"Writing is Thinking"—Do Students Who Rely on ChatGPT Lose Out on Learning?

"Writing is Thinking"—Do Students Who Rely on ChatGPT Lose Out on Learning?

2025年07月02日 15:30

Table of Contents

  1. Introduction

  2. Overview and Background of MIT Research

  3. Interpretation of Key Results

  4. Limitations and Critical Perspectives of the Research

  5. Japan's Generative AI Guidelines and Trends in Educational Settings

  6. Voices of Teachers and Students

  7. "Writing is Thinking" from the Perspective of Learning Science

  8. Overseas Policies and Platform Responses

  9. Reconstructing Learning Design—Cyclical Model

  10. Academic Integrity and Evaluation Reform

  11. Conclusion and Recommendations



1. Introduction

The thesis "Writing is Thinking" indicates a unique human learning process of organizing and internalizing concepts through writing. However, since the release of ChatGPT at the end of 2022, an environment where students can **"write without thinking"** has permeated, causing universities to face confusion over its usage. Associate Professor Leitzinger from the University of Illinois at Chicago revealed that "half of the 180 students used ChatGPT inappropriately," warning of the potential for AI dependence to erode critical thinking.ibtimes.com.au




2. Overview and Background of MIT Research

2.1 Experimental Design

Nataliya Kosmyna and colleagues divided 54 adult learners aged 18-39 into three groups: ① ChatGPT group, ② Search engine group, ③ Brain only group. Each group wrote three 20-minute essays, measuring network connectivity via EEG, with two teachers conducting blind evaluations. Additionally, conditions were switched for the fourth session to examine the short-term and cumulative effects of AI usage. The research was published as a preprint in June 2025.media.mit.edu



2.2 Research Aim

The aim was to quantify whether "tool dependence excessively reduces cognitive load" and "does not hinder learning transfer (memory and creativity)" across neural, behavioral, and linguistic layers.




3. Interpretation of Key Results

  1. Essay Evaluation: The ChatGPT group received the lowest scores in creativity, insight, and originality.

  2. Brain Activity: The same group showed the weakest frontal-posterior connectivity, with diminished interaction in attention and memory areas.

  3. Recollection Memory: Only 19% of the ChatGPT group could immediately recall what they wrote, compared to about 90% in other groups.

  4. Behavioral Observation: The copy-and-paste rate surged after the third session, and typing time halved.

  5. Switch Session: Even participants who switched from brain only to ChatGPT showed decreased connectivity, suggesting a risk of short-term dependence.media.mit.edu




4. Limitations and Critical Perspectives of the Research

  • Sample Size: The statistical power is limited with only 54 participants.

  • Task Monotony: The 20-minute essay differs from actual semester assignments.

  • Prompt Control: The quality of instructions was not standardized, possibly causing variations in generation accuracy.

  • Pre-Peer Review: As it is unpublished, the methods have not been sufficiently validated.


    Dr. Juavinett from UC San Diego points out that "it is premature to discuss neural impacts," criticizing sensational media reports.ibtimes.com.au




5. Japan's Generative AI Guidelines and Trends in Educational Settings

In December 2024, the Ministry of Education, Culture, Sports, Science and Technology published the **"Generative AI Utilization Guidelines Ver.2.0."**


  1. Mandatory Submission of Prompts and Outputs

  2. Clear Indication of AI-Generated Sections

  3. Strict Citation Rules

  4. Recommendation for Instruction to Visualize Thought Processes
    were formalized. The guidelines emphasize alignment with "active, interactive, and deep learning," with pilot schools advancing demonstrations.mext.go.jp



Private high schools have introduced a phased assignment of "Summarize with ChatGPT → Students critically re-edit → Submit source verification report." Universities are increasingly adopting new rubrics with AI usage sections, with **"AI as an aid, evaluation as a process"** becoming a catchphrase.




6. Voices of Teachers and Students

  • Proponents (First-year University Student): "Using it as a hint for English composition speeds up understanding."

  • Concerns (High School Teacher): "Focusing on copying halts vocabulary acquisition. Accompanying guidance from draft to revision is essential."

  • Neutral (Education ICT Specialist): "There are false positives with AI detectors. Metacognitive development should be prioritized over penalties."


    These voices indicate a shift from an era of drawing a "black and white" line on AI usage to an era ofcompeting in utilization literacy.




7. "Writing is Thinking" from the Perspective of Learning Science

The process model by researchers Flower & Hayes defines the essence of writing as a"cycle of idea generation → verbalization → reconstruction." When ChatGPT takes over the initial draft, the "verbalization phase" is skipped, leading to shallow knowledge construction—this is the **"risk of externalizing thinking."**


On the other hand, addingcritical review → rewriting could increase metacognitive load, potentially enhancing learning effects.




8. Overseas Policies and Platform Responses

  • UK Ofqual: Clarified AI usage standards in exams and assignments starting in the 2025 academic year.

  • OECD: Added "creativity and responsible use" to the AI literacy framework.

  • Learning Management Systems (LMS): Canvas and Moodle provide AI detection logs to teachers and incorporate ethics education modules for students.




9. Reconstructing Learning Design—Cyclical Model

  1. Setting Subjective Questions: Students redefine tasks in their own words

  2. AI Utilization (Search and Generation): Information gathering and hypothesis generation

  3. Critical Re-examination: Detecting contradictions and conducting additional research

  4. Output: Distinguish between AI output and handwritten parts, and write reflections


By repeating these four stages, AI functions not as a "shortcut" but as a "detour thinking support tool."




10. Academic Integrity and Evaluation Reform

The true positive rate of detectors like Turnitin and GPTZero is only around 70%, with issues of false positives and evasion prompts. The evaluation side is shifting focus from "finished products" to "processes,"


  • Log Submission (version history and prompts)

  • Oral Examination

  • Peer Review


    are being combined intomulti-layered evaluation.




11. Conclusion and Recommendations

ChatGPT can be both a "learning shortcut device" and a "thinking expansion tool."



  • Educators should shift task design to process-oriented and guide transparent AI usage.

  • Students should use AI drafts as "starting points," elevating them into their own words through reconstruction and critical examination.

  • Policy Makers should support flexible guidelines and ongoing empirical research.


Building a learning ecosystem that co-evolves with AI while preserving the uniquely human activity of "writing is thinking" will be the core of next-generation education.


List of Reference Articles

  • Daniel Lawler "‘Writing Is Thinking’: Do Students Who Use ChatGPT Learn Less?" International Business Times Australia (July 2, 2025)ibtimes.com.au

  • Nataliya Kosmyna et al. “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task” MIT Media Lab Preprint (June 10, 2025)media.mit.edu##HTML_TAG_

← Back to Article List