Exploring adjustable autonomy in online tutoring systems

Learning and teaching have been influenced greatly by the rapid development of technology. For instance, through the use of soft computing techniques, it would be possible to create an artificially intelligent autonomous tutor agent, which can ease the burden on teachers and enhance learning outcomes through its more personalised interaction with students. Providing students with automated guidance, such as directing students through the most appropriate content sequence is one aim of online tutoring systems. However, in most of the available tutoring systems, users neither have the ability to adjust the tutor agent’s autonomy level nor fully control the rules applied by the tutor agent. Thus, this thesis has sought to overcome these shortcomings by proposing a system called the ‘Adaptive Course Sequencing Approach’ (ACSA) which enables students to adjust the autonomy level of the tutor agent and gives teachers the ability to directly communicate with the tutor agent to create the sequencing rules and alter them at any time during the learning experience. This is achieved with fuzzy logic, which has the capability of producing human-readable sequencing rules as well as managing the uncertainty of measuring some students’ levels of knowledge. We hypothesise that by equipping intelligent educational environments with adjustable autonomy mechanisms, the students’ learning outcomes will be enhanced. This research was divided into seven phases and involved a large number of participants (1725 in total) to assess the need for adjustable autonomy mechanisms in online tutoring systems and to explore the way of providing these mechanisms in ACSA, thereby demonstrating the hypothesis by two empirical experiments. The results showed that applying adjustable autonomy mechanisms significantly improved the students’ learning outcomes and that the students who adjusted the autonomy level more than once performed slightly better than those who adjusted it once only. In addition, applying the collaborative-driven agent method, which relies on machine learning to generate and optimise the sequencing rules, led to improving the students’ learning outcomes and highly satisfying the teachers.