AI-ED Society News ------------------ Contents: 1. Introduction 2. Report on the AI-ED 93 Conference 3. Plan to Attend AI-ED 95 in Washington, DC Introduction ------------ The AI-ED Society News serves as a vehicle for the dissemination of news on AI-ED Society activities and items of interest to AI-ED Society members and others. Readers are encouraged to send items for this news section. Items should be sent preferably by e-mail to JAIED@ comp.lancs.ac.uk or by post to John Self, Department of Computing, Engineering Building, Bailrigg University of Lancaster, LA1 4YR, United Kingdom. We welcome all submissions and hope that readers will actively contribute to this news section for the benefit of all. Report on the AI-ED 93 Conference --------------------------------- AI-ED 93 was held in Edinburgh in August, 1993. This report provides details about the conference. About Edinburgh --------------- The AI-ED 93 conference was held in Edinburgh August 23-27, 1993. Edinburgh is a small city 5 hours north of London by train. It houses three universities: University of Edinburgh, Napier University, and Heriot-Watt University. The conference was held in the conference centre at Heriot-Watt University. The Conference -------------- This is the first AI-ED conference conducted by the newly formed AI-ED society. The conference was significant in that it started the renewal of a series of AI-ED conferences after the earlier series in which 4 were held (the last one being in Amsterdam in 1989). The AI-ED society plans to have these conferences biannually. The society is a part of the Association for the Advancement of Computing in Education, which sponsors the conference. The conference included 3 tutorials, 6 workshops, 3 panels and a number of invited, contributed, and poster papers. There were also demonstrations of ITS systems (many of them related to the corresponding paper/poster presentations). The conference had about 69 papers (including 6 invited talks) and 60 posters. The bulk of the papers were from Europe and North America; however, there were a few papers from Australia and Asia. These papers cover the whole gamut of areas in Artificial Intelligence in Education. The largest category of papers were related to student modelling. Table 1, which is based on the contents of the proceedings, gives a distribution of the 63 contributed papers category-wise. Table 1. Category Wise Distribution of AI-ED 93 Papers Architectures 3 Authoring Tools 4 Case Based Reasoning 3 Dialogue and Explanation 5 Exploratory Environments 4 Formal Reasoning 3 Learning Design Skills 3 Meta/Situational 3 Models of Learning 5 Pedagogical Tools 2 Pedagogical Planning and 3 Instructional Dialogues Problem Solving 5 Representations and Notations 4 Student Modelling (numeric) 3 Student Modelling (symbolic) 9 Troubleshooting 4 The conference was held in three parallel streams with the demonstrations in addition. I will therefore not be able to give a comprehensive report on the conference, but will focus on some of the paper presentations I attended. There will be a personal bias in the selection of papers, which is unavoidable. The report also covers two of the six workshops which were held during the conference and which I attended. The conference presentations are not given in chronological order, but are broadly grouped together according to the topic. Where available, I have given the e-mail addresses of authors to allow for ease of communication with them. Prologue -------- Before we get on to a description of the conference, it would be useful to broadly summarize the conference. Some of the key features of the conference were that: (1) Most of the papers which dealt with implementation of systems were backed up by empirical evaluations. This is in contrast to some of the earlier AI-ED conferences. (2) The number of people who attended the conference was rather large. This seems to indicate a healthy growth in the size of the AI-ED community. (3) Like in earlier conferences, not many systems presented at the conference seemed to have been used in the real-world (See a later section for a short description of a workshop on real world deployment of ITS systems). (4) On the positive side, a number of papers expressed concern about theories of learning. This could perhaps indicate a stronger link to the field of education. Some of the prominent issues at the conference were: * The tractability of student modelling. * Advocation of situated learning. * Psychological validity of programming plans (in presentations related to computer programming). Based on the papers at the conference, it seems that computer programming and mathematics continue to be the major domains currently being looked at by ITS researchers. The organization of the conference was exemplary. The local organizing committee deserve credit for this. Presentations ------------- Kurt VanLehn (University of Pittsburgh, USA, vanlehn+@pitt.edu) gave the keynote address at AI-ED 93. He talked about a system named CASCADE which his group is working on. This is a machine learning program which aims to provide a good simulation of human students' learning. The system works in the domain of high school physics and aims to learn 10 chapters (half a semester of a course). Some of the fairly ambitious goals of the system are to: - gradually shift from a generic weak method of problem solving to abstract planning with deep domain principles, - gradually be able to classify problems according to major characteristics of their solutions, and - become more accurate in estimating the difficulty of problems. The best paper award this year was given to a paper by Joel D. Martin and Kurt VanLehn (University of