From Robots to Books: An Introduction to Smart Applications of AI in Education (AIEd)

AI-Powered Teacher Support Systems

Context: Mathematics, geometry, and general education courses. Provides feedback on classroom discourse, assists with doubt clearing, reduces teacher workload, facilitates prerequisite identification, simplifies resource creation from LaTeX documents, and aids in overall curriculum design.

Implementation: Utilizes a variety of AI techniques including SOA-NLP, LSTM, ResNet-101, supervised learning, deep learning for OCR, and knowledge graph-based systems. These are applied to analyze classroom discourse, solve geometry problems, identify prerequisites, convert LaTeX documents, and provide intelligent question answering.

Outcomes: Offers new methods for feedback collection, enables doubt clearing, reduces teacher workload, facilitates resource creation, and supports curriculum design. Improves feedback quality and overall educational experience.

Challenges: None mentioned in the text.

Adaptive Learning Systems

Context: Mathematics tutoring (grades 5-11), and personalized article recommendation. Used to tailor curriculum based on student needs and expertise.

Implementation: Employs smart curriculum based tutoring, deep learning for article classification, and end-to-end pipelines for dependency graph generation. Also includes analysis of student frustration using virtual agents.

Outcomes: Enables adaptive curriculum setting based on student needs and expertise, improving interest and knowledge. Facilitates personalized article recommendations.

Challenges: None mentioned in the text.

Intelligent Tutoring Systems

Context: General education. Providing immediate feedback to students on their work, analyzing student data, and improving the student studying experience.

Implementation: Uses intelligent ecosystems of teachers and students, knowledge graph-based systems for answering questions, and BERT-based language models for math assessments.

Outcomes: Provides immediate feedback to students, analyzes student data, and improves the student studying experience. Used for automatic scoring systems based on maths assessments.

Challenges: None mentioned in the text.

AI-Driven Assessment and Planning

Context: Computer science (setting personalized study plans), and various educational fields for assessments. Focuses on adaptive curriculum setting based on student progress, finding similar exercises, and planning assessments based on students’ knowledge levels.

Implementation: Employs neural transducer models, AI-based assessment systems, BERT-based models for exercise similarity, and language/neural network models for knowledge tracing.

Outcomes: Enables personalized study plans, adaptive curriculum setting, easy assessment setting, and improved assessment planning based on student knowledge. Provides improvised evaluation and testing in diverse coursework.

Challenges: None mentioned in the text.

Handwritten Text Recognition

Context: Conversion of handwritten textual information into digital format or vice-versa.

Implementation: Neural Network based Handwritten Text Recognition model

Outcomes: Can be used for conversion of handwritten textual information into digital format or vice-versa

Challenges: None mentioned in the text.

AI Assessments in Lab Education

Context: For providing improvised evaluation and testing in diverse coursework.

Implementation: Cognitive theory based assessments with feedback ensuring qualitative evaluation

Outcomes: For providing improvised evaluation and testing in diverse coursework.

Challenges: None mentioned in the text.

Lack of Benchmarking Datasets

The need for better benchmarking datasets, similar to ImageNet, to provide a consistent baseline for model improvement.

Proposed Solutions: Curation of consensus benchmark datasets.

Lack of Interpretable AI and Integration of Symbolic/ML Methods

AI systems need to be able to articulate their thought processes in a way that humans can understand. Finding ways to combine ML with symbolic methods to create AI systems that are interpretable and scalable.

Proposed Solutions: Develop AI systems whose thought processes are transparent. Research is needed to figure out how to combine ML and symbolic methods to AI, i.e. near symbolic AI that can ensure better AI for Education systems.

Multimodal Learning Challenges

Many STEM fields require the ability to digest knowledge from various modalities, including texts and visuals. More development in multimodal DL is required.

Proposed Solutions: Much development in multimodal DL is required.

Need for Long-Term Research and Attention to Under-Explored AIED Technologies

Research must be broadened to examine AIED over longer time periods and at the institutional, regional, and national levels. While some AIED technologies have drawn more attention from researchers than others, there is a need to explore AIED technologies that have received less focus over time.

Proposed Solutions: Utilizing cutting-edge technology like text mining, learning analytics, and data visualisations is also required to progress AIED research. More AIED technologies should be covered in future research, especially ones that have gotten less attention over time.

Project Contact: Dr. Jianhua Yang

LLM Model Version: gemini-2.0-flash-lite