Developing reasoning capabilities for intelligent agents that facilitate adaptive learning

适应性学习系统（AL系统）能够通过个性化的教育课程和内容，以及根据学习者个人的特点提供补救或辅导，使其行为适应学习者的需要。为了适应其行为，人工智能系统必须考虑学习者之间在学习目标、知识、经验、学习风格和学习偏好方面的差异。在基于智能体架构的人工智能系统中，智能体与每个学习者或特定任务相关联，以模拟执行教学任务的课程讲师或导师。由于其分布式特性和任务复杂性，基于智能体的人工智能系统被构造为一个多智能体系统。智能体在开放环境中合理适应的能力取决于是否拥有事实性知识，如课程知识、学习者知识、教育资源知识，最重要的是是否拥有推理能力。推理能力包括“理解”学习者特征和知识领域；规划和优化学习计划和学习活动的能力；更新学习模式的能力；一组特定任务代理的干预协调；以及组成联盟进行协作学习等。然而，现有的学习系统有其特定领域的推理机制，并且推理机制的特定于领域的性质限制了系统本身的可重用性。本文提出了一种新的方法来形式化人工智能系统中智能体的推理模型和机制。在短期内，本研究将侧重于设计自适应课程规划、自适应测试和自适应联盟形成的算法，并将不同的模型和机制应用于不同的任务，以确定每种情况下哪种选择是最合适的。从长远来看，这项研究将尝试开发一个完全开放的、可扩展的、基于代理的学习环境。本研究所开发的技术将促进现代分布式学习的进一步发展。此外，透过此计划所训练的学生，将成为以智能体为基础的智能系统的先驱工程师或研究人员。

An adaptive learning system (AL system) is able to adapt its behavior to the learner's needs by personalizing educational curricula and contents, and by providing remedial or tutorial tailored to the properties of the individual learner. To adapt its behavior, an AL system must consider the differences among learners in terms of learning objectives, knowledge, experience, learning styles, and learning preferences. In an AL system with an agent-based architecture, an intelligent agent is associated with each learner or a specific task to simulate a course instructor or tutor performing pedagogical tasks. Due to its distributed nature and task complexity, an agent-based AL system is structured as a multi-agent system. The abilities of the agents to adapt rationally in an open environment hinge on possessing factual knowledge, such as knowledge of curricula, learners, educational resources, and, most importantly, possessing reasoning capabilities. The reasoning capabilities include "understanding" learner characteristics and knowledge domains; the ability to plan and optimize study plans and learning activities; the ability to update learning models; coordination of the intervention of a set of task-specific agents; and coalition formation for collaborative learning, etc. Existing learning systems, however, have their reasoning mechanisms hardwired for specific domains, and the domain-specific nature of the reasoning mechanisms restricts the reusability of the systems themselves. This proposal highlights a new methodology for formalizing the reasoning models and mechanisms for the agents of AL systems. In the short-term, this research will focus on designing algorithms for adaptive course planning, adaptive testing, and adaptive coalition formation for collaborative learning, and applying different models and mechanisms to different tasks to determine which option is the most appropriate in each instance. In the long-term, this research will attempt to develop a fully open and scalable, agent-based learning environment. The technology to be developed in the research will facilitate further R&D in modern distributed learning. Also, the students to be trained through this project will become pioneer engineers or researchers in agent-based intelligent systems.