In our previous blog, we provided an insightful overview of AI-assisted cheating, exploring its various forms and potential countermeasures. In this follow-up blog, we delve deeper into the detection techniques employed by e-assessment solutions, designed to identify and alert administrators of AI-driven dishonesty. Join us as we demystify these detection methods and their effectiveness in maintaining exam integrity.
Awarding bodies and professional testing organisations realise the need to adopt an e-assessment platform, especially post-covid. The need for robust cheating detection methods has become paramount, not only because at-home online testing increases the temptation to cheat, but also because AI has made cheating much easier. To maintain the integrity of online assessments, various tools have emerged to recognise when candidates have used AI, such as ChatGPT, to answer exam questions.
A handy dandy guide to arming yourself against AI is desperately needed (stay tuned for our “comprehensive guide on AI” ebook), and this is a good start. This blog post explores some of the most effective methods of detecting the use of AI, including online proctoring, Turnitin, and software upgrades that enhance cheating detection capabilities. Knowledge is power, and detection is the first line in your arsenal in defending your exams from AI. Next week’s post will cover all the tools you have at your disposal to prevent the use of AI to cheat.
Online Proctoring
AI proctoring, such as Proctorio, is an excellent approach in the fight against AI during online exams by using features that detect and report suspicious activity. In this recorded proctoring method, the system captures video and audio of the candidate throughout the exam and employs advanced algorithms to analyse their actions. By detecting suspicious activities such as unusual eye movements, unauthorised materials or devices, or additional individuals in the testing area, AI proctoring provides instructors with comprehensive reports highlighting potential cheating incidents. It is then up to the instructors to review these reports and make the final judgement call on whether cheating occurred, ensuring a human touch remains in the evaluation process while harnessing the power of AI for effective exam monitoring.
Features to watch out for in online proctoring:
ID verification
What it is: Software that enhances exam security by implementing a robust identity verification process. As part of this process, the software captures an image of the test-takers ID card, along with images of the test-taker themselves. This method ensures that the person taking the exam is indeed the authorised individual, preventing impersonation or fraud and maintaining the integrity of online exams.
How it detects AI cheating: Thwarting attempts at impersonation or AI-generated faces, such in the case that a candidate overlays their face onto an unauthorised person in order to make it appear as though they took the exam.
Face detection
What it is: After an ID card is presented and the identification of the candidate is proven to be them indeed, the proctoring software then deploys face detection software. In an e-assessment context, this technology is designed to identify human faces within the camera’s field of view during an online exam. Unlike facial recognition software, which focuses on matching a specific individual’s facial features, face detection simply detects the presence of a face without identifying the person.
How it detects AI cheating: In preventing cheating with AI, face detection software helps ensure that candidates are physically present during the exam and do not use AI-generated faces or other deceptive means to impersonate themselves or others. This technology also helps detect the use of AI-driven tools that may attempt to manipulate the test-taking environment by identifying unusual or suspicious face-related activities. By ensuring the test-taker’s presence and maintaining a secure exam environment, face detection software contributes to online assessments’ overall integrity and fairness.
Gaze detection
What it is: Gaze detection is a technology that tracks and analyses a candidate’s eye movement and gaze direction during an online exam. By monitoring where a test-taker is looking, gaze detection can identify unusual patterns or behaviours that may suggest cheating attempts.
How it detects AI cheating: Gaze detection plays a crucial role in preventing cheating with AI during online exams by monitoring test-takers’ eye movements and gaze patterns. It can identify when a test-taker’s gaze frequently shifts away from the screen, which may indicate attempts to access unauthorised materials, such as another device with AI tools on it (such as in the case of attempting to get around a lockdown browser) .
The software also monitors gaze duration, tracking how long a test-taker spends looking at a specific area on the screen to help identify potential attempts to read AI-generated content. Gaze detection is also helpful in flagging the use of AI to “fix” your gaze somewhere, such as in the case with AI that makes it appear that you are making “eye contact”. By alerting to suspicious gaze patterns or behaviours, gaze detection allows exam administrators to review recorded footage and assess if cheating with AI was attempted, ultimately ensuring a safer exam environment.
Lockdown Browser
What it is: A lockdown browser is a specialised web browser designed to secure the online exam environment by restricting access to unauthorised digital materials and resources.
