King's College London

Problem/Motivation 

Receiving timely and appropriate feedback on assignments and supporting the student individual requirements are critical parts of the learning journey. As digital learning platforms become the norm and we witness the advent of hybrid teaching, quizzes consisting of multiple-choice questions have become the preferred method for both formative and summative assessments and for tracking the learning progress. However, current approaches of providing feedback for both multiple choice and parametrized questions suffer from number of limitations. Firstly, the feedback is based on individual questions, therefore has a limited ability to inform; in a typical scenario, the answer to the question is the actual feedback, as it identifies the correct option and, potentially, provides a link to supporting material from the course. Secondly, the feedback does not cater for different learning needs; due to its limited scope, it provides the same output to all students. Finally, the feedback does not support adaptive learning due to its static, memoryless nature, as does not consider the previous learning experience, the progress, or the performance of the student. All these limitations can be addressed through the development of a feedback framework with personalised recommendations based on the previous learning experience of each student. 

 The flexibility and scalability properties of generative language modelling provide a unique solution for creating an adaptive learning system through providing personalised tutoring and feedback to students based on their individual learning needs and progress. The flexibility of such adaptive learning systems allows identification of common student mistakes, for the benefit of the lecturers, and for providing tailored feedback, for the benefit of the students.  

 Aim/Objectives 

The aim of this project is to integrate Generative AI into a learning system in order to provide repeatable, scalable, and adaptive automatically generated feedback which caters for the student performance and learning progress. The primary outcome of the project is an enhanced, inclusive student learning experience through personalized and immediate learning support by using the ability of a GenAI model to respond to user prompts and to generate highly original output. The framework will benefit both lecturers and students, as it will identify and address the challenges faced by individual students and aggregated cohorts throughout their learning journeys, which will subsequently inform the lecturer how to optimise their delivery plan. 

The project will consist of a software platform that collects individual student interaction with the material, performance in assessments, and input relating to queries and confidence in each delivered concept. The platform will use these as input into the AI based engine that will produce both personalised feedback for students and the general progress of the cohort for a more comprehensive performance picture for the lecturer.  

 As part of the investigation, the generative learning model will be trained on a diverse range of texts, including formative assessments (quiz questions), module material, books, forum discussions, and challenging concepts as flagged by the students, to understand user input, generate human-like responses, provide customized recommendations and feedback, and maintain coherent conversations on a wide range of topics. After each formative assessment, the AI based engine will provide personalised feedback using the profile of each student as input, as well as focus on areas where the students lack confidence, and aggregate all this information over time to provide further revision support prior to the summative assessment. The system will also aggregate formative performance across cohorts and inform the lecturer which concepts required further clarification for wider groups. The resulting adaptive system will consist of several feedback loops: 

1. A short-term feedback loop, to guide students for the current material, based on the performance of one assessment. 

1. A long-term feedback loop, to support student revision by highlighting the concepts that need further studying throughout the module. 

1. Short- and long-term feedback loops, to inform the lecturer and optimise their delivery of the material, according to the student performance and needs across multiple cohorts over time. 

The platform will be integrated with the current KEATS online learning platform and evaluated on current modules. On completion, student satisfaction rates in relation to the provided feedback and student achievement should see an increase, while the lecturer effort can be channelled towards specific concepts, reducing their overall workload but delivering a more impactful contribution. Due to its learning mechanism, the proposed approach is subject-agnostic and can therefore be applied to any subject area. 

 All ethical approvals will be collected according to the College rules