Pittsburgh, USA, jmartin+@pitt.edu). The paper described a student modelling system named OLAE. This system does modelling in the domain of physics using a Bayesian framework and can provide assessments to the assessor at multiple grain sizes. Computer Programming -------------------- There were a number of papers related to computer programming which were presented during the conference. Martin Mathieson (Napier University, Scotland) talked about a system named SWANN which supports novices learning to debug. He argued that novices have significant problems debugging programs because they are - not familiar with the programming language, - ignorant of the facilities offered by the debugging environment, - inexpert at understanding the structure of their programs, and/or - inexperienced at fixing errors. The SWANN system is based on the Marcel model (proposed by Spohrer as a part of his PhD work at Yale University). SWANN is interesting because there are not many systems which have tried to focus on teaching program debugging to students. SWANN, which is currently in a prototype phase, has knowledge about non-programming plans. It provides a buggy program and possible operations which can be done to correct the program out of which the student needs to select the right one. Simon P. Davies (Nottingham University, UK) talked in his presentation about the relationship between design experience and programming skills. He presented results from two experiments. The first experiment dealt with the roles of cues to program structures in debugging, and the second one dealt with recall of programs by design-experienced (candidates who had attended a course on program design) and non-design-experienced subjects. In the first experiment, subjects (with and without design experience) were given a buggy program with up to two bugs in it. The program given had either no cue, a control cue (indenting the code), or a plan cue (colouring the plan segments). The results seemed to indicate that design experience does not appear to enhance the programmer's overall ability to detect bugs. Design experience, however, does have a significant impact upon a programmer's ability to detect plan-related bugs. In the second experiment, subjects had to recall plan-like and unplan-like programs. It was found that the design-trained group recalled significantly more plan structures than the non-design-trained group. Andrew Bowles (University of Edinburgh, andy@aisb.ed.ac.uk) argued for teaching novices programming techniques instead of programming plans. He mentioned the limitations of programming plans and then advocated programming techniques as an alternative. By programming techniques he meant knowledge that is language dependent but independent of the task. Examples of techniques in Prolog, for instance, are list head, list subgoal, and so forth. He argued that the list of techniques in a language should be small and should cover the constructs novices will encounter during initial stages of programming. Further, these techniques should be generalizable to cover other more advanced constructs later on. He illustrated the idea of techniques by taking examples in Prolog. Teaching Mathematics -------------------- Mathematics was another important domain covered by systems presented at the conference. Gautam Biswas (Vanderbilt University, USA, biswas@vuse.vanderbilt.edu) talked about a video-based environment to help teach problem solving and mathematical skills. He emphasized the importance of teaching these skills in the context of real world problems while at the same time making the learning fun. The system he described involved the adventures of Jasper Woodbury (a make-believe character) who needed to plan a trip while keeping in mind some constraints forced by the situation. The tutor provided different levels of interaction with the student. These included - asking the student to consider a particular option, - asking the student a question relevant to the situation, - posing a challenge to the student, - providing a hint to the student, and - providing instruction to the student on how the problem needs to be solved. It was observed that students would often generate sub-optimal plans for the problem at hand but would fail to step back and look at their solutions. They would therefore not be able to detect the sub-optimality. Valerie Shute (Brooks Air Force Base, USA, shute@eis.brooks. af.mil), in her colourful presentation, described a system named Stat Lady which teaches basic probability. She described results of the use of Stat Lady in two experiments. The first experiment compared learning via traditional lectures and learning with Stat Lady. The second experiment compared learning using Stat Lady and learning using a paper-and-pencil workbook (which contained the same material Stat Lady used). Some of the results follow: - Stat lady did not seem to contribute more than the workbook condition in experiment 2. - Stat lady was good for high aptitude students. - Low aptitude students also benefited from Stat lady; however, they seemed to require a combination of lecture and Stat lady. - Stat lady seems to be better for teaching declarative knowledge, and traditional lectures for procedural knowledge. There was an interesting panel on a comparative analysis of learning environment design in the context of mathematics problem solving. The panelists were Baruch Schwarz (Hebrew University, Israel), Mitchell J. Nathan (University of Pittsburgh, USA), and Kenneth R. Koedinger (Carnegie Mellon University, USA). All three of them had been involved in the development of word problem tutors in mathematics and described issues related to the design of these systems. The