1. Changes in evaluation methods
➀ Assessing based on process: Propose dividing the assignment into stages for completion.
- As process-centered evaluation checks every step of the assignments up to the final result, it can prevent issues such as plagiarism and the absence of fact-checking that may arise from using ChatGPT without sufficient planning for the final result.
➁ Open-book tests, oral evaluations, and non-evaluative evaluations during class
- Using non-evaluative evaluation methods, such as peer review, self-evaluation, and mutual evaluation, which do not reflect on grades but can improve the learning process and outcomes, can prevent plagiarism by having students review each other's assignments.
③ Evaluating critical thinking skills
-
An assessment method can be utilized by posing questions for students to analyze and evaluate the validity or inadequacy of ChatGPT's responses based on the information they have gathered through ChatGPT.
(e.g., having computer science students find flaws in code generated by ChatGPT)
2. Control and Management of the testing environment
➀ Conducting offline exams
-
By conducting exams in an offline environment where it is easier to control the process of students writing their answers, cheating rates can be significantly reduced compared to online exams and assignment evaluations.
Additionally, a fair evaluation environment can be provided by strictly supervising the exams.
(e.g., requiring the power-off or prohibition of electronic devices such as cell phones, tablets, and smartwatches during the exam, and placing all belongings, except for ID, writing utensils, and watches (including outerwear), in a bag in an empty space in front or behind the examination room)
➁ Avoiding online exams and assignment evaluations
③ Checking ChatGPT's answers in advance during the exam question creation process
3. Utilizing AI Detection Programs
AI-generated text detection programs can be used to check for the use of ChatGPT, Bing AI, etc.
- AI detection programs: GPT Zero, Detect GPT, Originality.AI, etc.
However, even when using AI detection programs, it is not possible to be certain that the written text was created entirely through generative AI like ChatGPT.
A review of the detection program's results by the professor is necessary.
As an example, an AI text classifier released by OpenAI, the developer of ChatGPT, correctly identified 26% of AI-generated texts in exam evaluations as 'possibly AI-generated' and mistakenly labeled 9% of human-written texts as 'AI-generated texts.'
How to use AI Detect Tools
It is recommended to avoid online exams and assignment evaluations which are relatively easy to cheat by using ChatGPT.
For conducting such of that exams, it is advised to create questions that require not only knowledge acquisition but also multidimensional thinking to measure students' academic achievement.
Additionally, if conducting online exams, it is crucial to strictly supervise the process, including requiring students to record their screens simultaneously with camera monitoring.
After creating exam questions, you can check the content of ChatGPT's answers in advance to review whether the exam questions are appropriate for evaluating students' academic achievements.