Skip to main content Skip to navigation

Experimental Evidence on Negative Impact of Generative AI on Scientific Learning Outcomes

Project Overview

The document examines the role of Generative AI in education, focusing on its influence on academic reading and writing tasks. It finds that students who rely entirely on AI for writing experience a notable decrease in accuracy, with a decline of 25.1%, while those using AI for reading tasks see a 12% drop in accuracy. Conversely, tools like ChatGPT can improve writing quality, especially for summarization tasks, suggesting that AI can be beneficial when used appropriately. The findings emphasize the importance of educators guiding students on the responsible integration of AI in their learning processes to mitigate risks of over-dependence and misinformation. Overall, while Generative AI holds potential for enhancing educational outcomes, careful oversight and instruction are crucial to maximize its benefits and minimize drawbacks.

Key Applications

Generative AI tools (e.g., ChatGPT)

Context: Higher education, targeting college-educated participants engaged in reading and writing tasks.

Implementation: Participants used AI tools for writing summaries and answering comprehension questions after reading academic materials.

Outcomes: Improved writing quality and productivity in AI-assisted groups, but reduced accuracy in comprehension tasks.

Challenges: Over-dependence on AI tools leading to misinformation and diminished learning engagement.

Implementation Barriers

Cognitive Engagement

Students may rely too much on AI for understanding material, bypassing critical engagement.

Proposed Solutions: Educators should guide students on the optimal use of AI tools to enhance comprehension without sacrificing engagement.

Misinformation Risk

AI can produce plausible but inaccurate information, which students may accept uncritically.

Proposed Solutions: Instructors need to provide clear guidelines and encourage critical thinking when using AI-generated content.

Equity in Learning

Not all students benefit equally from AI-assisted learning; prior knowledge and skills influence outcomes.

Proposed Solutions: Customizing AI integration in educational frameworks based on students' backgrounds and providing support for less experienced users.

Project Team

Qirui Ju

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Qirui Ju

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

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