Personalized Teaching with AI: Revolutionizing Traditional Teaching Methods

Introduction

The class had 35 students in all. Among them, Xen was a slow learner, and his teacher was in a dilemma. Should she slow down for Xen? But that would cause boredom to others in the class. In her normal pace of teaching, Xen failed to match up. And about Xen, she was sure that once the concepts were clear, he would do well. Figuring out a solution to handle the different learners in her class was rather a tough task. 

And then there was Yash in another class. He had trouble recollecting the concepts. In his case, the teacher tried various methods – reading aloud, story-based learning, and so on. Finally, she asked him to create mind maps. And that worked like magic! With mind maps, he could quickly recollect the concepts. 

Both these cases point towards ‘personalized’ teaching that involves understanding the strength and weaknesses of a particular student. This customization of pedagogy to cater to individualistic needs ensure maximum outcomes. However, this is neither scalable nor, very often, affordable. AI (Artificial Intelligence) has been playing a key role in disrupting and complimenting some of the traditional teaching practices. In this blog, we will explore concepts like ‘Intelligent Tutoring Systems’ and how AI is playing a key role in this field. 

Intelligent Tutoring Systems (ITS) involve capturing various parameters of a learner and then scientifically analyzing and applying AI algorithms to them to provide adaptive and specific learning methodologies for individual learners. ITS could capture simple interactions on the learning platform, non-invasive biometric features, or other characteristics of the learner. Out of the several dimensions of ITS, let us explore a few in this article.  

Knowledge Tracing 

One of the key concepts in AI-empowered ITS is “Knowledge Tracing.” This involves providing several types of content to the learner and capturing the interactions. These interactions along with their past performances can be used to predict their future performances. This can be understood better by looking at the following dataset from the ‘RIID Answer Correctness Prediction’ competition hosted by Kaggle. 

The data contains learner interactions over a period. This includes –  

  • User Id 
  • The timestamp of the interaction,  
  • The type of interaction (content_type_id),  
  • The multiple-choice question that was presented to the learner (or) a lecture that the learner is viewing (content-id),   
  • The answer provided by the learner (user-answer),  
  • Boolean that implies has the learner answered it correctly or not (answered_correctly), 
  • Other details like ‘time elapsed since the last question, if the previous question had an explanation, to what group did the question belong (task_container_id), and so on

Figure 1: RIID Dataset 

Besides this, the details (tags, answers, type) of the questions and lectures are also made available. The goal was to create algorithms for “Knowledge Tracing,” modeling student knowledge over a period of time. Once this is done, the model created would be able to predict if the subsequent questions will be answered correctly or not.  

This is typical “time-series” data. There is a set of inputs that spans across time. Using this, we can predict the same series in the future. “Sequence Models” are AI models that can be applied to time-series data to predict the next sequence. Several AI-based algorithms can be used to solve this type of problem. However, one of the winning solutions was based on “Transformer” technology. Note that this is the same technology that is powering the famous ChatGPT tool. “Transformer” is a concept that was first introduced in a path-breaking paper “Attention Is All You Need” by Ashish Vaswani and Team. “Attention” refers to a mechanism that allows the model to focus on specific parts of the input sequence. This can be useful in cases where the model needs to understand the relationships between various parts of the input sequence to make a prediction. As the user interacts increasingly with the KT system, The AI system will be able to capture the “pattern” and thus make predictions for future interactions.  

Curriculum Alignment 

We live in an era that can be best described as a one with “Content Explosion.” This is especially true in the field of education. The process of ‘curriculum alignment’ or discovering the best resources to match the curriculum is currently manual. As new content is created, additional efforts are needed to find the topics that they align with. Organizations like UNHCR have been exploring options for the use of machine learning to support educators and students around the world in accessing relevant learning materials. 

Recently the company “Learning Equality” jointly with “Learning Agency Lab” hosted a competition on Kaggle. The dataset contained two artifacts – Content and Topic. Content could be video, document, exercise, html5, and so on. The content has a title and description that best describes it. For example, “Pythagoras’ Theorem,” “Ordinary Least Square” and so on. The goal of the competition was – if a Topic, for example, “Trigonometry” is provided, the content (in the order of relevancy) should be identified and correlated with the topic. The topics internally could be part of a ‘hierarchy’ and could have the appropriate topic parent. For example, the topic ‘Trigonometry’ could have the parent Topic as ‘Mathematics’ and so on.  

Problems like these also involve applying AI-based algorithms to read details of the “content” to establish the relationship between different content. The solution involves exploring topics, reading the text (title, description, and so on), tokenizing them, representing them using “embeddings,” establishing the relationships, and identifying the neighborhood and correlations. The competition demonstrated that simple unsupervised algorithms like KNN (K-Nearest Neighbour algorithm) can be used to set the baseline. Advanced ‘Transformer’ based solutions can be used to obtain better scores.  

Student Engagement  

One other aspect of ITS is the ability to detect the “interest” level of the learner and this is one of the toughest tasks to evaluate. The “student engagement detection” technology is in its nascent stage. There have been several studies that show that it is possible to assess learner engagement using AI and Biosensors. This consists of web cameras that capture various postures, movements, facial expressions, eye positions, and so on. It can also use wearables to capture the vitals. These are stored and processed on high-end machines. The data is then used to train AI models to understand the engagement of the students in the classroom. This is used to estimate behavioral and emotional engagement. This also will help in evaluating different teaching methods and approaches. Several AI-based models have shown the ability to predict with good accuracy. However, there have been several privacy concerns with this type of technology.

Conclusion 

Organizations like UNESCO and UNHCR have been actively involved in ensuring quality education for all. To provide scale and quality in education, it is important to adopt and deploy artificial intelligence-based tools and technologies. Technologies like “ChatGPT” has demonstrated the power of Artificial Intelligence and Large Language Models. Augmenting the existing education framework with AI-based tools and technologies with the right regulations and standards in place will definitely be the way forward.  

 

Related Articles

loader
Please wait while your application is being created.
Request Callback