Ebook: Artificial Intelligence in Education
The field of Artificial Intelligence in Education has continued to broaden and now includes research and researchers from many areas of technology and social science. This study opens opportunities for the cross-fertilization of information and ideas from researchers in the many fields that make up this interdisciplinary research area, including artificial intelligence, other areas of computer science, cognitive science, education, learning sciences, educational technology, psychology, philosophy, sociology, anthropology, linguistics, and the many domain-specific areas for which Artificial Intelligence in Education systems have been designed and built. An explicit goal is to appeal to those researchers who share the perspective that true progress in learning technology requires both deep insight into technology and also deep insight into learners, learning, and the context of learning. The theme reflects this basic duality.
The 12th International Conference on Artificial Intelligence in Education (AIED-2005) is being held July 18–22, 2005, in Amsterdam, the beautiful Dutch city near the sea. AIED-2005 is the latest in an on-going series of biennial conferences in AIED dating back to the mid-1980's when the field emerged from a synthesis of artificial intelligence and education research. Since then, the field has continued to broaden and now includes research and researchers from many areas of technology and social science. The conference thus provides opportunities for the cross-fertilization of information and ideas from researchers in the many fields that make up this interdisciplinary research area, including artificial intelligence, other areas of computer science, cognitive science, education, learning sciences, educational technology, psychology, philosophy, sociology, anthropology, linguistics, and the many domain-specific areas for which AIED systems have been designed and built.
An explicit goal of this conference was to appeal to those researchers who share the AIED perspective that true progress in learning technology requires both deep insight into technology and also deep insight into learners, learning, and the context of learning. The 2005 theme "Supporting Learning through Intelligent and Socially Informed Technology" reflects this basic duality. Clearly, this theme has resonated with e-learning researchers throughout the world, since we received a record number of submissions, from researchers with a wide variety of backgrounds, but a common purpose in exploring these deep issues.
Here are some statistics. Overall, we received 289 submissions for full papers and posters. 89 of these (31%) were accepted and published as full papers, and a further 72 as posters (25%). Full papers each have been allotted 8 pages in the Proceedings; posters have been allotted 3 pages. The conference also includes 11 interactive events, 2 panels, 12 workshops, 5 tutorials, and 28 papers in the Young Researcher's Track. Each of these has been allotted a one-page abstract in the Proceedings; the workshops, tutorials, and YRT papers also have their own Proceedings, provided at the conference itself. Also in the Proceedings are brief abstracts of the talks of the four invited speakers: Daniel Schwartz of Stanford University in the U.S.A., Antonija Mitrovic of the University of Canterbury in New Zealand, Justine Cassell of Northwestern University in the U.S.A., and Ton de Jong of the University of Twente in the Netherlands.
The work to put on a conference of this size is immense. We would like to thank the many, many people who have helped to make it possible. In particular we thank the members of the Local Organizing Committee, who have strived to make sure nothing is left to chance, and to keep stressing to everybody else, especially the program co-chairs, the importance of keeping on schedule! Without their concerted efforts AIED-2005 would probably have been held in 2007! As with any quality conference, the Program Committee is critical to having a strong program. Our Program Committee was under much more stress than normal, with way more papers than expected, and a shorter time than we had originally planned for reviewing. Thanks to all of the Program Committee members for doing constructive reviews under conditions of extreme pressure, and doing so more or less on time. Thanks, too, to the reviewers who were recruited by Program Committee members to help out in this critical task. The committees organizing the other events at the conference also have helped to make the conference richer and broader: Young Researcher's Track, chaired by Monique Grandbastien; Tutorials, chaired by Jacqueline Bourdeau and Peter Wiemer-Hastings; Workshops, chaired by Joe Beck and Neil Heffernen; and Interactive Events, chaired by Lo- ra Aroyo. Antoinette Muntjewerff chaired the conference Publicity committee, and the widespread interest in the 2005 conference is in no small measure due to her and her committee's activities. We also thank an advisory group of senior AIED researchers, an informal conference executive committee, who were a useful sounding board on many occasions during the conference planning. Each of the individuals serving in these various roles is acknowledged in the next few pages. Quite literally, without them this conference could not happen. Finally, we would like to thank Thomas Preuss who helped the program co-chairs through the mysteries of the Conference Master reviewing software.
