Skip to main content Skip to navigation

An intelligent tutor for planning in large partially observable environments

Project Overview

The document explores the application of generative AI in education through the development of an intelligent tutoring system designed to enhance planning skills in partially observable environments. Utilizing a metareasoning algorithm known as MGPO, this system effectively discovers and teaches optimal planning strategies, thereby improving decision-making abilities among learners. An experimental study conducted with participants demonstrated that those who engaged with the intelligent tutor showed marked improvements in their planning performance compared to those who learned without assistance. The findings highlight the potential of AI-driven educational tools to facilitate better learning outcomes and develop critical cognitive skills in students.

Key Applications

Intelligent Cognitive Tutor for Planning

Context: Educational context for learners planning in partially observable environments

Implementation: The tutor provides feedback based on the planning operations learners choose, adjusting difficulty and offering demonstrations of optimal strategies.

Outcomes: Participants showed significant improvement in planning strategies, achieving higher resource rationality scores compared to control groups.

Challenges: Participants struggled with repeating computations and understanding when to terminate planning.

Implementation Barriers

Technical Barrier

The complexity of modeling real-world partially observable environments accurately.

Proposed Solutions: Integrate robust strategy discovery methods and domain expert knowledge to improve model accuracy.

User Understanding Barrier

Participants had difficulty learning when to repeat computations and how to plan effectively under uncertainty.

Proposed Solutions: Utilize feedback and shaping methods to gradually teach planning strategies.

Project Team

Lovis Heindrich

Researcher

Saksham Consul

Researcher

Falk Lieder

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Lovis Heindrich, Saksham Consul, Falk Lieder

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies