[DigitalToday reporter Yoonseo Lee] Testimony from university classrooms after the spread of generative artificial intelligence (AI) said professors' roles are tilting more toward detecting misconduct than teaching.

On April 13, local time, IT outlet Ars Technica said a university instructor who has taught asynchronous online classes pointed out that student assessment and assignment management have effectively become an entirely different job since ChatGPT emerged.

The instructor said recorded online classes were already prone to student disengagement. Now more unmotivated students are submitting AI-generated work that looks like an assignment, rather than skipping assignments altogether. Unlike simple plagiarism, it is hard to clearly determine whether a large language model (LLM) was used, the instructor said, adding that the burden of justification has grown in case of disputes. The instructor also said that while plagiarism used to be a question of whether something was misconduct, instructors now have to judge like detectives.

The problem, the instructor said, is that the learning process itself breaks down. Students can easily mistake the goal of education for submitting correct answers or earning grades, and essays and descriptive answers are about the thinking process rather than the result, the instructor said, adding that if an LLM takes over there is nothing left for students. The instructor compared writing an essay with an LLM to bringing a forklift into a gym, saying it may lift the weight but it does not build fitness.

The instructor said the shift is being seen in real classrooms. One question the instructor has given students since 2019 requires them to extend their thinking based on lecture content, and before ChatGPT about 30 percent of students answered correctly. Over the past two years, the correct-answer rate has exceeded half, the instructor said. The instructor said certain repeated phrasing that appears when the same question is entered into ChatGPT also often shows up in student answers. Students may see it as a simple information search, the instructor said, but typing a question and copying the response is no different from copy and paste.

The change is also weakening the meaning of formative assessments, used to check learning progress. Simple quizzes or descriptive assignments are a tool to confirm whether students properly understand concepts. But if these are left to an LLM, it can become a waste of time for both instructors and students. Recently, agent-type browsers have reached the level of processing quizzes for an entire course at once.

Universities are responding by bringing back oral exams and proctored handwritten tests. But such methods are effectively hard to apply in asynchronous online classes. Online classes matter for students with disabilities, students living far from campus, and students juggling full-time work or caregiving. The instructor said if online classes are abandoned, the harm will fall on these students first.

The situation is similar in in-person classes. The instructor said adaptations to prevent LLM-related misconduct often end up lowering the quality of education. Activities that used to be effective writing assignments are disappearing first. As one example, students were once asked in a class to write a Hollywood-style disaster movie plot mixing scientific and unrealistic elements, but the instructor said the assignment can no longer be maintained because an LLM can generate a similar piece in 10 seconds.

Faculty concerns are also significant. In a survey of about 3,000 university instructors, 85 percent said LLMs weaken the development of students' critical thinking skills, and 72 percent said they are having difficulty managing LLM use. Even so, higher education often tells instructors to teach students how to use AI effectively. The instructor said many such class cases end up having students critique AI-written text, but the goal is to make students realise they should not leave writing to an LLM.

The fatigue felt by teachers in the field is different from narratives about technology anxiety. The view is strong that it is not a problem that can coexist by restricting some functions, like debates over introducing calculators, but one that shakes almost all areas of education. Concerns are growing that today's AI, rather than innovating education and strengthening learning, is making activities that have long supported student learning much harder to carry out.