AI and ML can reduce exam fraud

Within the larger field of education, the administration and supervision of examinations is one area where AI/ML can be helpful. We already have access to many software such as Proctor UProctor IO, exam room.AI, exam ban, etc., which enable web-based monitoring of exams. We can think of these systems as digital invigilators that use a combination of rule-based and ML-Drive heuristics to try to determine if the examinee is performing an action that is not allowed on the exam. For example, by measuring a student’s eye movements and keystroke frequency, E-Invigilator can determine if the student is attempting to open a new tab or window in the current system or is moving their eyes to read a secondary screen. The speed of answering questions and the enrolled student’s keystroke patterns can be used to determine whether or not it is the same student who is writing their exams. In an online environment, one can also use a network analyzer to find out what applications are using bandwidth on the student’s system; This can also help determine if the student is using more than just screens. With offline settings, we can still enlist the help of object and motion detection systems to keep track of all students and report real-time analysis to the regulator. In both online and offline environments where students are taking exams outside of testing centers, we can consider using real-time plagiarism detection to determine if identical answers are being circulated among students within the center.

Technical solutions
Although states like UP and Bihar have low technology penetration, national-level exams that use digital oversight, such as JEE are already available. Technology can help human supervisors better monitor a larger group of students. In order to bring technical solutions closer to poorly equipped test centers, one has to think in the direction of stand-alone hardware systems that can be placed in each class and which can stream data from devices such as e.g Bluetooth, video surveillance cameras and sound detection devices. Such support can be extended not only during the implementation of the tests, but also during the offline testing of subjective questions and their entry into the portal. However, their use has to be extrapolated with a pinch of salt, as the problem in states like UP and Bihar is not just technical, but needs to be addressed at institutional and societal levels.

The need for well-maintained exam centers
While AI-powered systems are regularly used to conduct and monitor investigations, they are limited in their ability to monitor like humans. One such issue is the limited scope that cameras can provide and contextual information for exams with different rules, such as open-book exams. Another issue is the student’s ability to challenge an incorrectly predicted instance of cheating, as any attempt to interact with the system during the exam will be blocked and reported. Second, the infrastructure is missing. Ensuring students who are sitting at home have rooms, can block them out for the duration of exams, and ensure access to power and stable WiFi during exam times is a challenge. To overcome this, well-maintained examination centers should be opened across the country that cater to different examination requirements. Such centers should be available to candidates of all backgrounds and requirements. The use of AI-powered systems for surveillance and fraud detection opens up a broader debate about human surveillance and cybersecurity that is currently beyond the scope of this discussion, but still needs to be considered. Therefore, we cannot completely eliminate the human aspect of surveillance, whether online or offline. However, we can reduce the burden on human invigilators and provide students with an environment where exams can be administered fairly.

(The author is Associate Professor, IIIT Delhi)