panel was interesting because the audience got a rare opportunity to compare and contrast different approaches of design for the same domain. The panelists covered the theoretical foundations of their work and also empirical evaluations they had carried out to determine the effectiveness of the systems. Other Presentations ------------------- In this section, I report on presentations in some of the other areas covered at the conference. G. Frosini (Universita di Pisa, Italy, frosinin@iep.unipi.it) described a tool for building expert systems which carry out academic examinations. This tool, which was developed using the Advisor-2 expert system shell, adapts the test to the ability of the student. It automatically determines the level of difficulty of the questions and collects statistics on the questions in an examination. J-Y Djamen (Universite de Montreal, Canada) talked about an interactive planning and learning system named PIF. This system tries to force a learner to reflect and think in advance of trying to solve a problem. The system supports multiple views of a domain and promotes interactive planning. It can detect sub-optimal plans and point out when a wrong step is made by the user. Diana Laurillard (Open University, UK), in her invited talk discussed the issue of the link between learning needs on the one hand and the teaching strategy on the other. She argued that the range of teaching strategies explored by current tutoring were limited and also that the "fill-the-gap" strategy using a buggy model was pedagogically unsound. She said that the teaching strategy should define - the presentational discourse, - the forms of tasks for students to carry out, and - types of feedback to be given. She presented a possible architecture to explore in a search for this link. Hans Spada (University of Frieburg, Germany, spada@psychologie.uni-freiburg.de) gave an invited talk on the role of cognitive modelling in computerised instruction. He said that it is difficult to develop effective student models and that there are few systems which have a successful and full-fledged on-line working diagnostic component. He distinguished between 3 types of cognitive model--idiographic models, prototypical models--and individualized models and argued for the use of prototypical models. Peter Brusilovsky (ICSTI, Russia, plb@plb.icsti.su) talked about the need for educational systems to provide adaptive interfaces. He argued that systems should provide interfaces which can adapt to the knowledge level of the student. He described his system for programming named ITEM/IP which provides for adaptive interfaces. Geoffrey I. Webb (Deakin University, Australia, webb@deakin.edu.au) talked about an approach to student modelling which he called feature-based modelling (FBM). This modelling uses machine learning techniques and task and action features. He said that FBM had been applied in four different domains and has been able to predict 92% of student's solutions in the domain of subtraction. Thierry Chanier (Universite de Clermont, France, chanier@cfdvax. unix-bpclermont.fr) described a system for teaching French interrogatives to English speakers. This system aims to situate teaching in the appropriate context and employs usage and acquisition models of language. Peter Reimann (University of Frieburg, Germany, reimann@psychologie. uni-freiburgh.de) talked about a learning strategy model for worked out examples named AXE. AXE provides a computational model of learning from worked out examples in the domain of mechanics. AXE is also being developed as an on-line model which can help students understand how to more effectively study worked out examples. Two observations he made were: (a) students initially rely heavily on examples but often gain little from them and (b) novices stick to specifics of examples while trying to solve similar problems. Barbara White (University of California at Berkeley, USA), in her invited talk, advocated an approach to science and engineering education that introduces students to scientific domains via progressions of increasing complex causal models. She said that causal models provide a link between higher order abstractions and the real world situations. She also outlined properties required by causal models to make them instructionally effective. Conclusion ---------- AI-ED 93 was definitely a conference worth attending. A combination of a good, well-planned technical program, local organizers who knew what hospitality was all about and participants spanning the whole range of AI and Education areas to interact with made it successful. The next AI-ED conference will be in Washington DC in August, 1995. Acknowledgements: I would like to thank the organizing committee of AI-ED 93 who provided partial support for my attendance at the conference. I would also like to thank my colleague, V. Suresh Kumar for his comments on this report. Reported by: KSR Anjaneyulu National Centre for Software Technology Gulmohar Cross Road No. 9 Juhu, Bombay 400 049, India Email: anji@saathi.ncst.ernet.in Plan to Attend AI-ED 95 in Washington, DC ----------------------------------------- Proposal Deadline: Jan. 6th AI-ED 95--7th World Conference on Artificial Intelligence in Education will be held in Washington, D.C., August 16-19th. The deadline for proposal submissions is January 6th. The technical program will focus on research activities linking Artificial Intelligence theories and techniques with Educational theory and practice. Chris Dede, George Mason University, will serve as the Local Organizing Committee Chair. See the Call for Paper section of the AACE Information Server main menu for further details. +------+ | AACE | +------+ .