For those who enjoyed the contributions in this Proceedings, we recommend considering joining the International Society for Artificial Intelligence in Education, an active scientific community that helps to forge on-going interactions among AIED researchers in between conferences. The Society not only sponsors the biennial conferences and the occasional smaller meetings, but also has a quality journal, the AIED Journal, and an informative web site: http://aied.inf.ed.ac.uk/aiedsoc.html.
We certainly hope that you all enjoy the AIED-2005 conference, and that you find it illuminating, entertaining, and stimulating. And, please also take some time to enjoy cosmopolitan Amsterdam.
Chee-Kit Looi, Program Co-Chair, Nanyang Technological University, Singapore; Gord McCalla, Program Co-Chair, University of Saskatchewan, Canada; Bert Bredeweg, LOC-Chair, University of Amsterdam, The Netherlands; Joost Breuker, LOC-Chair, University of Amsterdam, The Netherlands; Helen Pain, Conference Chair, University of Edinburgh, United Kingdom
Schools aren't the only places people learn, and in the field of educational technology, informal learning is receiving increasing attention. In informal learning peers are of primary importance. But, how do you discover what works in peer learning? If you want to discover what peers do for one other so that you can then set up situations and technologies that maximize peer learning, where do you get your data from? You can study groups of children and hope that informal learning will happen and hope that you have a large enough sample to witness examples of each kind of peer teaching that you hope to study.
Or you can make a peer Unfortunately, the biological approach takes years, care and feeding is expensive, diary studies are out of fashion, and in any case the human subjects review board frowns on the kind of mind control that would allow one to manipulate the peer so as to provoke different learning reactions. And so, in my own research, I chose to make a bionic peer.
In this talk I describe the results from a series of studies where we manipulate a bionic peer to see the effects of various kinds of peer behavior on learning. The peer is sometimes older and sometimes younger than the learners, sometimes the same race and sometimes a different race, sometimes speaking at the same developmental level – and in the same dialect – and the learners, and sometimes differently. In each case we are struck by how much learning occurs when peers play, how learning appears to be potentiated by the rapport between the real and virtual child, and how many lessons we learn about the more general nature of informal learning mediated by technology.
Inquiry learning is way of learning in which learners act like scientists and discover a domain by employing processes such as hypothesis generation, experiment design, and data interpretation. The sequence of these learning processes and the choice for specific actions (e.g., what experiment to perform) are determined by the learners themselves. This student centeredness makes that inquiry learning heavily calls upon metacognitive processes such as planning and monitoring. These inquiry and metacognitive processes make inquiry learning a demanding task. When inquiry is combined with modelling and collaboration facilities the complexity of the learning process even increases. To make inquiry learning successful, the inquiry (and modelling and collaborative) activities need to scaffolded. Scaffolding can mean that the learning environment is structured or that learners are provided with cognitive tools for specific activities. AI techniques can be used to make scaffolds more adaptive to the learner or to developments in the learning process. In this presentation an overview of (adaptive and non-adaptive) scaffolds for inquiry learning in simulation based learning environments will be discussed.details will follow.
Constraint-based modelling (CBM) was proposed in 1992 as a way of overcoming the intractable nature of student modelling. Originally, Ohlsson viewed CBM as an approach to developing short-term student models. In this talk, I will illustrate how we have extended CBM to support both short- and long-term models, and developed methodology for using such models to make various pedagogical decisions. In particular, I will present several successful constraint-based tutors built for a various procedural and non-procedural domains. I will illustrate how constraint-based modelling supports learning and meta- cognitive skills, and present current project within the Intelligent Computer Tutoring Group.
Two claims for artificial intelligence techniques in education are that they can increase positive interactive experiences for students, and they can enhance learning. Depending on one's preferences, the critical question might be “how do we configure interactive opportunities to optimize learning?” Alternatively, the question might be, “how do we configure learning opportunities to optimize positive interactions?” Ideally, the answers to these two questions are compatible so that desirable interactions and learning outcomes are positively correlated. But, this does not have to be the case – interactions that people deem negative might lead to learning that people deem positive, or vice versa. The question for this talk is whether there is a “sweet spot” where interactions and learning complement one another and the values we hold most important. I will offer a pair of frameworks to address this question: one for characterizing learning by the dimensions of innovation and efficiency; and one for characterizing interactivity by the dimensions of initiative and idea incorporation. I will provide empirical examples of students working with intelligent computer technologies to show how desirable outcomes in both frameworks can be correlated.
