Saturday 29 March 2025
For years, computer scientists have been working on a solution to detect and identify code generated by artificial intelligence (AI) languages, also known as large language models (LLMs). These AI-powered tools are capable of producing human-like code, making it challenging for developers to determine whether the code was written by a human or an AI. In a significant breakthrough, researchers have developed a method to detect and identify LLM-generated code using coding style features.
The detection system, called LPcode, is based on the idea that humans and AI languages exhibit distinct patterns in their coding styles. By analyzing these patterns, the system can determine whether a piece of code was written by a human or an AI. The researchers collected a dataset of human-written code and LLM-generated code, known as LPcode, which consists of over 20,000 lines of code.
The team used machine learning algorithms to identify coding style features that are unique to human-written code. They found that humans tend to use more consistent naming conventions, indentation styles, and commenting patterns than AI languages. These features were then used to train the detection system.
To test the effectiveness of LPcode, the researchers paraphrased a set of human-written code using four different LLMs: ChatGPT, Gemini-Pro, WizardCoder, and DeepSeek-Coder. They then compared the original code with its paraphrased versions to determine how well LPcode could detect the AI-generated code.
The results were impressive. LPcode was able to accurately identify AI-generated code in over 90% of cases. The system also demonstrated a high degree of accuracy in identifying which LLM generated the code, even when the code was highly similar to human-written code.
This breakthrough has significant implications for the development and maintenance of software. With LPcode, developers can now confidently detect and identify AI-generated code, ensuring that intellectual property is protected and preventing plagiarism.
The detection system also opens up new possibilities for collaborative coding between humans and AI languages. By identifying which LLM generated a piece of code, developers can understand the strengths and weaknesses of the AI language and adjust their own coding style accordingly.
In the future, LPcode has the potential to be integrated into popular code editors and IDEs, making it easier for developers to detect and identify AI-generated code. As AI languages continue to evolve, this detection system will play a crucial role in maintaining the integrity of software development.
Cite this article: “Breakthrough Detection System Identifies Artificial Intelligence-Generated Code with High Accuracy”, The Science Archive, 2025.
Ai Languages, Large Language Models, Coding Style Features, Machine Learning Algorithms, Human-Written Code, Lpcode, Detection System, Intellectual Property, Plagiarism, Collaborative Coding.
