AVERT (Authorship Verification and Evaluation Through Responsive Testing): an LLM-Based Procedure that Interactively Verifies Code Authorship and Evaluates Student Understanding

The rapid development of generative artificial intelligence challenges traditional plagiarism detection systems. Classic methods, based primarily on similarity, are more and more ineffective against content that is AI-generated, and which resembles authentic student submissions. This requires the use of detection methods that not only establish authorship but also augment student engagement and motivation, thus discouraging future plagiarism attempts. In this work-in-progress paper, we present AVERT (Authorship Verification and Evaluation through Responsive Testing), a proof-of-concept approach for establishing the authorship of programming assignments. Integrated in an LLM-based chatbot, it proactively assesses students' understanding of programming assignments by interactively questioning them about the process of developing their own code. This way the reasons for which students may be dishonest, such as motivation and lack of subject comprehension, are addressed. Using advanced prompting techniques and formulating questions that employ the Socratic method, AVERT dynamically generates questions to verify the students' understanding of the coding process. By asking students to explain the logic behind their solutions, it also improves analytical skills and engages students in the learning process. By automating the assessment of authorship, AVERT alleviates the logistical problems encountered by educators in large classroom environments where individualized assessment demands large resources. It also helps instructors in grading students by making the details of each conversation available for evaluation. AVERT provides a scalable and efficient method for establishing authorship, overcoming the shortcomings of other techniques by directly addressing the primary goal of programming assignments: assessing students' abilities to write functional code.