The REDEEM authoring tool allows teachers to create adapted learning environments for their students from existing material. Previous evaluations have shown that under experimental conditions REDEEM can significantly improve learning. The goals of this study were twofold: to explore if REDEEM could improve students' learning in real world situations and to examine if learners can share in the authoring decisions. REDEEM was used to create 10 courses from existing lectures that taught undergraduate statistics. An experimenter performed the content authoring and then created student categories and tutorial strategies that learners chose for themselves. All first-year psychology students were offered the opportunity to learn with REDEEM: 90 used REDEEM at least once but 77 did not. Students also completed a pre-test, 3 attitude questionnaires and their final exam was used as a post-test. Learning with REDEEM was associated with significantly better exam scores, and this remains true even when attempting to control for increased effort or ability of REDEEM users. Students explored a variety of categories and strategies, rating their option to choose this as moderately important. Consequently, whilst there is no direct evidence that allowing students this control enhanced performance, it seems likely that it increased uptake of the system.
Given the important role that meta-cognitive processes play in learning, intelligent tutoring systems should not only provide domain-specific assistance, but should also aim to help students in acquiring meta-cognitive skills. As a step toward this goal, we have constructed a Help Tutor, aimed at improving students' help-seeking skill. The Help Tutor is based on a cognitive model of students' desired help-seeking processes, as they work with a Cognitive Tutor (Aleven et al., 2004). To provide meta-cognitive tutoring in conjunction with cognitive tutoring, we designed an architecture in which the Help Tutor and a Cognitive Tutor function as independent agents, to facilitate re-use of the Help Tutor. Pilot tests with four students showed that students improved their help-seeking behavior significantly while working with the Help Tutor. The improvement could not be attributed to their becoming more familiar with the domain-specific skills being taught by the tutor. Although students reported afterwards that they welcomed feedback on their help-seeking behavior, they seemed less fond of it when actually advised to act differently while working. We discuss our plans for an experiment to evaluate the impact of the Help Tutor on students' help-seeking behavior and learning, including future learning, after their work with the Help Tutor.
The authors of the Web-based courseware typically face problems such as how to locate, select and semantically relate suitable learning resources. As the concept of the Semantic Web has not yet matured, the authors resort to a keyword-based search and bookmarking. This paper proposes a tool that supports the authors in their tasks of selection and grouping the learning material. The “à la” (Associative Linking of Attributes) in Education, enhances the search engine results by extracting the attributes (keywords and document formats) from the text. The relationships between the attributes are established and visualised in a novel hypertext paradigm using the ZigZag principles. Browsing the related metadata provides a quick summary of the document that can help in faster determining its relevancy. Also, the proposed solution enables better understanding why some resources are grouped together as well as providing suggestions for the further search. The results of a user trial indicate high levels of user satisfaction and effectiveness.
A student's goals and attitudes while interacting with a tutor are typically unseen and unknowable. However their outward behavior (e.g. problem-solving time, mistakes and help requests) is easily recorded and can reflect hidden affect status. This research evaluates the accuracy of a Bayesian Network to infer a student's hidden attitude toward learning, amount learned and perception of the system from log-data. The long term goal is to develop tutors that self-improve their student models and their teaching, dynamically can adapt pedagogical decisions about hints and help improve student's affective, intellectual and learning situation based on inferences about their goals and attitude.
In this study we examined the effectiveness of self-regulated learning (SRL) and externally-regulated learning (ERL) on adolescents' learning about the circulatory system with hypermedia. A total of 128 middle-school and high school students with little knowledge of the topic were randomly assigned either to the SRL or ERL condition. Learners in the SRL condition regulated their own learning, while learners in the ERL condition had access to a human tutor who facilitated their self-regulated learning. We converged product (pretest-posttest shifts in students' mental models) with process (think-aloud) data to examine the effectiveness of self- and externally-regulated learning about a science topic during a 40-minute session. Findings revealed that the ERL condition facilitated the shift in learners' mental models significantly more than did the SRL condition. Verbal protocol data indicated that learners in the ERL condition regulated their learning by activating prior knowledge, engaging in several monitoring activities, deploying several effective strategies, and engaging in adaptive help-seeking. By contrast, learners in the SRL condition regulated their learning by using fewer monitoring activities, and using several ineffective strategies. We present design principles for adaptive hypermedia learning environments designed to foster students' self-regulated learning of complex and challenging science topics.
