Smart Tutors: Delivering Personalized Learning

As the pandemic has accelerated the trend toward self-directed and lifelong learning, smart tutors have become an important tool for supporting both independent learning and student-instructor communication.

Smart tutor apps use artificial intelligence (AI) to guide learners in understanding a given question or lesson, while also mapping out the entire learning path based on their individual progress and needs. AI analyzes learning patterns to help students become better learners, so they can master the skills they need.

Smart tutor apps are focused on tackling the biggest obstacles to education. Instead of a “pie-in-the-sky” aspiration, a personalized learning experience – available anywhere and anytime – is now a common expectation. Designed for all students, from elementary age to adult, across all subjects, smart tutors enable deep personalization and allow teachers to create a more complex learning environment, augmenting their communication with students. Such apps also enable just-in-time learning and tailored support for those who want to develop the skills and knowledge needed for success in the workplace and in life. 

Smart Tutors Replicate Personalized Human Tutoring

The main idea behind a smart tutor is automation of the human tutor's role and the teaching process. A tutor is a personal teacher that provides one-on-one instruction, which has proven to be more effective than the traditional model of one teacher per classroom. Unfortunately, not everyone can hire a personal teacher, but a smart tutor is more affordable and available 24/7.

Typical human tutoring is effective because it is highly personalized. A tutor presents and explains content, teaches a student to apply knowledge and provides examples. They interact with the student, ask questions and assign tasks to check how the student understands the material. A tutor explains mistakes, analyzes the steps in problem-solving and offers hints to guide students toward the correct solution. If a student is confused, the tutor demonstrates the solution in another way or in more detail, at the student’s individual pace. Depending on how easily the student solves the problem, the tutor can either progress to the next topic or continue to delve into the current one.

The steps and methods of a human tutor are programmed into smart tutors. The app explains the topic, shows how knowledge can be used to solve problems, provides tests based on the student's performance and offers the next learning activity, when appropriate.

How Does a Smart Tutor Work?

The typical smart tutor architecture (Source) 

Conceptually, smart tutors include the following components:

1. Content structure – Created by the subject-matter expert, the content structure provides a visual representation of each course and the relationship between topics. All topics depend on each other and have a hierarchy: without completing one topic, a student cannot start another. Based on the content structure, a smart tutor can build an individualized learning plan for each student.
2. Question bank – Connected to the content structure, the question bank allows for assessment of a learner’s mastery of each topic. Test scores are recorded in the student's learning history and used for subsequent analysis. The question bank includes feedback – hints for next steps in solutioning and recommendations for any topics to repeat based on the student’s answers. Both hints and recommendations can be generated automatically using machine learning (ML) algorithms or coded manually by the content developer.
3. Student model – Personal data (age, grade or mastery level, language); learning goals; current knowledge and competencies; historical data on learning activity; schedule and performance comprise the student model. Based on all these data, the expert model (see below) offers the student the next content item. The smart tutor adapts to the student's interactions dynamically (rather than in a predetermined way) and provides personalized instruction, recommendations and specific exercises for each student throughout the learning process.
4. Expert model – Combining  domain knowledge and AI, the expert model is what makes the smart tutor "smarter." The first part of the model processes test results and analyzes scores. The second part plans personalized learning paths based on the data from the first part. 
   The modern expert model includes a CAT (Computer Adaptive Testing) component based on IRT (Item Response Theory) technology. The iterative algorithm calculates the individual student's ability and corresponding question difficulty. Analyzing the assessment data, the model chooses the next question for the student according to their performance level. Each time the student answers a question, their ability level is recalculated and the difficulty of the next question is adjusted.
   When a learner completes the testing items and demonstrates their knowledge of a topic, the expert model analyzes all associated data: number of attempts, use of hints, learning goals, previous experience with the topic, grade or mastery level and dozens of other factors. Based on this information, the model selects the subsequent content piece to continue the course, or to repeat the current or a dependent topic. Thus, the student moves at a comfortable pace until reaching their learning goal.
   
5. Course manager – Like a conductor, the course manager orchestrates the "relationships" between all other components of the smart tutor. For example, when a student starts a course, the course manager transfers the student’s data and information about the available content items to the expert model. The expert model chooses the content and questions for the learner and returns the reference to the course manager. Finally, the course manager pulls the content item out of the database and sends it to the user interface.
6. User interface (UI) – The student learns and interacts with the system via a user-friendly UI, which might include a search engine, personal account, schedule, links, chats, guides, surveys and performance statistics. 

Benefits of Smart Tutors for Learners and Instructors

Smart tutors not only address learners’ knowledge gaps, for younger students they also equip parents to better support their children’s learning. Further, they optimize instructor labor and address teacher retention and development. While a teacher has limited time for student support, smart tutors are always available. They help to stay in touch with the student 24/7 and provide an uninterrupted learning process, so students get support even outside the classroom. Moreover, some types of tests and grading can also be implemented by smart tutors. That frees teachers from routine tasks, which means they can use their time more effectively and take learning to the next level.

Smart tutors also aid the optimization of learning paths and decision tree development. When AI applications in learning were rather limited, “adaptive learning” meant that a student followed one of the pre-set learning paths. Building decision trees to cover everything we might want to teach would be a prohibitively expensive endeavor. By contrast, modern smart tutors don’t have pre-set learning paths. With AI modeling, the "new adaptive" means tailoring to each student’s performance and optimizing teaching strategies accordingly.

Smart tutors simulate one-to-one human tutoring, provide targeted just-in-time feedback and build personalized learning paths to best meet students’ individual learning needs. The more users learn, the more smart tutors adapt to each learner. The apps help each learner work on their personal challenge areas, knowing what extra help they need to progress. Smart tutor apps can make learning more personalized, flexible, inclusive and engaging, leading to improved learning outcomes. Using smart tutors, students can take charge of their own learning.