This post is also available in:

‘AI Cheating’ Again… Is This Truly a New Incident?

Following the news that some students shared questions and answers via an anonymous chat room and Google Docs during a non-face-to-face online quiz at Yonsei University, the media once again brought up the term ‘AI cheating.’ At the University of Ulsan, an online mid-term exam consisting of 44 questions to be solved in 50 minutes was ultimately invalidated due to an influx of answers suspected of utilizing Generative AI. There were even reports of high school Korean performance assessments where students used ChatGPT to create plots and critiques, which they then simply copied.

Each time an incident occurs, headlines invariably include terms like “shock,” “great confusion,” and “mass cheating.” The campus and educational field are depicted as if they have suddenly collapsed. However, viewing these incidents from a slight distance suggests that they are closer to the result of allowing time to pass while leaving ‘pre-existing vulnerabilities’ intact after a new technology emerged. Although the method and tool of cheating have changed, the moral and systemic structure has long been present in the educational setting.

The direction of the subsequent discussion changes entirely depending on whether we view the current events merely as “a new crime created by AI” or as “the failure of an assessment system unprepared for a predictable change.” The former tends to flow towards holding students’ ethics and morality responsible, while the latter focuses on fundamentally reviewing the assessment systems designed by universities, schools, and educational authorities. Unfortunately, the discussion thus far has been skewed toward the former.

The Playing Field Was Already Tilted Before AI Intervened in Performance Assessments

The incident where high school students used ChatGPT to write a book’s plot and critique for a Korean performance assessment seems to have shocked many. However, a slight recollection reveals the numerous ways in which performance assessments have been distorted. Cases where high-cost private tutoring services essentially ghostwrite assignments under the name of ‘consulting,’ specialized performance assessment companies target students and parents, or parents are heavily involved in report and project creation are not new. Performance assessment was originally introduced to meticulously observe the student’s process and reflect everyday learning achievements.

However, in reality, it was often operated by merely receiving the final product and assigning grades. This is often due to teachers lacking the time and capacity to meticulously check individual students’ writing processes and actual comprehension. Since only the outcome is reviewed, it is near impossible to confirm who wrote it, what help was received, or whether the student actually understood the content. Even before the advent of AI, performance assessment was criticized as an evaluation heavily influenced by parental privilege and private education.

광고

The crucial point here is that AI didn’t create new cheating but rather acutely exposed the structural vulnerability that performance assessments have harbored for a long time. An assessment that already struggled to ensure fairness is now augmented with a ghostwriting tool accessible to anyone for free or at a low cost. Consequently, the doubt surrounding “Did the student write it themselves?” is no longer an exceptional incident but a shadow cast over the entire performance assessment system.

Assessment Based on Competition and Ranking Structurally Induces Temptation to Cheat

Assessments exist for various reasons, but the most potent function operating in school assessment remains selection and ranking. CSAT and GPA grades, performance assessment scores, and university grades are all interpreted within a competitive structure. A student’s relative position within a class or grade level determines their career path and opportunities. As long as exams function as a tool to differentiate ‘pass and fail’ or ‘top and bottom tiers,’ the temptation to cheat will inevitably always exist.

National examinations are the sector that knows this fact best. The bar exam, national medical examination, and various qualification exams are operated under the premise of strict invigilation and control mechanisms. Metal detectors, prohibition of electronic devices, mandated distance between seats, constant supervisor patrols, and camera installations are all minimal measures established to ensure the fairness of the exam. The more significant the stakes linked to the exam, the more meticulously the system has built safeguards to prevent cheating.

However, realistically, it is impossible to manage every university lecture and every high school subject at this level. We cannot introduce national exam-level surveillance for every class, nor can we change all performance assessments to face-to-face interviews or oral exams. This leaves only two options: one is to strengthen surveillance and punishment, and the other is to change the assessment method itself. The former is unsustainable in terms of cost and administrative burden. Ultimately, what remains is to redesign the content of the assessment to suit the times.

It was reported that in one course at Seoul National University, some students used ChatGPT during a coding midterm conducted on in-classroom computers, despite a pre-announced “No AI Use” rule. As similar incidents followed at Yonsei and Korea Universities, the media emphasized the frame of “students who violated the AI use ban.” This was followed by the diagnosis that “the ethical awareness of today’s university students has collapsed.”

But a crucial question is omitted here: How realistic is the “No AI Use” regulation? Today’s university and high school students already routinely use Generative AI for assignments, translation, summarization, and drafting personal statements. Schools have often promoted the benefits of AI and encouraged its use as part of digital literacy education. Then, suddenly, at the moment of an exam, they say, “You shouldn’t use AI,” without explaining where the line lies between permissible use and cheating. In this contradictory message, what is required of students is closer to an attitude of ‘just knowing better.’ The system strongly demands ethical responsibility without providing clear standards. Simple prohibition rules like “Zero points for using AI” or “F grade if AI traces are found” may temporarily raise awareness. However, as AI increasingly and naturally integrates into various services and functions, the scope of this ban becomes ambiguous, and its enforceability decreases. This is a typical example of norms failing to keep up with reality.

What to Test: From Knowledge Reproduction to Understanding and Thinking

In an era where AI can generate text and solve problems, merely instructing “Don’t use AI” is insufficient to protect the essence of education. The more fundamental question lies in “What should we be testing?” In an environment saturated with searchable information and automated calculations, maintaining assessments centered on memorization and simple reproduction capacity is already losing its social validity. The shift must be toward assessing what students have understood and what judgments they have made in the process, how they verified and modified the AI’s answer, and whether they can apply and critically analyze it, even if they received help from AI. In other words, assessment needs to focus on the thinking process that produces the outcome and the competencies revealed during that process, rather than the outcome itself. This is not merely a new method forced upon us by technology but aligns with the goals education has always aimed for.