Formalizing a student model for an educational system requires an engineering effort that is highly domain-specific. This model-specificity limits the ability to scale a tutoring system across content domains. In this work we offer an alternative, in which the task of student modeling is not performed by the system designers. We achieve this by using a reciprocal tutoring system in which peer-tutors are implicitly tasked with student modeling. Students are motivated, using the Teacher's Dilemma, to use these models to provide appropriately-difficult challenges. We implement this as a basic literacy game in a spelling-bee format, in which players choose words for each other to spell across the internet. We find that students are responsive to the game's motivational structure, and we examine the affect on participants' spelling accuracy, challenge difficulty, and tutoring skill.
Students approach the learning opportunity offered by intelligent tutoring systems with a variety of goals and attitudes. These goals and attitudes can substantially affect students' behavior within the tutor, and how much the student learns. One behavior that has been found to be associated with poorer learning is gaming the system, where a student attempts to complete problems and advance through an educational task by systematically taking advantage of properties and regularities in the system used to complete that task. It has been hypothesized that students game the system because of performance goals. In this paper, however, we find that the frequency of gaming the system does not correlate to a known measure of performance goals; instead, gaming is correlated to disliking computers and the tutor. Performance goals, by contrast, are shown to be associated with working slowly and avoiding errors, and are found to not be correlated to differences in learning outcomes.
The current work examined the influence of pedagogical agents as social models to increase females' interest in engineering. Seventy-nine female undergraduate students rated pedagogical agents on a series of factors (e.g., most like themselves, most like an engineer, and most prefer to learn from). The agents were identical with the exception of differing by appearance/image in four aspects (age, gender, attractiveness, “coolness”). After selecting the agent from which they most preferred to learn, participants interacted with it for approximately 15 minutes and received a persuasive message about engineering. Results indicated that the women were more likely to choose a female, attractive, young, and cool agent as most like themselves and the one they most wanted to be like. However, they tended to select male, older, uncool agents as the most like engineers and tended to choose to learn about engineering from agents that were male and attractive, but uncool. Interacting with an agent had a positive impact on math-related beliefs. Specifically, the women reported more positive math and science related beliefs compared to their attitudes at the beginning of the semester and compared to a group of women who did not interact with an agent. Further, among the women who viewed an agent, the older version of the agent had a stronger positive influence on their math-related beliefs than the younger agent.
Mitigating frustration is important within computer-based learning contexts. In this experimental study where participants were purposefully frustrated, the interface agent message (apologetic, empathetic, or silent) was manipulated to investigate its impact on student attitude toward the task, attitude toward the agent, and attribution toward the cause of frustration. Fifty-seven undergraduate students responded to an invitation to participate in a web-based survey and to receive a movie ticket for their effort. An animated interface agent, “Survey Sam,” was present as students answered survey items and were confronted with a frustrating obstacle – an error message pop-up window that blocked them from answering the survey items. Survey Sam delivered either an affective message (apologetic or empathetic) or remained silent to the thirty students who actually completed the survey. Results revealed that the presence of an affective message (either apologetic or empathetic) led participants to report significantly greater frustration, suggesting that the affective message reinforced and validated their frustration. However, and more importantly, they attributed the cause of their frustration to the program instead of to themselves (as did the no message group). A comparison of message type (apologetic or empathetic) indicated that participants receiving the empathetic message rated Survey Sam as significantly more believable and sincere. Implications of these findings as a catalyst for further research in the development of frustration-mitigating support for computer-based contexts are discussed.
Intelligent tutoring systems customize the learning experiences of students. Because no two students have precisely the same learning history, traditional analytic techniques are not appropriate. This paper shows how to compare the learning histories of students and how to compare groups of students in different experimental conditions. A class of randomization tests is introduced and illustrated with data from the AnimalWatch ITS project for elementary school arithmetic.
Time on task is an important predictor for how much students learn. However, students must be focused on their learning for the time invested to be productive. Unfortunately, students do not always try their hardest to solve problems presented by computer tutors. This paper explores student disengagement and proposes an approach, engagement tracing, for detecting whether a student is engaged in answering questions. This model is based on item response theory, and uses as input the difficulty of the question, how long the student took to respond, and whether the response was correct. From these data, the model determines the probability a student was actively engaged in trying to answer the question. The model has a reliability of 0.95, and its estimate of student engagement correlates at 0.25 with student gains on external tests. We demonstrate that simultaneously modeling student proficiency in the domain enables us to better model student engagement. Our model is sensitive enough to detect variations in student engagement within a single tutoring session. The novel aspect of this work is that it requires only data normally collected by a computer tutor, and the affective model is statistically validated against student performance on an external measure.
