M.S. AAI Capstone Chronicles 2024

EDUCATION ASSISTANCE THROUGH A.I.

4

Background Information

The goal of an A.I. teaching assistant in this context can be broken down into two parts. The first is to provide assistance in a more accessible manner than the professor. This can manifest as a 24/7 chatbot that answers simple questions. This would grant more time for the instructor to answer difficult issues. The second part involves grading assignments. At the time of writing the bot in theory has the ability to do so as a LLM however this was never tested. Combining both aspects is the stated goal of the over-arching project. A simple example of a technology that solves the first issue was a study done where they used an Amazon Echo as a virtual assistant in the classroom to assist both the teacher and students (Parab, 2020). The study concluded that while students trusted the human teacher more, the assistant did help the teacher, allowing the teacher to assist students more effectively. Another example is ChatGPT. The use of this A.I. has become commonplace since the release of GPT-3.5. According to a survey conducted by Nav in 2024, 43% of students admit to having used ChatGPT or similar A.I. tools (Nav, 2024). Unfortunately, Nav also states that two-thirds of teachers have grown distrustful of students due to these tools. Turning to the topic of automated grading mechanisms, this has been a goal for quite some time. In the 1960s, Stanford attempted to create an automatic grader; however, this implementation was prone to quite a few flaws (Forsythe & Wirth, 1965). Since then, there have been many attempts. Blackboard-Anthology(n.d.) and Canvas(n.d.) are both common programs that have the ability to automatically grade certain multiple choice assignments. This is typically done through static criteria and any self written answers still need to be verified by the professor in question. They require well maintained databases of questions and answers and from experience it is sometimes difficult to keep them up to date. Grading code requires significantly more manual effort than when using these two systems in their base form.

77