Logo

ARDEN

Insights Beyond the Headlines

Science

Exploring the Cognitive Cost of Using Large Language Models in Education


A team of researchers from MIT Media Lab investigates the impact of Large Language Models (LLMs), like ChatGPT, on cognitive engagement in educational settings. Their findings reveal important implications for integrating AI tools in learning environments and their potential effects on long-term learning skills.

  

Published on 19/06/2025 13:31


    • The study provides valuable insights into the cognitive effects of using large language models (llms) in educational contexts.
    • It uses a robust methodology with eeg monitoring to objectively assess cognitive engagement and load across different groups.
    • The research highlights potential areas of improvement for educational practices involving ai.
    • The study's setup might not fully capture real-world educational environments and diverse student interactions with llms.
    • Limited sample size (54 participants) may affect the generalizability of the findings to broader populations.
    • Participants in the brain-only group showed the highest cognitive engagement, suggesting more profound learning processes.
    • The study identifies the varying levels of cognitive load associated with different methods, offering a detailed understanding of how students interact with various tools.
    • Low engagement in the llm group points to potential drawbacks of over-relying on ai tools, which can diminish critical thinking and memory skills.
    • Those who transitioned from llm usage to no-tools faced difficulties re-engaging, indicating potential long-term dependency issues.
    • Essays from the llm group were consistently acceptable, highlighting the capability of ai to aid in producing standard quality content efficiently.
    • The research may prompt educators to reconsider the balance between ai assistance and independent student work to foster creativity.
    • Llm-written essays were more homogeneous and formulaic, potentially stifling originality and creative expression.
    • Reduced sense of ownership reported by llm users can lead to decreased personal engagement and retention of learnt material.

    • Introduction to the Study

    Nataliya Kosmyna and her co-authors, representing MIT Media Lab and collaborating institutions, have embarked on a research endeavor to explore the cognitive implications and impacts on educational practices when utilizing Large Language Models (LLMs) such as ChatGPT. This research centers specifically on the domain of essay writing, a crucial component of educational processes, to assess how these advanced AI tools affect cognitive XXYPLACEHOLDER0YXX engagement and overall learning outcomes.

    Methodology and Participant Breakdown

    The study involved 54 participants who were divided into three distinct groups, each employing different tools and methods for essay writing tasks across four sessions. The first, the LLM group, was provided with ChatGPT as their primary tool. The second group, known as the Search Engine group, utilized traditional search engines like Google but had no access to AI assistance. The third, termed the Brain-only group, relied entirely on their cognitive faculties without external tools. During a fourth session, some participants switched groups to examine potential carry-over effects of their previous tool usage on cognitive performance.

    Measurement Techniques

    To gauge cognitive engagement and cognitive load, the researchers employed Electroencephalography (EEG) to monitor and record brain activity patterns of participants during their essay XXYPLACEHOLDER1YXX writing tasks. Additionally, the essays produced were subjected to analysis using Natural Language Processing (NLP) techniques and were evaluated by both human teachers and an AI judge to ensure comprehensive assessment criteria encompassing quality, originality, and thematic deviations.

    Key Findings on Cognitive Engagement

    The EEG data revealed compelling differences between the three groups in terms of cognitive load and neural activity. The Brain-only group consistently demonstrated the highest levels of brain connectivity and active engagement, indicating they exerted more cognitive effort compared to the other groups. The Search Engine group was positioned in the middle, exhibiting moderate levels of engagement. In stark contrast, participants relying on the LLM exhibited the lowest neural connectivity, a signifier of reduced cognitive effort and engagement during the essay writing process.

    Challenges in Switching Cognitive Tools

    XXYPLACEHOLDER2YXX

    The study further explored the repercussions of dependency on LLM tools. Participants who transitioned from the LLM group to the Brain-only group in the fourth session faced significant challenges. These individuals struggled to re-adapt to cognitive engagement without the support of AI tools. Their EEG readings showed reduced brain connectivity, and they experienced difficulty with memory recall, reinforcing concerns regarding the potential consequences of over-relying on AI for cognitive tasks.

    Insights on Essay Quality and Homogeneity

    Analysis of the essays pointed to notable patterns distinguishing the outputs of different groups. Essays generated by participants in the LLM group were found to be more homogeneous, with less deviation from central topics and a tendency toward formulaic phrasing. This consistency, while perhaps producing acceptable essays in the short term, raises concerns about originality and XXYPLACEHOLDER3YXX creative thought in educational practices relying heavily on AI-generated content.

    Personal Engagement and Memory Retention

    Interview insights gathered from LLM users further underscored the potential drawbacks of AI reliance. Many reported a diminished sense of ownership over their work and found it challenging to recall specific details or quotes from their own essays. Such findings raise important considerations regarding engagement and personal involvement in educational tasks, which are crucial for meaningful learning and retention.