A well-known challenge of adaptive educational systems is the need to develop intelligent content, which is very time and expertise consuming. In traditional approaches a teacher is kept at a distance from intelligent authoring. This paper advocates the involvement of teachers in creating intelligent content. We are presenting an approach to the development of intelligent content as well as an authoring tool for teachers that support our approach. This approach has two main stages: elicitation of concepts from content elements and the identification of a prerequisite/outcome structure for the course. The resulting sequence of adaptive activities reflects the author's view of the course's organization. The developed tool facilitates concept elicitation in two ways: it provides an author with an automatic indexing component and also allows her/him to edit the index using the domain ontology as an authoring map.
Open learner models to facilitate reflection are becoming more common in adaptive learning environments. There are a variety of approaches to presenting the learner model to the student, and for the student to interact with their open learner model, as the requirements for an open learner model will vary depending on the aims of the system. In this paper we extend existing approaches yet further, presenting three environments that offer: (i) haptic feedback on learner model data; (ii) a handheld open learner model to support collaboration amongst mobile learners; (iii) an approach which allows students to open their model to selected or to all peers and instructors, in anonymous or named form.
We have developed a method to identify when a student essay is off-topic, i.e. the essay does not respond to the test question topic. This task is motivated by a real-world problem: detecting when students using a commercial essay evaluation system, CriterionSM, enter off-topic essays. Sometimes this is done in bad faith to trick the system; other times it is inadvertent, and the student has cut-and-pasted the wrong selection into the system. All previous methods that perform this task require 200-300 human scored essays for training purposes. However, there are situations in which no essays are available for training, such as when a user (teacher) wants to spontaneously write a new topic for her students. For these kinds of cases, we need a system that works reliably without training data. This paper describes an algorithm that detects when a student's essay is off-topic without requiring a set of topic-specific essays for training. The system also distinguishes between two different kinds of off-topic writing. The results of our experiment indicate that the performance of this new system is comparable to the previous system that does require topic-specific essays for training, and conflates different types of off-topic writing.
In this work we present a thread-based approach for analyzing synchronous collaborative math problem solving activities. Thread information is shown to be an important resource for analyzing collaborative activities, especially for conducting sequential analysis of interaction among participants of a small group. We propose a computational model based on thread information which allows us to identify patterns of interaction and their sequential organization in computer-supported collaborative environments. This approach enables us to understand important features of collaborative math problem solving in a chat environment and to envisage several useful implications for educational and design purposes.
This paper describes current work directed at dealing with students' learning impasses that can arise when they are unable to make further learning progress while interacting in a 3D virtual world. This kind of situation may occur when group members do not possess the requisite knowledge needed to bootstrap themselves out of their predicament or when all group members mistakenly believe that their incorrect conceptual understanding of a science phenomenon is correct. The work reported here takes place in C–VISions, a socialized collaborative learning environment. To deal with such learning impasses, we have developed multiple embodied pedagogical agents and introduced them into the C–VISions environment. The agents are used to trigger experientially grounded cognitive dissonance between students and thereby to induce conceptual conflict that requires resolution. We describe the design and implementation of our agents which take on different functional roles and are programmed to aid students in the conflict resolution process. A description of multi agent-user interaction is provided to demonstrate how the agents enact their roles when students encounter a learning impasse.
This paper reports a pilot study of how to utilize simulated animal companions to encourage students to pay more effort in their study in the classroom environment. A class of students is divided into several teams. Every student keeps her own individual animal companion, called My-Pet, which keeps a simple performance record of its master for self-reflection. Also, every team has a team animal companion, called Our-Pet, kept by all teammates. Our-Pet has a collective performance record formed by all team members' performance records. The design of Our-Pet intends to help a team set a team goal through a competitive game among Our-Pets, and promotes positive and helpful interactions among teammates. A preliminary experiment is conducted in a fifth-grade class with 31 students in an elementary school, and the experimental results show that there are both cognitive and affective gains.
Physical manipulatives have been applied in traditional education for a long time. This paper proposes that by embedding computing power in manipulatives computers can monitor students' physical manipulations to support learning. This paper also describes the design of a digital desk prototype, called ArithmeticDesk to illustrate the vision of computer embedded manipulatives and takes learning fractions as an example. The study is an attempt to accommodate physical and virtual manipulations, as well as to eliminate the gap between traditional and computer-based learning activities. More experiments and studies will be conducted in the future.