Unlocking Personalized Education with Teachable AI

Table of Contents

1. Introduction
2. Fraction Arithmetic Tutor Building System
3. Val the Verbal Apprentice Learner
4. How Can AI Tutors be Tailor-Made for Students?
5. Intelligent Agents Utilizing Hierarchical Task Network Planners
6. Interactions with the Teacher
7. Construction of Hierarchical Plans
8. Personalization at the Classroom Level
9. Val: A Combination of HTNs and Language Models
10. Natural Verbal Instruction of AI Agents
11. General forms of Tasks and Their Applications
12. Conclusion

📚 Introduction

In this article, we will explore the cutting-edge advancements in AI-powered tutors developed by the Teachable AI Lab at Georgia Tech in collaboration with AI.allo. We will delve into two fascinating projects: the Fraction Arithmetic Tutor Building System and Val the Verbal Apprentice Learner. These innovations are a result of the continuous investigation into a fundamental question: How can we empower teachers to customize AI Tutor systems for their students? We will discuss the user-focused evaluations already planned for both systems and the endless possibilities these technologies offer.

📚 Fraction Arithmetic Tutor Building System

The Fraction Arithmetic Tutor Building System showcases the Teachable AI Lab's commitment to creating tailor-made tutoring interfaces. Teachers can effortlessly create a set of problems and initialize them to the AI agent. Through a series of interactions with the teacher, the agent learns to complete the problem. When faced with unfamiliar steps, the agent asks for clarifications or demonstrations. Underneath its functioning, the agent uses hierarchical task network planners to construct a knowledge representation of the tasks, enabling it to solve problems just as the teacher has shown it. This technology empowers teachers to teach the AI agent while having complete control over the teaching style and methodology at a classroom level.

📚 Val the Verbal Apprentice Learner

Val the Verbal Apprentice Learner takes the concept of hierarchical task network planners and incorporates modern language model technology. This innovative approach enables natural verbal instruction of an AI agent. In a video Game environment called Overcooked AI, Val learns how to cook soup through verbal interactions with the teacher. The teacher can break down the steps using natural language, and Val remembers the general forms of the tasks. The technology leverages hierarchical task networks (HTNs) to plan and reason effectively, surpassing the limitations of neural AI systems. HTNs enable Val to interpret new situations, substituting ingredients and adapting to different tasks without requiring explicit instructions. This development paves the way for teaching AI systems to perform various tasks, transcending the boundaries of video game environments.

📚 How Can AI Tutors be Tailor-Made for Students?

The primary focus of the Teachable AI Lab is to enable teachers to tailor-make AI tutor systems for their students. By incorporating user-focused evaluations in their development process, the lab ensures that the technology aligns with the unique needs of the students. Personalization at the classroom level allows teachers to impart their knowledge and teaching style into the AI system. This not only empowers teachers but also enhances the learning experience of the students. With AI tutors that can be personalized, students can receive instruction in a way that suits their individual learning preferences and abilities.

📚 Intelligent Agents Utilizing Hierarchical Task Network Planners

The use of hierarchical task network (HTN) planners forms the backbone of the intelligent agents developed by the Teachable AI Lab. These agents interact with teachers through various modalities, such as text, dialogue, and demonstrations. The agent constructs a hierarchical plan that contains tasks specified by the teacher and methods created by the agent based on teacher demonstrations. This dynamic process allows the agent to learn and adapt to the problem-solving requirements. Through the selection of appropriate methods, the agent can complete all tasks under a specific method. This approach ensures that the technology effectively incorporates the teacher's knowledge and teaching methodologies, resulting in an AI tutor that can successfully solve complex tasks.

📚 Interactions with the Teacher

The interactions between the intelligent agents and the teacher play a crucial role in the learning process. The agents engage with teachers through multiple modalities, including text, dialogue, and demonstrations. By incorporating a combination of these interaction modes, the agents facilitate effective knowledge transfer from the teacher to the AI system. The agents ask clarifying questions and Seek demonstrations when faced with unfamiliar steps. The teacher's responses enable the agents to construct a comprehensive hierarchical plan, ensuring a deep understanding of the problem-solving process. These interactions pave the way for a collaborative and adaptive learning environment, where the AI system assists the teacher in imparting knowledge effectively.

📚 Construction of Hierarchical Plans

The construction of hierarchical plans lies at the core of the intelligent agents' problem-solving abilities. Through a series of interactions, the agents learn from the teacher and construct a knowledge representation of the problem-solving tasks. The hierarchical plans consist of tasks specified by the teacher and methods created by the agent based on teacher demonstrations. This combination of the teacher's expertise and the agent's adaptability allows the AI system to effectively navigate complex problem-solving scenarios. As the agents construct the hierarchical plans, they select appropriate methods for task completion, ensuring efficient and accurate solutions.

