AI Tutor Basics
Background information on AI and Education
The concept of using AI in education has been explored for over three decades, with its roots in “Intelligent Computer Aided Instruction” (ICAI) and the subsequent emergence of “Intelligent Tutoring Systems” (ITS) in the research community. This evolution highlights a continuous effort to harness technology for more efficient and personalized learning, aiming to replicate the highly effective one-on-one human tutoring model, which often faces challenges in terms of scalability and cost. (Baillifard et al., 2025; Memarian & Doleck, 2023)
How AI Tutors Work
AI Tutors function as advanced educational tools that simulate human instruction, providing personalized and adaptable learning experiences. At their core, these systems are powered by a combination of artificial intelligence algorithms and machine learning models, which enable them to learn from data and continuously improve their performance. Natural Language Processing (NLP) and Large Language Models (LLMs), such as GPT-3, GPT-4, and ChatGPT, play a crucial role in facilitating human-like conversations and generating relevant educational content, including explanations, examples, and customized questions. Some AI tutors also utilize neural networks to predict a student’s evolving knowledge levels and adapt learning paths accordingly. Incorporating knowledge graphs can further enhance the system’s ability to provide accurate answers. These underlying technologies allow AI tutors to offer personalized learning by adjusting content based on real-time student performance, provide immediate and timely feedback on assignments, generate diverse learning materials, actively engage students through interactive dialogue, and monitor progress to identify knowledge gaps (Baillifard et al., 2025; Kim & Kim, 2020; Maity & Deroy, 2024).
Setting up an AI Tutor
Setting up an effective AI tutor system requires careful planning, design, and ongoing management. The specific setup of your tutor depends on the underlying software you choose, but there are four key parts to the setup process.
- Choosing and preparing content to train the tutor. In most cases, the process begins with thorough content and data preparation, which involves curating content such as research articles and PowerPoints, providing outlines of learning objectives and outcomes, and reviewing past example questions from homework, previous exams, or other assignments.
- Prompt engineering. For LLM-based tutors, meticulous prompt engineering is essential. This includes clearly defining the AI’s role (e.g., “an encouraging tutor”), stating its goal (e.g., “help students understand concepts”), providing step-by-step instructions (e.g., “ask one question at a time, wait for a response”), tailoring explanations to the student’s level (e.g., “a college freshman”), guiding without directly giving answers (e.g., “Socratic methods”), and adding specific constraints (e.g., “limit responses to 150 words”). Prompts often require thorough testing to ensure proper responses and mitigate incorrect information, also known as “hallucinations,” which are further discussed in a later section of this chapter.
- Deciding how students will access the tutor. In terms of infrastructure, using web-based chatbots or integrating AI into existing Learning Management Systems (LMS), such as Brightspace, is the most common approach. Students should have clear instructions on how to access the tutor for support.
- Testing and validation. It is essential to review AI-generated responses for bias and inaccuracies, design assignments to minimize over-reliance on AI, ensure data privacy and security, and emphasize human oversight to effectively combine AI’s computational power with human creativity and judgment. We will continue to discuss each of these issues in more detail in the “Considerations and Concerns” section of this chapter. Ultimately, continuous testing, validation, and research are crucial for refining the AI tutor and validating its effectiveness (Bailey & Warner, 2024; Kestin et al., 2024; Mollick & Mollick, 2024).
