Stop the AI: detect texts generated by Chat GPT in bizExaminer

Nearly everyone in the education sector is currently talking about the AI chatbot “ChatGPT” by OpenAI and what implications it has in the evaluation of exams.

The innovative AI technology is a blessing and a curse at the same time. What makes this new approach of Chat GPT special is that it can create text which can hardly be distinguished from human-written ones. This generative AI’ technology will help a lot with the creation of exam content, and make it easier, faster, and more efficient.

Generally, when a generative, text-based AI is used, a query is given to the AI, which generates an answer… in bizExaminer the candidates have to write their answers directly on the screen into a text box.

Should we worry about artificial intelligence in education?

A productive application of AI engines can be found in evaluating written text answers. This works well as long as the exam questions have a specific answer which could be verified automatically, i.e. list all capitals of the EU. Questions which have a free-form text answer, and are targeting the candidate’s knowledge, are much harder to automatically verify through an AI based approach.

While it is not possible to predict where things will go, it is almost certain that generative AI technologies will deeply affect the digital learning and examination sector. What can be seen in terms of individualized & personalized learning and knowledge transfer, as well as AI generated content are only the beginning.

The task with the highest urgency in the field of education currently is to detect when generative AI has been used, most importantly in the context of homework, exams, or other work used for evaluation of the learner. Not to much surprise, a tool has been created. The tool GPTZero is able to detect AI patterns in AI generated texts. However, it is hard to guess whether this option will be viable, and for how long.

How does bizExaminer support a solution?

Generally, when a generative, text-based AI is used, a query is given to the AI, which generates an answer. This answer is then copied and pasted into the exam.

For the question type ‘essay question’ in bizExaminer the candidates have to write their answers directly on the screen into a text box. If the candidate is found to copy paste large parts (or all) of the answer into the text box, bizExaminer detects this event and displays a fraud notification to the person responsible for marking. The final decision on how to evaluate the answer is up to the grader. This would allow the permitted use of AI for exams, and could prevent abuse in a case where the use of AI tools is prohibited.

An additional security level exists when bizExaminer lockdown client or SafeExamBrowser is used for locking down the examination computer. These modes prevent the exam taker from opening any other browser tab besides the one with the exam.

Is artificial examination possible?

It is not yet clear where this AI journey in the education sector is headed. AI based, personalized eLearning, and AI generated education content are surely just the beginning of a much larger, tectonial transformation. In the future, examination could maybe done easily and efficiently by real time conversations between the students and a AI based chatbot, preconfigured on evaluation models calibrated by educators.

On the other hand, it might become near impossible to verify whether a given text – question or answer – has been generated by an AI. We will need new approaches to determine that conclusively.

At bizExaminer, we are following these developments very closely, and are constantly evaluating possible options of how to instrument AI in bizExaminer, and ensure the integrity of our product for our customers.

Not one step back, but two steps ahead

Request a free live demo and discover how easy and stress-free creating, managing and delivering online exams can be for your organization.

Fill out the form and be amazed yourself.