Supporting Learners in Open Ended and Ill-Defined Domains

点击翻译按钮获取中文摘要

My research is in “artificial intelligence in education” (AIED), an area of computer science that explores how to create learning environments that are deeply supportive of learners. AIED draws on (and contributes to) methodologies from artificial intelligence, of course, but also human-computer interaction, data mining, recommender systems, multi-agent systems, the semantic web, and cognitive science. One key AIED issue is personalization: how the learning environment can adapt to individual differences among learners and tailor its interactions to maximize each learner’s achievements. Personalization is the central concern of another area of computer science: “user modeling, adaptation, and personalization” (UMAP). I contribute to both AIED and UMAP research.One of the issues my graduate students and I have been exploring is simulation of learners and learning environments. We have created simulations to explore various hypotheses using the ecological approach, a novel learning environment architecture (developed in my earlier research) that, over time, can adapt its actions based on patterns discovered in learner behaviour. We have shown that simulations can provide insight into a learning environment without the need for experimentation with actual learners. Simulation could thus lead to reductions in the time and cost of building learning environments.Historically, AIED has been focused on domains like algebra, where there is a well-defined body of knowledge. In contrast, my interests have been drawn to ill-defined domains, where new techniques are needed. One such domain is professional ethics, with two projects. A serious game has been built where software engineering students take the role of a project manager and make decisions (with an ethical dimension) and face consequences, thus illuminating subtle trade-offs they will encounter in their careers. In another project we have been developing a system to support learners in analyzing ethics case studies through a novel visualization (an “open learner model”) that shows them the breadth and depth of their analyses, and how these compare to other learners’ analyses. This stimulates reflection, a metacognitive skill necessary in many ill-defined domains.In another ill-defined domain, language, a system has been built to help people learn Russian pronunciation, with the aid of an innovative “historic learner model” that allows them to visualize their progress over time. Another system we have been developing, in reading comprehension, data mines learners’ behaviour as they read documents and answer questions. Patterns have been found that allow accurate predictions of how well they will answer these questions. These patterns don’t emerge, however, without first classifying the questions as to the depth of knowledge needed to answer them (their “Bloom level”), a novel idea in educational data mining.Going forward we are undertaking new iterations of our work on simulation, ethics, and reading comprehension. We also wish to move further into ill-defined domains to look at lifelong learning. Lifelong learning is a special challenge because learners must be tracked over years, and learning must be supported both proactively and reactively as part of people’s busy lives. Lifelong learning is an emerging area of AIED and UMAP, and our efforts here should pay off both in research impact and real world impact given the need for constant upgrading of personal and professional skills in the information age. The techniques we have been exploring, such as the ecological approach and educational data mining, are helpful in tackling the challenges of supporting lifelong learning, and simulation is crucial to explore implications and test new ideas and techniques.