Predicting and Understanding Human Action Decisions: Insights from Large Language Models and Cognitive Instance-Based Learning
Project Overview
The document examines the application of Large Language Models (LLMs) in the educational sector, particularly their role in predicting human decision-making strategies and addressing cognitive biases in sequential tasks. By comparing LLMs to a cognitive instance-based learning (IBL) model, it highlights LLMs' ability to integrate feedback effectively, thereby improving prediction accuracy. This capability not only demonstrates the potential of LLMs to enhance understanding of complex human behavior but also suggests a beneficial synergy between these models and cognitive architectures in educational settings. The findings emphasize the significance of leveraging generative AI to support decision-making processes, optimize learning outcomes, and tailor educational experiences, showcasing its transformative potential in enhancing educational methodologies and strategies. Overall, the document underscores the promising implications of generative AI in fostering better decision-making and learning pathways in education.
Key Applications
LLM-powered decision support systems for predicting human behavior
Context: Educational settings involving decision-making tasks for students or researchers
Implementation: Used open-source LLMs (Mistral 7B and Llama-3 70B) to analyze decision-making in gridworld environments
Outcomes: Mistral-7B outperformed other models in predicting human strategies and behaviors, demonstrating adaptability to feedback.
Challenges: Complex decision environments pose challenges for accurate predictions, particularly when human behavior is highly variable.
Implementation Barriers
Technical Barrier
Limited understanding of how LLMs capture human cognitive biases and decision-making processes.
Proposed Solutions: Integrate cognitive models with LLMs to enhance prediction accuracy and understanding of human-like behaviors.
Data Barrier
LLMs require extensive training data to accurately model human decision-making. Utilize instance-based learning models that can perform well with fewer samples to complement LLMs.
Proposed Solutions: Utilize instance-based learning models that can perform well with fewer samples to complement LLMs.
Project Team
Thuy Ngoc Nguyen
Researcher
Kasturi Jamale
Researcher
Cleotilde Gonzalez
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Thuy Ngoc Nguyen, Kasturi Jamale, Cleotilde Gonzalez
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