Of course, this change is not easy. It requires redesigning exam questions, reorganizing assessment criteria, and enhancing the assessment capabilities of teachers and professors. However, if we resort to repeating only ‘prohibition’ and ‘ethics’ instead of change—because it is difficult and because results are not quickly visible—educational assessment will eventually become an increasingly detached ritual, failing to catch up with social changes.

Only Process-Visible Assessment Survives the AI Era

In an era where AI can generate text and answers, judging a student’s learning based only on the submitted outcome is virtually impossible. This applies to assignments, performance assessments, and online exams alike. The assessment design must now center on the criterion of ‘how transparently the process can be revealed.’

For example, consider a writing assignment. If the current method of submitting only a completed essay is maintained, it is difficult for the teacher to verify if the writing was done with AI assistance, guided by a tutor, essentially co-written by a parent, or genuinely written solely by the student through their own thought process. The situation changes if students are required to submit multiple stages of documents—such as a rough draft, a second version, a revised draft after feedback, and the final copy—along with a brief record of the student’s choices and reasons for modification at each stage. It is also possible to use a portion of class time to have students directly perform part of the writing or problem-solving process and compare it with the subsequently submitted outcome. Short oral interviews or presentations where students explain the text they wrote or the problem they solved can also be used. While this takes more time and increases the burden on the assessor, only such methods serve as a minimal mechanism to verify a student’s actual comprehension in the AI era. The transition to process-oriented assessment is closer to a necessity than an option.

The Responsibilities Universities and Schools Must Prioritize

In the current ‘AI cheating’ controversy, the term that most easily disappears is “systemic responsibility.” However, the responsibility for failing to manage foreseeable change first lies with the system designers. Nearly two years have passed since ChatGPT was publicly released, and during that time, numerous universities, research institutions, and companies have discussed AI utilization, ethics, and assessment changes. Despite this, many universities and schools have barely altered their assignment, exam, and performance assessment operational methods.

The first thing universities and schools must do is self-examine whether the assessments they designed are still valid in the AI era, before questioning students’ ethics. If non-face-to-face exams are to be maintained, concrete principles must be established regarding how to reduce cheating; if performance assessments are to continue, how process verification mechanisms can be secured; and if AI is to be partially permitted, where the line is drawn between ‘learning tool’ and ‘cheating.’ In this process, the responsibility should not be solely placed on individual teachers and professors. Assessment design and operation are issues that transcend the level of individual courses. University-wide guidelines, departmental and school support, and structural adjustments to reduce the workload of teachers must be discussed together. Otherwise, the demand to “change assessment methods to prevent AI cheating” will only come across as another pressure on already overworked teachers and professors in the field. Assessment reform can only be implemented in reality when supported and restructured at the policy and systemic level.

Following the high school performance assessment incident, the Ministry of Education announced plans to distribute a “Guideline for Safe AI Introduction and Utilization in Schools” next March. Following successive incidents at Seoul National, Yonsei, Korea, and Ulsan Universities, plans to urgently devise “Measures to Prevent AI Cheating” were also mentioned. These responses are late but necessary. The problem, however, is that if this guideline remains merely a declarative statement, such as “No AI use during exams” or “Must disclose AI use on assignments,” the current confusion will not change significantly. The media also needs to focus deeper on the structural limitations revealed by the system and how unprepared universities, schools, and educational authorities were for the AI era, rather than framing students’ actions purely in moral terms. Some articles already point out that the rapid spread of non-face-to-face classes created a vacuum in management and supervision, and that traditional exams and assignments are virtually unrealistic in liberal arts courses with hundreds or even thousands of students. However, such analysis remains in the minority.

What is needed now is a new frame that centers on the “epochal inadequacy of the assessment system,” instead of the familiar narrative of “the collapse of student morality.” Only then can policies, school settings, and university assessment systems gradually shift direction. And in that process, the voices of students can also be invited not merely as ‘objects of regulation’ but as agents collaborating in the discussion of new assessment methods.

The AI cheating controversy superficially appears to be about student misconduct, but a much larger question lies beneath it. In a society where AI is commonplace, are we still administering tests and grading in the ways of the past? Are we relying on easily manageable methods and outdated conventions instead of assessments that genuinely reflect student learning? Shifting the responsibility for unpreparedness for predictable change onto students’ lack of ethics and morality is choosing the easiest solution in the name of education. But the easy solution usually leaves the fundamental problem unresolved, inviting crises in other forms. The ‘AI cheating’ situation we now face is the same. Technology will continue to evolve, and AI will integrate more naturally into our learning and work. If we merely repeat, “This time we will prohibit it more strictly” and “We will punish it more severely,” educational assessment will increasingly remain a norm detached from reality.

Now, the questions must change. We must shift focus from “Why did students cheat with AI?” to “What assessment structure tempted such a choice?” from “How to prohibit AI?” to “What and how should we assess in a world where AI exists?” and from “What ethics should we demand of students?” to “What responsibility should the system and schools take first?” Only when this transition occurs can educational assessment in the AI era provide a convincing answer to students, teachers, schools, and society alike.

#AICheating #GenerativeAI #NonFaceToFaceExam #PerformanceAssessment #UniversityEducation #EducationalAssessment #LearningFairness #AIEducationGuidelines #SpotlightU