AI-Generated code detection: an examination of current tools in education
Uloženo v:
| Název: | AI-Generated code detection: an examination of current tools in education |
|---|---|
| Autoři: | Cuellar Argotty, Juan Esteban |
| Přispěvatelé: | Manrique Piramanrique, Rubén Francisco, Facultad de Ingeniería |
| Informace o vydavateli: | Universidad de los Andes Ingeniería de Sistemas y Computación Facultad de Ingeniería Departamento de Ingeniería de Sistemas y Computación |
| Rok vydání: | 2025 |
| Sbírka: | Universidad de los Andes Colombia: Séneca |
| Témata: | AI-generated code, AI-Generated Code Detection, Software Engineering Education, Ingeniería |
| Popis: | This document explores the challenge of detecting AI-generated Python code in education, highlighting limitations of current detection tools, especially against simple obfuscation techniques. It emphasizes the need for advanced, resilient detection methods and ethical AI use in academic settings. ; This document explores the challenge of detecting AI-generated Python code within educational settings, focusing on first-semester student solutions on the Senecode platform. It outlines the creation of a dataset combining both human-written and AI-generated code (across multiple obfuscation variants) and evaluates seven widely used AI detectors. Despite each tool’s strengths in certain areas—such as high precision or high recall—none consistently excels, and simple code modifications substantially reduce detection accuracy. The study underscores the trade-off between minimizing false positives and maximizing true detection, highlighting the risk of unjustly penalizing students or overlooking AI misuse. Recommendations include developing more advanced, code-specific detection methods, employing a multi-layer approach that integrates human oversight, and fostering ethical AI use through clear academic policies. ; Pregrado |
| Druh dokumentu: | bachelor thesis |
| Popis souboru: | 29 páginas; application/pdf |
| Jazyk: | English |
| Relation: | Uchendu, Adaku, Venkatraman, Saranya, Le, Thai, and Lee, Dongwon. "Catch Me If You GPT: Tutorial on Deepfake Texts". Available at: https://aclanthology.org/2024.naacl-tutorials.1.pdf; Weber-Wulff, Debora, Anohina-Naumeca, Alla, and Bjelobaba, Sonja. "Testing of detection tools for AI-generated text". SpringerLink, 2023. Available at: https://link.springer.com/article/10.1007/s40979-023-00146-z; Mitchell, Eric, Lee, Yoonho, Khazatsky, Alexander, Manning, Christopher D., and Finn, Chelsea. "DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature". Available at: https://openreview.net/pdf?id=UiAyIILXRd; Pan, Wei Hung, Chok, Ming Jie, and Wong, Jonathan Leong Shan. "Assessing AI Detectors in Identifying AI-Generated Code: Implications for Education". IEEE Xplore, 2023. Available at: https://ieeexplore.ieee.org/document/10554754; Li, Yafu, Li, Qintong, and Cui, Leyang. "MAGE: Machine-generated Text Detection in the Wild". Available at: https://arxiv.org/pdf/2305.13242; Xu, Zhenyu, and Sheng, Victor S. "Detecting AI-Generated Code Assignments Using Perplexity of Large Language Models". AAAI Conference on Artificial Intelligence, 2023. Available at: https://ojs.aaai.org/index.php/AAAI/article/view/30361; New Era of Artificial Intelligence in Education: Towards a Sustainable Multifaceted Revolution. Available at: https://www.mdpi.com/2071-1050/15/16/12451; The role and impact of ChatGPT in educational practices: insights from an Australian higher education case study. Available at: https://link.springer.com/article/10.1007/s44217-024-00126-6; https://hdl.handle.net/1992/75503; instname:Universidad de los Andes; reponame:Repositorio Institucional Séneca; repourl:https://repositorio.uniandes.edu.co/ |
| Dostupnost: | https://hdl.handle.net/1992/75503 |
| Rights: | Attribution-NonCommercial-NoDerivatives 4.0 International ; http://creativecommons.org/licenses/by-nc-nd/4.0/ ; info:eu-repo/semantics/embargoedAccess ; http://purl.org/coar/access_right/c_f1cf |
| Přístupové číslo: | edsbas.95086CC5 |
| Databáze: | BASE |
Nájsť tento článok vo Web of Science | Abstrakt: | This document explores the challenge of detecting AI-generated Python code in education, highlighting limitations of current detection tools, especially against simple obfuscation techniques. It emphasizes the need for advanced, resilient detection methods and ethical AI use in academic settings. ; This document explores the challenge of detecting AI-generated Python code within educational settings, focusing on first-semester student solutions on the Senecode platform. It outlines the creation of a dataset combining both human-written and AI-generated code (across multiple obfuscation variants) and evaluates seven widely used AI detectors. Despite each tool’s strengths in certain areas—such as high precision or high recall—none consistently excels, and simple code modifications substantially reduce detection accuracy. The study underscores the trade-off between minimizing false positives and maximizing true detection, highlighting the risk of unjustly penalizing students or overlooking AI misuse. Recommendations include developing more advanced, code-specific detection methods, employing a multi-layer approach that integrates human oversight, and fostering ethical AI use through clear academic policies. ; Pregrado |
|---|