Generative Artificial Intelligence (GenAI) is rapidly reshaping coding practices in higher education, raising critical questions about how student learning is reflected in assessed work. This study presents a large-scale quantitative analysis of nearly 6,500 Python summative assessment coding submissions from large cohorts of several hundred postgraduate MSc students at Imperial College London, across programmes in computational science and engineering, machine learning, and data science. Comparing submissions before and after the widespread adoption of GenAI tools, the results reveal a clear shift in the nature of student work: although submissions have become longer on average, they exhibit significantly lower structural complexity across established metrics, including cyclomatic (McCabe) complexity, cognitive complexity, and Halstead measures. This reduction in code complexity suggests a move away from constructing algorithmic solutions towards less cognitively demanding approaches, with potential implications for the development of core programming and computational thinking skills. By providing robust empirical evidence from a large, discipline-specific dataset, this study highlights a measurable transformation in coding assessment outputs in the GenAI era, raising questions about the validity of current assessment practices.

Authors:
Marijan Beg, Imperial College London, United Kingdom
Tiffany Chiu, Imperial College London, United Kingdom

About the Presenter(s)
Dr Marijan Beg is a Senior Teaching Fellow at Imperial College London (UK), teaching on MSc computational science and engineering courses.

Connect on Linkedin
https://www.linkedin.com/in/marijanbeg/

See this presentation on the full schedule – Saturday Schedule

Bookmark