anderson, Hadley & Mark, 1997), is a comprehensive secondary mathematics curricula and computer-based tutoring program developed by John R. Anderson, professor at the Carnegie Mellon University. The Cognitive Tutor is an example of how research prototypes can be evolved into commercial solutions, as it is nowadays used in 1,500 schools in the United States. On the business side, Read-On! is presented as a product that teaches reading comprehension skills for adults. It analyzes and diagnoses the specific deficiencies and problems of each student and then adapts the learning process based on that features (Read On, 2007). It includes an authoring tool that allows course designers to adapt course contents to different student profiles in a fast and flexible way.
Automatic Evaluation Systems
Automatic Evaluation Systems are mainly focused on evaluating the strengths and weaknesses of students in different learning activities through assessment tests (Conejo, Guzman, Millan, Trella, Perez-de-la-Cruz. & Rios, 2004). In this way, these systems not only perform the automatic correction of the test, but also derive automatically useful information about the competences and skills obtained by the students during the educational process.
Among the automatic evaluation systems, we could highlight ToL (Test On Line) (Tartaglia & Tresso, 2002), which have been used by Physics students in the Polytechnic University of Milano. The system is composed of a database of tests, an algorithm for question selection, and a mechanism for the automatic evaluation of tests, which can be additionally configured by the teachers. CELLA (Comprehensive English Language Learning Assesment) (Cella, 2007) is another system that evaluates the student competence on using and understanding the English language. The application shows the progress carried out by the students and determines their proficiency and degree of competence on the use of foreign languages. As for commercial applications, Intellimetric is a Web-based system that lets students to submit their work online (Intellimetric, 2007). In a few seconds, the AI-supported grading engine automatically provides the score of the work. The company claims a reliability of 99%, meaning that 99 percent of the time the engine’s scores match those provided by human teachers.
Computer Supported Collaborative Learning
The environments of computer supported collaborative learning are aimed at facilitating the learning process providing the students both the context and tools to interact and work in a collaborative way with their classmates (Soller, Martinez, Jermann & Muehlenbrock, 2005). In intelligent-based systems, the collaboration is usually carried out with the help of software agents in charge of mediating and supporting student interaction to achieve the proposed learning ob.ectives.
The research prototypes are the suitable test-beds to prove new ideas and concepts, to provide the best collaborative strategies. The DEGREE system, for instance, allows the characterization of group behaviours as well as the individual behaviours of the people constituting them, on the basis of a set of attributes or tags. The mediator agent utilizes those attributes, which are introduced by students, in order to provide recommendations and suggestions to improve the interaction inside each group (Barros & Verdejo, 2000). In the business domain there exist multiple solutions although they do not offer intelligent mediation to facilitate the collaborative interactions. The DEBBIE system (DePauw Electronic Blackboard for Interactive Education) is one of the most popular (Berque, Johnson, Hutcheson, Jovanovic, Moore, Singer & Slattery, 2000). It was originally developed at the beginning of year 2000 at the University of Depauw, and managed later by the DyKnow company, which was specifically created to make profit with DEBBIE (Schnitzler, 2004). The technology that currently offers DyKnow allows both teachers and students to instantaneously share information and ideas. The final goal is to support student tasks in the classroom by eliminating the need of performing simple tasks, as for instance backing up the teacher’s presentations. The students could therefore be more focused on understanding as well as analyzing the concepts presented by the teacher.
Game-Based Learning
Learning based on serious games, a term coined to distinguish between learning-oriented games used in education and purely entertaining-oriented games, deal with the utilization of the motivational power and attractiveness of games in the educational domain in order to improve the satisfaction and performance of students when acquiring new knowledge and skills. This type of learning allows to carry out activities in complex educational environments that would be impossible to implement, because of budget, time, infrastructure and security limitations, with traditional resources (Michael & Chen, 2005; Corti, 2006).
NetAid’s is an institution that develop games to teach concepts of global citizenship and to sensitize to fight against poverty. One of its first games, released in 2002, called NetAid World Class, consists on taking the identity of a real child living in India and to resolve the real problems that confront the poor children in this region (Stokes, 2005). In 2003 the game was used by 40.000 students in different Schools across the United States. In the business and entertainment arena, many games exist that can be resorted to reach educational goals. Among the most popular ones, Brain Training of Nintendo (Brain Training, 2007) challenges the user to improve her mental shape by doing memory, reasoning and mathematical exercises. The final goal is to reach an optimal cerebral age after some regular training.
AI TECHNIQUES IN EDUCATION
The intelligent educational systems reviewed above are based on a diversity of artificial intelligence techniques (Brusilovsky & Peylo, 2003). The most frequently used in the field of education are: (1) personalization mechanisms based on student and group models, (2) intelligent agents and agent-based systems, and (3) ontologies and semantic web techniques.