    Concluding Remarks and Future Directions

    The conclusions drawn from this study underscore the intricate landscape of integrating AI tools like LLMs in educational settings. While these tools can facilitate certain tasks and may yield short-term gains in essay production, the potential accumulation of “cognitive debt” cannot be ignored. The researchers stress that indiscriminate use of AI in XXYPLACEHOLDER4YXX education may adversely affect long-term learning skills, critical thinking, and memory retention. As educational institutions continue to explore innovative tools, there is an urgent call for measured approaches that balance AI utilization with fostering deep cognitive engagement and sustainable learning outcomes. It is crucial to ensure that the next generation of learners does not compromise critical educational foundations for the sake of convenience and efficiency brought by technological advancements.


    The article explores a study conducted by researchers from MIT Media Lab, which investigates the cognitive costs associated with using Large Language Models, such as ChatGPT, in educational contexts, specifically during essay writing. The study highlights differences in cognitive engagement among participants using different tools (LLMs, search engines, and no tools) and raises concerns about potential negative impacts on long-term learning skills when relying on AI tools. The article suggests careful integration of AI into educational practices to ensure that the enhancement of writing tasks does not undermine crucial cognitive development.

    • Subjectivity: The article is mostly objective, presenting findings from a scientific study with an emphasis on data-driven insights and observed implications. however, it includes some subjective elements in discussing potential risks and recommendations for integrating ai tools in educational settings.
    • Polarity: The overall tone of the article is cautionary, highlighting potential negative outcomes associated with using ai tools excessively in education. it suggests a balanced approach to leveraging technology, emphasizing the importance of preserving cognitive engagement and critical thinking skills.

      Nataliya Kosmyna is a researcher affiliated with the MIT Media Lab, known for her work in exploring the cognitive implications and impacts of technology, particularly in the context of educational practices and the integration of artificial intelligence tools.

      Large Language Models are advanced artificial intelligence systems designed to understand and generate human-like text, capable of performing a wide range of language-based tasks. They are trained on vast amounts of data and can produce coherent, contextually relevant text, often used in applications such as chatbots and automated writing tools.

      Electroencephalography is a non-invasive method used to record electrical activity of the brain. EEG measures voltage fluctuations resulting from ionic current within the neurons of the brain, often used to assess brain activity related to different cognitive tasks and states.

      Natural Language Processing is a field of artificial intelligence focused on the interaction between computers and humans through natural language. It involves the application of algorithms and models to analyze, understand, and generate human language in a way that is valuable for tasks like translation, sentiment analysis, and information extraction.

      Cognitive engagement refers to the mental effort and investment required for processing information and completing tasks. It involves active participation and concentration, and is essential for effective learning, problem-solving, and critical thinking.

      Cognitive load is the amount of mental effort being used in the working memory. It refers to the demands imposed on cognitive resources during information processing and can affect an individual's ability to learn, remember, and perform tasks efficiently.

      Cognitive debt is a metaphorical term describing the potential negative consequences of relying on technology to reduce cognitive effort. It suggests that while technology may provide immediate benefits, it can impair long-term skills such as critical thinking, memory retention, and independent problem-solving.

    Highest in brain-only group, intermediate in search engine group, lowest in llm group

    Cognitive Engagement (EEG) - Brain Connectivity

    EEG recordings showed that participants in the Brain-only group exhibited the highest levels of brain connectivity and engagement, suggesting more cognitive effort. The Search Engine group had moderate engagement, while the LLM group demonstrated the lowest neural connectivity, indicative of reduced cognitive effort.

    Lower connectivity and poorer memory recall

    Struggle with Cognitive Engagement after Switching from LLM to Brain-only

    Participants who switched from using the LLM to the Brain-only method in the fourth session showed significantly decreased brain connectivity and struggled with memory recall. This suggests that reliance on LLMs can lead to difficulties in re-engaging cognitively without AI assistance.

    More homogeneous, less topic deviation, more formulaic phrasing in llm group

    Essay Homogeneity and Phrasing

    Essays produced by the LLM group were noted to be more homogeneous in terms of content, with less deviation from the main topic and more formulaic phrasing. This indicates that LLMs may guide users to produce consistent but less diverse and original content.

    Lower sense of ownership and weaker ability to recall quotes

    Sense of Ownership and Quote Recall in LLM Users

    Participants using LLMs reported a lower sense of ownership over their essays and had more difficulty recalling specific quotes from their work. This suggests that the use of AI can diminish personal engagement and retention of information, which are critical for learning and development.