基於大型語言模型的生成式人工智慧除了可以進行問答、翻譯和創作等任務外，在程式開發領域能生成程式碼、註解、測試資料或檢查錯誤等協助程式開發，許多程式開發者開始在軟體開發工作中使用生成式人工智慧。儘管人工智慧已經可以取代一部分程式撰寫工作，程式開發者在使用人工智慧生成程式碼時仍然存在一些問題。受限於自動化偏見，程式開發者對生成程式碼的視覺關注少於自行撰寫的程式碼，甚至可能在未經檢查的情況下直接使用，導致程式開發者需要面對陌生的人工智慧程式碼混合在程式碼專案中，提升理解程式碼的困難。本研究提出一種基於程式語言的機制分離人工智慧生成程式碼，將人工智慧程式碼從原始碼檔案中分離出，以改善整體程式碼可讀性。我們使用認知負荷量表和程式碼複雜度指標進行評估，結果顯示程式碼可讀性確實得到改善。;The generative AI based on large language models can not only perform tasks such as Q&A, translation, and creation, but also assist in software development by generating code, comments, test data, or checking for errors. Many developers have begun using generative AI in their software development work. Although AI can already replace some parts of coding, there are still some issues when developers use AI to generate code. Due to automation bias, developers pay less visual attention to AI-generated code compared to code they write themselves, and they might even use it directly without checking, leading to difficulties in understanding the code when unfamiliar AI-generated code is mixed into their projects. This study proposes a mechanism based on programming languages to separate AI-generated code, isolating it from the original code files to improve overall code readability. We used cognitive load scales and code complexity metrics for evaluation, and the results show that code readability has indeed improved.