Personalization Mechanisms
The personalization techniques, which are the basis of intelligent tutoring systems, involve the creation and use of student models. Broadly speaking, these models imply the construction of a qualitative representation of student behavior in terms of existing background knowledge about a domain (McCalla, 1992). These representations can be further used in intelligent tutoring systems, intelligent learning environments, and to develop autonomous intelligent agents that may collaborate with human students during the learning process. The introduction of machine learning techniques facilitates to update and extend the first versions of student models in order to adapt to the evolution of each student as well as the possible changes and modifications of contents and learning activities (Sison & Shimura, 1998). The most popular student modeling techniques are (Beck, Stern, & Haugsjaa, 1996): overlay models and bayesian network models. The first method consists on considering the student model as a subset of the knowledge of an expert in the domain on which the learning is taking place. In fact, the degree of learning is measured in terms of the comparison between the knowledge acquired and represented in the student model with the background initially stored in the expert model. The second method deals with the representation of the learning process as a network of knowledge states. Once defined, the model should infer, from the tutor-student interaction, the probability of the student on being in a certain state.
Intelligent Agents and Agent-Based Systems
Software agents are considered software entities, such as software programs or robots, that present, with different degree, three main attributes: autonomy, cooperation and learning (Nwana, 1996). Autonomy refers to the principle that an agent can operate on their own (acting and deciding upon its own representation of the world). Cooperation refers to the ability to interact with other agents via some communication language. Finally, learning is essential to react or interact with the external environment. Teams of intelligent agents build up MultiAgent Systems (MAS). In this type of systems each agent has either incomplete information or limited capabilities for solving the problem at hand. Other important aspect concerns with the lack of centralized global control; therefore, data is distributed all over the system and computation is asynchronous (Sycara, 1998). Many important tasks can be carried out by intelligent agents in the context of learning and educational systems (Jafari, 2002, Sanchez, Lama, Amorim, Riera, Vila & Barro, 2003): the monitoring of inputs, outputs, and the activity outcomes produced by the students; the verification of deadlines during homework and exercise submission; automatic answering of student questions; and the automatic grading of tests and surveys.
Ontologies and Semantic Web Techniques
Ontologies aim to capture and represent consensual knowledge in a generic way, and that they may be reused and shared across software applications (Gomez-Perez, Fernandez-Lopez & Corcho, 2004). An ontology is composed of concepts or classes and their attributes, the relationships between concepts, the properties of these relationships, and the axioms and rules that explicitly represents the knowledge of a certain domain. In the educational domain, several ontologies have been proposed: (1) to describe the learning contents of technical documents (Kabel, Wielinga, & de How, 1999), (2) to model the elements required for the design, analysis, and evaluation of the interaction between learners in computer supported cooperative learning (Inaba, Tamura, Ohkubo, Ikeda, Mizoguchi & Toyoda, 2001), (3) to specify the knowledge needed to define new collaborative learning scenarios (Barros, Verdejo, Read & Mizoguchi, 2002), (4) to formalize the semantics of learning obj ects that are based on metadata standards (Brase & Nejdl, 2004), and (5) to describe the semantics of learning design languages (Amorim, Lama, Sanchez, Riera & Vila, 2006).
FUTURE TRENDS
The next generation of adaptive environments will integrate pedagogical agents, enriched with data mining and machine learning techniques, capable of providing cognitive diagnosis of the learners that will help to determine the state of the learning process and then optimize the selection of personalized learning designs. Moreover, improved models of learners, facilitators, tasks and problem-solving processes, combined with the use of Ontologies and reasoning engines, will facilitate the execution of learning activities on either online platforms or traditional classroom settings.
Research in this field is very active and faces ambitious goals. In some decades it could be possible to dream about sci-fi environments in which the students would have brain interfaces to directly interact with an intelligent assistant (Koch, 2006), which would play the role of a tutor with a direct connection with learning areas of the brain.
CONCLUSION
In this paper we have reviewed the state-of-art of the application of Artificial Intelligence techniques in the field of Education. AI approaches seem promising to improve the quality of the learning process and then to satisfy the new requirements of a rapidly changing society. Current AI-based systems such as intelligent tutoring systems, computer supported collaborative learning and educational games have already proved the possibilities of applying AI techniques. Future applications will both facilitate personalized learning styles and help the tasks of teachers and students in traditional classroom settings.
KEY TERMS
Automatic Evaluation Systems: Applications focused on evaluating the strengths and weaknesses of students in different learning activities through assessment tests.
Computer Supported Collaborative Learning (CSCL): A research topic on supporting collaborative learning methodologies with the help of computers and collaborative tools.
Game-Based Learning: A new type of learning that combines educational content and computer games in order to improve the satisfaction and performance of students when acquiring new knowledge and skills.
Intelligent Tutoring Systems: A computer program that provides personalized/adaptive instruction to students without the intervention of human beings.
Ontologies: A set of concepts within a domain that capture and represent consensual knowledge in a generic way, and that they may be reused and shared across software applications.
Software Agents: Software entities, such as software programs or robots, characterized by their autonomy, cooperation and learning capabilities.
Student Models: Representation of student behavior and degree of competence in terms of existing background knowledge about a domain.
