Volume 4,Issue 3
The Effects of AI Assistance Granularity on Writers' Cognitive Load: An Empirical Study in Human-AI Co-Writing
As more and more writers use large language models (LLMs) in their work, they face a big problem: how to enjoy the time savings that the technology brings without losing their own deep thinking and creative input. This is the key to making human-machine synergy truly effective. This paper methodically presents "AI-assisted granularity" as a key design element. By combining this idea with the cognitive load theory (CLT), we look at how it affects writers' intrinsic, extraneous, and germane cognitive loads in different ways. We modified the Leppink scale to create an AI-driven cognitive load assessment instrument for writing. We had 21 college students do argumentative writing tasks with AI help in three different ways: sentence-level, paragraph-level, and full-text structural guidance. The results show that full-text structural prompts greatly lower both intrinsic and extrinsic cognitive load, but they also make people less interested. On the other hand, prompts at the paragraph level are the best at increasing generative cognitive load, which shows a "moderate guidance" effect. On the other hand, sentence-level prompts make the brain work harder while giving the writer more creative freedom. At different levels of AI assistance, this study shows the ongoing conflict between "cognitive offloading" and "preserving agency." It suggests that writing tools in the future should have adjustable granularity mechanisms to make the best experiences for people and machines working together. These results offer a novel theoretical framework and design insights for AI writing research through the lens of cognitive load.
[1] Wu T, He S, Liu J, et al., 2023, A Brief Overview of ChatGPT: The History, Status Quo and Potential Future Development. IEEE/CAA Journal of Automatica Sinica, 10(5): 1122–1136.
[2] Marzuki, Widiati U, Rusdin D, et al., 2023, The Impact of AI Writing Tools on the Content and Organization of Students' Writing: EFL Teachers' Perspective. Cogent Education, 10(2): 18.
[3] Dhillon PS, Molaei S, Li J, et al., 2024, Shaping human-AI collaboration: Varied scaffolding levels in co-writing with language models, Proc. [Conf. Abbrev.], [Page range].
[4] Fu L, et al., 2023, Comparing Sentence-Level Suggestions to Message-Level Suggestions in AI-Mediated Communication. arXiv preprint arXiv:2304.05678.
[5] Zhao X, 2023, Leveraging Artificial Intelligence (AI) Technology for English Writing: Introducing Wordtune as a Digital Writing Assistant for EFL Writers. RELC Journal, 54: 890–894.
[6] Sweller J, 1988, Cognitive Load During Problem Solving: Effects on Learning. Cognitive Science, 12(2): 257–285.
[7] Sweller J, 1994, Cognitive Load Theory, Learning Difficulty, and Instructional Design. Learning & Instruction, 4(4): 295-312.
[8] Leppink J, Paas F, Van der Vleuten CP, et al., 2013, Development of an Instrument for Measuring Different Types of Cognitive Load. Behavior Research Methods, 45(4): 1058-1072.
[9] Zhang S, Wu Y, Fu Z, et al., 2020, Psychometric Properties of the Chinese Version of the Instrument for Measuring Different Types of Cognitive Load (MDT-CL). Journal of Nursing Management, 28(2): 277-285.
[10] Kirschner PA, Sweller J, 2006, Why Minimal Guidance During Instruction Does Not Work: An Analysis of the Failure of Constructivist, Discovery, Problem-Based, Experiential, and Inquiry-Based Teaching. Educational Psychologist, 41(2).