TLDR: A study surveyed 388 university students to predict ChatGPT usage in assignments using the XGBoost algorithm. It found that students’ learning habits, subject preferences, and attitudes towards AI are key predictors, with experiential familiarity (using ChatGPT for new learning) being the strongest factor. The research suggests that AI use is driven by cognitive style and pedagogical needs rather than demographics, highlighting the need for nuanced, AI-aware assessment policies that differentiate responsible use from misuse.
The rapid emergence of generative AI tools like ChatGPT has significantly transformed the educational landscape, sparking widespread discussions about their impact on learning and academic integrity. While previous research has explored the opportunities and risks, there has been a notable gap in quantitative analysis of how students actually behave when completing assignments using these tools. Understanding the influence of AI on real-world academic practices, especially assignment preparation, is a critical and timely research area.
A recent study, titled Predicting ChatGPT Use in Assignments: Implications for AI-Aware Assessment Design, by Surajit Das and Aleksei Eliseev, addresses this gap by analyzing survey responses from 388 university students, predominantly from Russia, with a subset of international participants. The researchers employed the XGBoost algorithm, a powerful machine learning technique, to model the predictors of ChatGPT usage in academic assignments.
Study Methodology and Findings
The study collected data through a structured self-reported survey, capturing various aspects such as demographics, daily practices, engagements, and attitudes towards AI. The core of the methodology involved two distinct classification tasks for predicting ChatGPT usage:
-
Binary Classification: This task categorized students into active ChatGPT users (those using it ‘Sometimes’, ‘Frequently’, or ‘Almost all the time’) and non-users (those using it ‘Never’ or ‘Rarely’). The model achieved a strong predictive performance with 80.1% test accuracy, demonstrating its ability to reliably distinguish between these two groups.
-
Multiclass Classification: This more granular task utilized the full five-level ordinal scale of usage frequency. While more complex due to data imbalance, this model achieved 64.5% test accuracy, providing detailed insights into varying levels of AI adoption.
Key predictive factors identified by the study included students’ learning habits, their subject preferences, and their overall attitudes toward AI. Interestingly, demographic variables such as gender, institution, or country showed minimal influence on ChatGPT usage, suggesting that the propensity to use AI tools transcends demographic boundaries and is more closely tied to cognitive style, attitudes toward technology, and pedagogical needs.
The most impactful variable in predicting ChatGPT use for assignments was found to be ‘ChatGPT_Used_in_New_Learning’. This indicates that students who frequently use generative AI tools for learning new concepts are substantially more likely to also use them for assignment preparation. Other significant factors included ‘ChatGPT_4_Non_Performer’, reflecting the use of ChatGPT in subjects of lesser interest or where students underperform, and ‘ChatGPT_Important_4_Good_Student’, which suggests that even high-performing or confident students use ChatGPT, potentially for optimizing time or enhancing productivity.
Also Read:
- Are LLMs Truly Collaborative Partners? A Study’s Insights into Human-AI Interaction
- University Students Grapple with Anxiety and Distrust Over AI Integration in Academia
Implications for Education
The findings suggest that while generative AI can enhance access to knowledge, unchecked reliance may erode critical thinking and originality. The study reveals a behavioral continuum in ChatGPT adoption, shaped more by cognitive orientation and learning strategies than by demographics. Students engage with ChatGPT both as a conceptual aid in difficult subjects and as a productivity tool in familiar ones.
This research highlights the need for educators and policymakers to move beyond blanket restrictions or permissive policies. Instead, it calls for a nuanced approach that differentiates responsible AI use from misuse. The authors propose discipline-specific guidelines and reimagined assessment strategies to balance innovation with academic rigor. Integrating AI literacy into curricula and setting clear expectations around its use could guide students toward ethical and effective engagement with these powerful tools.
