WEKO3
アイテム
【原著論文】生成AI とソフトウェア開発:中規模プロジェクトにおける人間とChatGPT のコミュニケーション分析
https://thu.repo.nii.ac.jp/records/2000682
https://thu.repo.nii.ac.jp/records/200068279c686ba-0af8-4c15-b56a-71d8d24e85bc
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
kiyou35_037.pdf (1.4 MB)
|
|
| Item type | 帝京平成大学紀要(1) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 公開日 | 2024-03-30 | |||||||||||
| タイトル | ||||||||||||
| タイトル | 【原著論文】生成AI とソフトウェア開発:中規模プロジェクトにおける人間とChatGPT のコミュニケーション分析 | |||||||||||
| 言語 | ja | |||||||||||
| その他のタイトル | ||||||||||||
| その他のタイトル | Generative AI and Software Development : An Analysis of Human-ChatGPT Communication in Medium-Scale Projects | |||||||||||
| 作成者 |
米澤, 直記
× 米澤, 直記
|
|||||||||||
| 主題 | ||||||||||||
| 主題Scheme | Other | |||||||||||
| 主題 | Conversation Analysis, Code Generation AI, Medium-scale Software Development, ChatGPT, Contextual Continuity | |||||||||||
| 内容記述 | ||||||||||||
| 内容記述タイプ | Abstract | |||||||||||
| 内容記述 | This study critically examines the utility and limitations of the generative AI (Artificial Intelligence) model, ChatGPT, in the development of medium-scale software, specifically focusing on a numerical puzzle game called “FunctorX". The research aims to shed light on how conversational AI can contribute to enhancing the efficiency and productivity of software development. While ChatGPT has shown promising results in small-scale programming tasks, it struggles to maintain consistency in larger projects from design to code generation, necessitating human intervention. In our methodology, we engaged in 643 dialogues with ChatGPT throughout the development of FunctorX, capturing conversation logs and categorizing them using 16 labels corresponding to various development stages, such as “Requirement Analysis", “Design", “Implementation", and “Deployment". The labels helped us determine the extent to which ChatGPT was useful or problematic at each stage. Notably, conversations around the “Implementation: Number Input" and “Implementation: Event Listener" were the most frequent, indicating a concentration of discussion during the implementation phase. Our lexical analysis, done via MeCab, revealed that the dialogues were primarily focused on technical aspects, as words like “button", “cell", “input" appeared frequently from the human side, whereas ChatGPT often used terms like “element" and “case". FunctorX's development also exposed difficulties in coordinating the event handler and the number input mechanism, revealing a nuanced understanding of where human expertise remains critical. Moreover, our research brings forth several recommendations for improving conversational code-generating AI. These include ensuring contextual continuity, clarifying the rationale behind any strategy shifts, reducing excessive apologies, and improving semantic accuracy. We posit that these enhancements are crucial for more effective code generation. Although AI in code generation offers exciting prospects, it is not without challenges that need addressing. By implementing these improvements, we expect to witness a significant leap in the AI's effectiveness and precision, making it more valuable in software development environments. | |||||||||||
| 出版者 | ||||||||||||
| 出版者 | 帝京平成大学 | |||||||||||
| 言語 | ja | |||||||||||
| 言語 | ||||||||||||
| 言語 | jpn | |||||||||||
| 資源タイプ | ||||||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_6501 | |||||||||||
| 資源タイプ | departmental bulletin paper | |||||||||||
| ISSN | ||||||||||||
| 収録物識別子タイプ | ISSN | |||||||||||
| 収録物識別子 | 13415182 | |||||||||||
| 書誌情報 |
ja : 帝京平成大学紀要 巻 35, p. 37-48, 発行日 2024-03 |
|||||||||||
