Skip to main content Skip to navigation

Can generative AI and ChatGPT outperform humans on cognitive-demanding problem-solving tasks in science?

Project Overview

The document explores the role of generative AI, specifically ChatGPT and GPT-4, in enhancing science education through their application in cognitive-demanding problem-solving tasks. Findings reveal that these AI models significantly outperform the average student on NAEP science assessments across various grade levels (4, 8, and 12). Unlike students, whose performance declines with increasing cognitive demands, the AI remains consistent, highlighting its capability to handle complex tasks effectively. This discrepancy suggests a critical need for educators to reassess educational objectives, shifting the focus from merely high-intensity cognitive tasks to fostering essential critical and creative skills in students. Overall, the study underscores the potential of generative AI tools to not only assist in educational assessments but also to inform a transformative approach to teaching and learning that embraces the strengths of AI while addressing the evolving needs of students in the educational landscape.

Key Applications

ChatGPT and GPT-4

Context: NAEP science assessments for K-12 students

Implementation: AI models were tested against 54 NAEP science assessment tasks, and their performance compared to average student scores.

Outcomes: ChatGPT and GPT-4 outperformed most students, demonstrating high accuracy and consistency in responses across various cognitive demands.

Challenges: AI models were not sensitive to the increased cognitive demands of tasks, which may not reflect the necessary educational outcomes for developing human cognitive skills.

Implementation Barriers

Technical barrier

GAI models like ChatGPT and GPT-4 are not sensitive to cognitive demands, which may lead to misalignment with educational objectives.

Proposed Solutions: Educational assessments should shift focus from cognitive intensity to creativity and application of knowledge.

Implementation barrier

Educators may lack preparedness to integrate GAI into educational settings effectively.

Proposed Solutions: Professional development opportunities should be provided to equip teachers with skills to guide students in the ethical use of GAI.

Project Team

Xiaoming Zhai

Researcher

Matthew Nyaaba

Researcher

Wenchao Ma

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Xiaoming Zhai, Matthew Nyaaba, Wenchao Ma

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