How it detects AI cheating: By ensuring that test-takers remain within the exam window, the lockdown browser effectively prevents them from browsing the internet, accessing AI-driven tools, or using unauthorised applications during the exam. This heightened level of security minimises the risk of AI-assisted cheating. It helps maintain the integrity and fairness of online assessments, ultimately providing a reliable and equitable testing experience for all students.
Websweep
What it is: A lesser-known feature in online proctoring designed to protect your exam items, this innovative technology continuously scans the internet for exam items being digitally posted or shared outside the authorised assessment platform.
How it detects AI cheating: By detecting and alerting exam administrators of potential leaks or unauthorised dissemination of exam content, WebSweep can assist admins in content takedowns, resulting in less chance of these items can be put into an AI program and answered. By maintaining the confidentiality of exam materials and ensuring that questions remain secure, WebSweep plays a vital role in preserving the fairness of online exams, allowing organisations to administer exams with confidence.
Originality Verification
What it is: Originality software is a plagiarism detection software.
How it detects AI cheating: Originality detection identifies AI-assisted cheating by flagging copied or paraphrased content and assessing citation misuse. These tools analyse submitted work, detecting similarities with sources in their database, which may reveal the use of AI content generators.
These AI proctoring systems can detect any unusual actions or prohibited resource usage, ensuring the assessment’s integrity is upheld.
Plagiarism detection software
Plagiarism detection tools like Turnitin use advanced algorithms and natural language processing to identify plagiarised content. As AI-driven plagiarism becomes more sophisticated, these tools are evolving to differentiate between human and AI-generated content. Platforms like Turnitin continually update their database with human-written essays, making it increasingly challenging for students to submit work generated by AI tools.
Features to watch out for:
Turnitin AI Writing Detection tool
Available for Originality users only, Turnitin’s AI writing detection tool is a new feature (launched in April 2023) that essentially works like the Similarity checker in that on your dashboard, there will be a little icon along the sidebar menu that specifically detects what percentage of a candidate’s essay/long-form answer is AI written.
Software Upgrades for Enhanced Cheating Detection
In addition, at Cirrus, we are considering several native features on the e-assessment platform side that can help detect and dissuade candidates from using AI:
Time Tracking
Tracking a candidate’s time to answer each question can help identify suspicious activity. For example, if a candidate answers complex questions too quickly, it could indicate using AI to answer the question for them, or alternatively, too long to answer a question could indicate trying to collude with other candidates.
Copy-Paste Prevention
E-assessment platforms can disable the copy-paste function or track when candidates use it excessively. This prevents students from copying answers from external sources or pasting AI-generated content into their responses.
Advanced Behavioral Analytics
By analysing patterns in mouse movements, keyboard usage, and on-screen interactions, e-assessment platforms can identify abnormal behaviour indicative of cheating. Unusual patterns in these actions can be flagged for further investigation.
AI Detection vs Prevention:
There are some possible downsides to these detection tools, the most glaring being that they detect AI cheating, but do not prevent AI cheating (other than candidates being aware that there are detection devices employed and potentially deterring cheating, of course).
Resource intensive
Reviewing the reports and alerts generated by these AI tools can be time-consuming for administrators. While AI can flag potential cheating, humans still have to review these instances to make a final decision. This could increase the workload for staff.
The time gap
Being that detection tools record or alert administrators after the exam is already over, the damage is already done and the candidate has already cheated. There is a lag between the act of cheating and when (and if) they get caught.
The “if”
There is no guarantee that the AI will always catch cheating 100%, meaning that it’s entirely possible that the candidate could get away with it. Alternatively, overzealous AI’s could falsely flag something as cheating, thereby punishing an innocent candidate. Over-reliance on AI tools could potentially lead to complacency in educators and administrators, believing that the AI system will catch all instances of cheating when that simply isn’t the case.
But fear not…look out for our next blog on preventing AI cheating, addressing all of these concerns and more.
The rise of AI has made maintaining academic integrity tricky. However, with the adoption of online proctoring like Proctorio, advanced plagiarism detection tools like Turnitin, and software upgrades that enhance cheating detection, organisations can safeguard the credibility of their assessments. By staying vigilant and making sure your e-assessment platform utilises these detection tools, you can ensure a culture of academic honesty in the digital age.
Stay tuned for our webinar on dealing with AI in assessment! Would you like to be notified? Sign up to our monthly newsletter on all things e-assessment below.
