Dear all,

This message is an introduction to the STAR project and its stated aims. We are working on building a new way of using AI to assist, rather than impose on humans. To do this, we are investigating how Large Language Models with multiple parameters which can generate text can be used to assist in a process that affects all of our Level 6 and above students; the process of dissertation supervision.

Many AI tools are currently focused on technology, but since we are all educators, we have opted to include the context of Learning Styles in the process by which our AI will be implemented. Many educators will be familiar with the CAST guidelines as enshrined by the UDL framework for designing learning materials, as linked here. These guidelines indicate how students thrive if the information that is presented to them is written in a way that takes their own latent learning styles into account.

The process by which the STAR research is being carried out is therefore in two phases. The first phase involves selecting a group of volunteer level 6 students, in our case from the ICT field, to participate in an online survey to determine their learning style as per the VARK questionnaire design linked here.

Once these students have filled in the questionnaire, the distribution of learning styles in our example cohort will be determined, and the learning styles in that cohort fully identified. This is the task that has been allocated primarily to the educators in our group, who are designing the online process by which students will provide consent to participate in this study, as well as the actual questionnaire which will be determining the learning style, a project that is proceeding apace and is looking to be completed during this month.

The next phase is AI related. Many are aware of the prevalence of large language models and how they are affecting the way that education is being carried out. This is leading to fundamental changes in education. One of the biggest issues affecting large language models today is defined by the broad term AI alignment. How can we tell whether or not an AI model is ‘hallucinating’ by linking words that may be part of its context window without being meaningfully part of the answer that makes sense? Although several initiatives with respect to large language model alignment have begun, including the OpenAI evals framework linked here, the advantage of an educational institution like MCAST doing this research is that expert evaluators are readily available; the lecturers who are teaching a specific subject.

In this case, the topic chosen is the Research Project, an accredited level 6 unit that is included in the ICT Institute’s IT and Multimedia programme curricula, as it is topic that is most closely allied to the process of dissertation supervision. Currently, research is being carried out in using the AI Large Language Models themselves to generate data for fine tuning, or additions to the context window, that are optimized for different learning styles. Such research can lead to the development of optimized and aligned models that present information based on the learning style which has been identified for the student, given different ways of representation of the content that is known to be correct.

Work is proceeding on both aspects, with a view to generating models that will create the right information. Access to these models through an online chatbot interface will be given to the students participating in the study, who will be requested to use the chatbot as an additional resource in support of their dissertation journey, and in collaboration with their lecturers.

All the students and their dissertation tutors will then be qualitatively interviewed about their experiences, to understand how they integrated the chatbot into their process and whether the addition of an AI chatbot based on this information helped them formulate better ideas in the light of the research process, and presented them with information that was correct, topical and useful.

We believe that this research will provide tangible and useful outcomes that will drive the adoption of AI in education, while ensuring that human supervision and insight is maintained in this context. As a team, we are looking forward to sharing our results with you all!

Warmest regards

Gerard Said Pullicino (PI, STAR project)