Digital examinations have taken away all the hurdles in the examination process and made them more accessible and efficient. However, with the advancements in the use of AI agents, examination authorities are now facing new challenges; the existing online proctoring systems that mainly monitor students' activities through webcam, browser controls, and screen sharing may not be enough.
Multi-tasking, analysing data, producing outputs, and engaging on digital platforms with little to no input from a human. "In an exam setting, this begs further questions about integrity, identity checks, and the integrity of exam systems,” says Manish Mohta, managing director of Learning Spiral.
The limitations of traditional proctoring
Current online proctoring employs behavioural cues to watch out for possible nefarious activity: screen saver use, reading off the screen, presence of an additional person and changing applications. Though these cues are effective at putting off most use of AI to cheat, they have gaps when it comes to more advanced forms of cheating using AI.
For instance, a system used to block access to the internet would prevent external screens, but it would not know whether a candidate was being helped by another device or whether the answers were being generated by a script running the AI software. And camera systems can track the identity and behaviour of a candidate, but cannot confirm all answers are being provided by the candidate.
"This therefore produces the necessity to move away from security-based surveillance to a broader approach considering the whole assessment process," Mohta says.
Building AI-ready examination security
New digital exam systems need to use multiple lines of defence and not a single point of monitoring. Secure browser technology, identity verification, behaviour analysis, device monitoring and exam log monitoring can all be used together to identify suspicious behaviour.
AI-powered analytics can help you flag anything suspicious. However, automation shouldn't be justification for ceding to suspicion. "First, allow the process to unfold without auto-flagging, which should always include a human reviewer and an opportunity for someone to present an alternative explanation for the behaviour," Mohta says.
Test delivery agencies and test publishers will need to consider how testing tools are used alongside AI. Certain technical and legal restrictions around access to testing tools and restrictions on unmonitored use of the system may reduce misuse risk.
Security also needs to be balanced. Monitoring all activity can be privacy-invasive, and poor-quality detection could effectively discriminate against certain disability and device combinations by denying access that others would not.
Rethinking assessment design
AI exams, not Cybersecurity, can win for you! Mohta says, "In the end, cybersecurity doesn't solve all the problems we have with AI-assisted exams." Educational systems should consider their assessments and should be moving away from conventional learning indicators (knowledge/ skills/ capacity).
Mohta said, "Application questions, case studies, scenario-based test questions, and complex multi-stage assessments are the best way to gain an understanding of a candidate." These methods move away from simply denying external support by looking at what they know and reasoning skills.
The way forward
As the capabilities of AI agents increase, the security of exams must increase alongside. The institutions and providers that deliver assessments require technology that integrates smart surveillance, safe use of AI, assessment design, and secure technology.
The future of digital exams won't be dictated by surveillance. It'll be defined by secure, transparent, scalable, and equitable assessment environments. It's time for a new chapter-one that takes a step back from everything proctoring and advances safe, fair assessments.
The views expressed in this article are those of the author and expert alone. The opinions and facts in this article do not represent the stand of Edexlive