📚 Personalization at the Classroom Level

One of the key advantages of the AI tutor systems developed by the Teachable AI Lab is the ability to personalize the learning experience at the classroom level. Teachers can impart their own knowledge and teaching style into the AI system, creating a model that aligns with their unique approach to instruction. This level of personalization enhances the effectiveness of the AI tutor in catering to the diverse needs of students. By allowing teachers to teach in their own style, the technology promotes a customized and engaging learning environment, fostering better student outcomes.

📚 Val: A Combination of HTNs and Language Models

Val, the verbal apprentice learner, combines the power of hierarchical task network planners with modern language model technology. By harnessing this synergy, Val's natural language understanding capabilities enable verbal interactions with the AI agent. The innovative approach allows the teacher to explain tasks using natural language, while Val comprehends the instructions with the assistance of chat GPT technology. The hierarchical task network (HTN) structures capture the learned tasks and facilitate reliable planning and reasoning. This combination of language understanding and interpretable plans empowers Val to adapt to new situations seamlessly, demonstrating incredible flexibility in its problem-solving capabilities.

📚 Natural Verbal Instruction of AI Agents

Val, the verbal apprentice learner, represents a significant breakthrough in AI tutoring technology through its ability to receive natural verbal instructions. In a video game environment, the teacher can verbally instruct Val to perform tasks. The technology leverages the power of chat GPT to comprehend natural language inputs and facilitate unconstrained interaction. As Val learns from the verbal instructions, it remembers the general forms of the tasks, enabling it to apply the acquired knowledge to new situations. This natural verbal instruction capability paves the way for a wide range of applications, extending beyond video games, and revolutionizing the field of AI tutoring.

📚 General Forms of Tasks and Their Applications

The hierarchical task network (HTN) structures learned by Val offer numerous benefits in terms of generalization and versatility. As demonstrated in the video game environment, Val can handle tasks beyond those explicitly taught, thanks to the comprehensive task structures. The HTNs allow the substitution of ingredients and steps, enabling Val to understand new recipes without requiring specific instructions. This adaptability makes Val an ideal candidate for a wide variety of settings, where the ability to reason and plan effectively is crucial. The combination of General Task structures and language understanding capabilities positions Val as a powerful tool for teaching humans various tasks in a personalized and dynamic manner.

📚 Conclusion

The advancements in AI-powered tutors developed by the Teachable AI Lab at Georgia Tech, such as the Fraction Arithmetic Tutor Building System and Val the Verbal Apprentice Learner, highlight the potential of tailoring AI tutor systems for students. The ability to personalize the learning experience at the classroom level empowers teachers and enhances student outcomes. The integration of hierarchical task network planners and modern language models enables natural verbal instruction of AI agents, opening up possibilities for versatile tutor building in various domains. As these technologies continue to evolve, the future of AI tutoring holds immense promise for transforming education and individualized learning experiences.

Highlights

1. The Teachable AI Lab at Georgia Tech has developed AI-powered tutors for personalized education.
2. The Fraction Arithmetic Tutor Building System enables teachers to create customized problem sets.
3. Val the Verbal Apprentice Learner combines hierarchical task network planners with language models for natural verbal instruction.
4. Teachers interact with AI agents via text, dialogue, and demonstrations to impart knowledge.
5. Hierarchical plans are constructed based on teacher specifications and agent-created methods.
6. Personalization at the classroom level allows teachers to tailor the AI tutor to their teaching style.
7. Val's hierarchical task network structures enable generalization and flexibility in problem-solving.
8. Val's language understanding capabilities facilitate unconstrained interaction and comprehension of instructions.
9. General forms of tasks learned by Val allow for efficient adaptation to new situations.
10. The future of AI tutoring holds vast potential for individualized and dynamic learning experiences.

Frequently Asked Questions (FAQ)

Q: What is the purpose of the AI-powered tutors developed by the Teachable AI Lab? A: The purpose of these tutors is to empower teachers by enabling them to tailor AI tutor systems for their students.

Q: How do the tutors interact with teachers? A: The tutors interact with teachers through various modalities, including text, dialogue, and demonstrations.

Q: What is the role of hierarchical task network planners in the intelligent agents? A: Hierarchical task network planners allow the agents to construct knowledge representations of tasks and methods based on teacher demonstrations.

Q: Can the AI tutors be personalized? A: Yes, the AI tutors can be personalized at the classroom level, allowing teachers to impart their own knowledge and teaching style into the system.

Q: What is the significance of Val the Verbal Apprentice Learner? A: Val combines hierarchical task network planners with language models, enabling natural verbal instruction and adaptability to new situations.

Resources:

• Teachable AI Lab
• AI.allo