British Journal of Educational Technology Vol 30 No 1 1999 79–81 Colloquium Artificial intelligence in educational software: has its time come? Paul Brna Computer Based Learning Unit, Leeds University, UK. Email: paul@cbl.leeds.ac.uk Introduction Can educational software developers learn anything from the artificial intelligence community based on thirty years of research into the use of AI in education? Has AI any role in the burgeoning educational software industry? What added value can it provide? On 12 June 1998, a colloquium was held to discuss the role that AI has in educational software development IEE (1998). A range of organisations were represented including artificial intelligence in education, funding agencies, schools, universities and educational quangos. Organised by Professor Alan Bundy, currently Head of the Division of Informatics at the University of Edinburgh, three key speakers and a range of demonstrations provided the groundwork. The debate that followed was organised around a panel featuring Nick Rushby of PA Consulting Group, John Gordon from the Scottish Council of Educational Technology and chaired by Professor Tim O’Shea, Master of Birkbeck College. This report summarises and briefly evaluates the arguments featured at the event. The fundamental arguments presented can be characterised as: • Time is needed for pulling through AI technologies • AI techniques can be applied satisfactorily but commitment is needed to see the development phase through to commercialisation • The “added value” is already evident, and there are now a number of detailed evaluations that demonstrate this value • As the use of ICT matures so the need for AI techniques will become apparent and will arise from the requirements of the educational software AI in educational software needs commitment Joost Breuker, from the University of Amsterdam, looked at the problems of bringing technology developed in a research context to maturity. He used the Esprit funded EUROHELP project as an example to show how intelligent help systems had had an influence on both the Apple operating system and Microsoft Word. However, there had been significant problems in attracting funding to pull through the techniques to maturity. © British Educational Communications and Technology Agency, 1999. Published by Blackwell Publishers, 108 Cowley Road, Oxford, OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA. 80 British Journal of Educational Technology Vol 30 No 1 1999 The paradox presented was that EUROHELP provided an architecture for developing coaching for “just in time” training—unlike many other attempts to build intelligent help systems. After about 120 person years of effort, there was a possibility that the system would be usable: ICL, one of the partners, intended to use the methodology to develop systems but around the end of the project the internal politics of ICL changed, and these plans were dropped. Long-term investment is still needed to bring such technologies to fruition but the basic ingredients for success are available. The added value is evident Ben du Boulay from the School of Cognitive and Computing Sciences, Sussex University, then demonstrated that there was strong evidence for an added value for the use of AI techniques in educational software. Starting from the educational value of standard computer assisted learning software, he argued that two kinds of added value have been sought for the use of AI: time gains, and achievement scores. Examples were given of systems that had claimed significant improvements in the use of AI. Of particular interest were the few studies that contrast “smart” and “dumb” versions of the same basic software. Shute’s intelligent SMART system produced significantly better learning than the dumb version which itself was very successful. However, there were costs associated with this added value with perhaps the better students receiving more benefit. Educational software needs artificial intelligence techniques John Self, from the Computer Based Learning Unit, Leeds University, provided an argument that many current educational software systems have a growing need for (some) use of AI techniques, and that these are required in order to satisfy the demands of the very people who want to deploy the software. It was argued that, unsurprisingly, few people wanted to “add AI”—rather, they wanted features which would naturally lead to the use of AI. Useful educational systems evolve over time, and sometimes generate requirements for software that adapts to the learning context, to the individual student. The rise of Web-based learning environments and the increased interest in distance learning has again provided the ground for a resurgence of the use of AI techniques. Taking individual learning styles into account, providing interactivity and immediacy of feedback, and designing learning environments that support both educator and student are issues have been core concerns of the AI in education community for many years. This argument was given flesh by an examination of several systems developed at the Computer Based Learning Unit, Leeds University including: CACTUS, a training system © British Educational Communications and Technology Agency, 1999. Artificial intelligence in educational software: has its time come? 81 for the police; MCOE, an environment for learning about ecology; CHALCS, a project to develop support tools for schoolchildren to be tutored out of normal school hours; and CLCV, an adaptive hypermedia system for the management of agricultural resources related to cotton leaf curl virus. The debate During the day, presentations were given of four systems featuring various techniques used in AI research (MR Tutor, ILEX, MENO, REDEEM), and videos supplied by US researchers using AI in current educational software (Anderson’s PAT, Munro and Towne’s DIAG, and work by Johnson, Rickel, Styles and colleagues on VET and STEVE). One major question was raised about how any community of practise organised around any specific set of technologies learns from the past and how these lessons can influence the future. Professor Mike Sharples asked whether each generation have to reinvent not just the systems but the basic techniques? Coupled to this is how to “keep the knowledge alive” through system development and maintenance is one of the major issues. The best model for influencing the future of educational software was discussed. Nick Rushby provided a strong argument for market led software development with a view to utilising AI techniques from the first. He favoured research with two levels of risk— low risk for industry and high risk for the main university involvement. Professor Alan Bundy suggested finding a popular piece of commercial software and “adding value”, but there is a need to ensure that the cost of the added value is low enough. John Gordon needed AI software now, and aimed to turn his programmers into “pedagogic engineers”. In conclusion The need for artificial intelligence in education is more urgent than ever because of complex interfaces and large volumes of information. AI can provide added value but this has to be done with due care for the conditions of the market. There are interesting opportunities to pursue especially in applying techniques developed in the AI in education field to education and training as well as to entertainment and consumer products. Reference IEE (1998) Artificial Intelligence in Educational Software Digest No 98/313. See also http://cbl.leeds.ac.uk/~paul/IEEAIED/ © British Educational Communications and Technology Agency, 1999.