LASRCOP: Lightweight Automated Session Recording to Certify Original Programming
Andrew Kramer Dakota State University
Abstract The proliferation of sophisticated artificial intelligence (AI) tools poses a significant threat to academic integrity in programming courses. Generative AI models capable of solving complex programming problems based only on a written prompt are a powerful tool but present a great temptation to students seeking a shortcut on their programming assignments. Worse yet, this creates a new challenge for instructors in detecting and substantiating instances of academic misconduct as each generated code sample is unique, reasonable, and difficult to differentiate from human work. A solution is needed, but must ensure fairness and privacy, while also being simple to implement and unobtrusive.
To solve this, we propose LASRCOP (Lightweight Automated Session Recording to Certify Original Programming), a novel system designed to mitigate cheating among programming students by recording their work in real time while also protecting their privacy, not requiring them to install any additional software, and being quick and easy for instructors to implement.
Since AI-assisted cheating can be difficult to detect and prove, ideally an instructor would observe a student’s work live to ensure the code is an original creation. This is clearly impractical in a classroom setting and is impossible for assignments which students complete outside the classroom. A naive solution might be to require that students record their screens as they program, however this presents many additional problems. First, students may have privacy concerns as they may not wish for an instructor to see other open windows or applications. Second, this will result in many gigabytes or terabytes of screen recording files which are difficult to transmit and store. Finally, students may not have the knowledge or ability to record their computer screens or may experience problems with software compatibility. As such, asking students to record their programming work is impractical.
To facilitate session recording while solving these problems, we present LASRCOP. LASRCOP is an application designed around “Asciinema”, an open-source tool for recording and replaying terminal sessions, which is automatically invoked upon user login and transparently creates a recording of the programmer’s work. It is designed to be installed in a centralized programming environment and can easily be enabled or disabled for specific students and classes based on a user/group architecture. Because recording files are text-only, they are also small and easy to store, and because session recording occurs entirely on the server-side there is no special software for the student to install. Additionally, the student’s privacy is respected as only their programming work is recorded. A student need only log in using a standard SSH client and LASRCOP handles the rest.
If academic dishonesty is suspected an instructor may review these session recordings to help determine whether the student wrote the code themselves or copied from another source. As recordings are viewable in real-time, copy-pasted code becomes immediately obvious.
In this session, the presenters will describe LASRCOP, share lessons learned from its use in programming classes, and show how attendees can implement it in their own programming classes to help detect AI-assisted academic dishonesty.