A funded research project by MCAST in collaboration with MCST

When ChatGPT was released, all of us were established AI researchers who are also teachers. We had been teaching python and training models for a long time at MCAST, which is a vocational institution, and were quite familiar with the issues that come with trying to train AIs. Some of us had experience with projects related to image processing, while others were more into the low level, electronics aspect, while still taking the machine learning element into account. All of us were experienced programmers and we all thought that this domain was our own knowledge, that had to be imparted gently online or through teaching in class.

My world was upended when I discovered ChatGPT’s potential. The fact that an AI based chatbot could generate plausible answers based on questions that I asked that could make sense gave me the immediate realisation that programming was dead, and that the whole concept of learning was going to be flipped on its head. I remember I was teaching a class about declarative programming at the time, and I started to use chatGPT to generate examples. It kept getting it wrong and at times gave the wrong answers, but it was enough to point me (and my students) in the right direction, and made a big difference to the students as well.

This realization prompted Stephan and myself, in collaboration with Lorna to write up a dissertation proposal that could align with MCAST’s AI strategy. Our intention was to try to build something like ChatGPT that would take the pedagogical fundamentals into account, allow us to moderate the answers that were generated by the AI, and also take another fundamental into account, which is the way that students prefer to have their content presented.

This is the basis of STAR, which is an investigation into the design and preparation of a Large Language Model (LLM) based on lecturer curated data, and the selection of a model based on the preferred learning style of a student (which then affects the way that the information is presented in the context of assisting students to solve a problem). The multidisciplinary team is composed of Mr. Gerard Said Pullicino, Dr. Stephane Role, Mr. Daren Scerri,Dr. Lorna Bonnici West, Mr. Alan Gatt, Mr Geoffrey Farrugia and Ms. Lisa Theuma, all of whom come with different backgrounds and expertise.

It is hoped that our work will help lecturers and students get better at bringing the best out of themselves with the help of AI.