Over 20% of Indians on GitHub Used AI to Write Python Functions: Study

A study published by the University of Utrecht, the Complexity Science Hub (Vienna), Corvinus University of Budapest, and the HUN-REN Centre for Economic and Regional Studies in Hungary outlined the use of AI to write code globally and its economic impact. 

The authors trained a neural classifier to detect AI-generated Python functions in 80 million GitHub commits by 200,000 developers over the past six years. 

“By December 2024, AI wrote an estimated 30.1% of Python functions from US contributors, versus 24.3% in Germany, 23.2% in France, 21.6% in India, 15.4% in Russia and 11.7% in China,” read the study. 

The authors also said that newer GitHub users use more AI than veterans, and using AI 30% of the time raises quarterly commits by 2.4%. 

By combining wage data with estimates of the amount of programming done in almost 900 US occupations, the study calculates that AI generates $9.6-14.4 billion annually in the US software sector. 

The researchers thoroughly trained the neural classifier using a vast dataset to ensure it correctly identifies AI-generated Python functions. They collected human-written Python functions from 2018 onwards, covering diverse examples across different time periods.

They also created synthetic AI-written functions using a two-step process: one LLM described human functions, and another LLM generated corresponding code based on these descriptions.

The researchers then used GraphCodeBERT, a cutting-edge model, to tokenise and embed the code before fine-tuning it into a classifier. The model was assessed with an out-of-sample ROC AUC score of 0.964, indicating high detection precision.