Skip to main content Skip to navigation

Elucidating STEM Concepts through Generative AI: A Multi-modal Exploration of Analogical Reasoning

Project Overview

This document explores the application of generative AI, specifically large language models, to revolutionize STEM education. The research introduces a multimodal system that simplifies complex concepts by transforming them into understandable metaphors, visualized as storyboards, and then animated videos. The system leverages generative AI to create analogies and visual representations, aiming to enhance student comprehension and engagement with STEM subjects. The study employs a randomized A/B/C test to evaluate the effectiveness of this AI-driven approach in improving learning outcomes.

Key Applications

AI-powered Visual Analogy Generation

Context: STEM education, specifically mathematics, physics, and programming concepts, for learners of all levels. Transforming analogies into engaging visual storyboards.

Implementation: Employs GPT-4 (or similar LLMs) to generate analogies for STEM concepts. These analogies are then transformed into visual storyboards and animated videos using a text-to-image generative model.

Outcomes: Aims to enhance learners’ understanding of STEM concepts, improve their learning engagement, and improve comprehension of complex STEM algorithms.

Challenges: Generating effective visual representations of complex concepts, ensuring the generated content accurately and comprehensively represents the concepts, and the generative model frequently falling short in representing each part comprehensively.

Implementation Barriers

Technical Limitations

The generative model struggles to accurately and comprehensively represent complex STEM concepts in visual form, including difficulties representing multiple components of a concept within a single image. There are also challenges in creating dynamic visual analogies, specifically in generating suitable transitions or motions to articulate the analogy dynamically.

Proposed Solutions: Further refinement of the prompts used to guide the image generation process and exploring different generative models. Further development of the video generation process to include better animation and transitions.

Project Team

Chen Cao

Researcher

Zijian Ding

Researcher

Gyeong-Geon Lee

Researcher

Jiajun Jiao

Researcher

Jionghao Lin

Researcher

Xiaoming Zhai

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Chen Cao, Zijian Ding, Gyeong-Geon Lee, Jiajun Jiao, Jionghao Lin, Xiaoming Zhai

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gemini-2.0-flash-lite