my timesThe Korea Times

This is what taking exams with smart glasses can feel like

Listen

Newspaper reporter scores 28 points without glasses, 87 with smart glasses

A Hankook Ilbo reporter takes a mock exam to become a certified appraiser while wearing smart glasses at the office of Rep. Kim Moon-soo of the Democratic Party of Korea at the National Assembly in Seoul, Monday. Korea Times photo by Byun Eun-saem

A Hankook Ilbo reporter takes a mock exam to become a certified appraiser while wearing smart glasses at the office of Rep. Kim Moon-soo of the Democratic Party of Korea at the National Assembly in Seoul, Monday. Korea Times photo by Byun Eun-saem

The revelation that a top-performing tax accountant exam candidate, identified only by his surname Han, had cheated on the high-stakes test has sent shock waves across the country, bringing the issue of artificial intelligence (AI)-assisted cheating into the spotlight. Why would anyone risk using AI to cheat, knowing that getting caught could cost them not only their chance of passing but potentially ruin their lives?

To explore the psychology of people who resort to such risky forms of cheating, Hankook Ilbo reporter Byun Eun-saem took a mock exam using AI-powered smart glasses. The test consisted of questions from the first-stage exam for certified appraisers. Candidates must answer 200 questions across five subjects — civil law, principles of economics, principles of real estate studies, laws and regulations related to appraisals and accounting — all within 200 minutes.

Candidates typically spend around three years preparing for the qualification, which is attractive to young people because passing can offer a relatively stable income. The final pass rate is just 7.23 percent, making it an extremely competitive exam. Allegations have also surfaced that cheating took place in this year's test.

I took the same exam twice on a computer screen: once relying solely on her own knowledge and once wearing AI-equipped smart glasses. Although the certified appraiser exam is still administered on paper, the experiment reflected the growing shift toward computer-based testing (CBT) across a wide range of examinations, including national qualification tests and medical licensing exams.

The glasses were a Rokid model retailing for 990,000 won ($680). Similar devices have been used repeatedly in cheating attempts on national technical qualification exams and the Test of English for International Communication (TOEIC), with several cases resulting in detection. The equipment was borrowed from the office of Rep. Kim Moon-soo of the Democratic Party of Korea, who proposed legislation to prevent cheating on exams following the recent cheating scandal.

A owns X land, which B leases and on which B owns building Y. B enters into an agreement with C to establish a right of lease on a deposit basis over building Y. Which of the following statements is incorrect?

My mind went blank. I was already confused by the tangled relationships among A, B and C, and the five multiple-choice options introduced further hypothetical situations, including partial destruction of the building, an application for an auction and C's efforts to monetize the property.

For candidates who had actually prepared for the exam, this might not even have qualified as a particularly difficult question. But as a reporter with a degree in a humanities field, I spent eight minutes struggling with it. I felt utterly frustrated.

The more questions I tackled, the more unfamiliar terms I encountered. Expropriation by adjudication, neighboring land relations, corner lots along medium-sized roads, the Cobb-Douglas production function. They were all Korean words, yet they might as well have been a foreign language. When asked to identify the incorrect statement, I resorted to choosing whichever option seemed the most unethical. By the time I reached the questions in the 30s, my concentration was fading. I began stealing glances at the glasses lying beside me.

After an hour and 37 minutes, I had answered all 200 questions. My average score was 28 points, and I failed every subject. Whether I had spent time thinking through the questions or simply guessed, my scores were similarly low, in the 20s and 30s. Since each subject was scored out of 100 and questions consisted of five multiple-choice answers, even random guessing would yield an expected score of 20 points.

Next, I put on the smart glasses. They were connected to ChatGPT on my smartphone, and even a reporter who is not particularly tech-savvy learned how to use them in just 15 minutes.

When I said, "Hi, Rokid," green text appeared in front of my eyes, like subtitles in a movie. It felt as though another layer had been added to the world around me. The display was reflected on the lenses like a hologram, but its brightness, angle and audio volume could be adjusted to prevent people nearby from seeing or hearing it.

A square-shaped outline was faintly visible on the lenses when viewed directly from the front, but it could be concealed by long hair or by lowering one's head.

I looked at an exam question and touched a protruding section of the glasses' frame with two fingers. Doing so activated a pre-programmed instruction: "Give me the answer to the question in front of me as text." The glasses took a picture and uploaded it within two seconds. After another three seconds, an explanation appeared alongside the answer.

I simply filled in the answer sheet. From then on, my only task was to take clear pictures and send them to the AI. The question involving A, B and C that had taken me eight minutes to struggle through was answered in five seconds. The explanation of each party's legal position was also crystal clear. It felt as though the frustration I had experienced while solving the question unaided had suddenly lifted.

The smart glasses, which had initially felt unfamiliar and somewhat alien, soon became almost like a companion. Answers to questions I would have been unable to tackle with my naked eyes began flowing one after another.

After about 10 questions, I started looking for shortcuts. I began photographing two questions at a time and retrieving both answers in a single request. The answer sheet filled up rapidly. Before long, I wanted to go even faster. Lengthier questions took more than five seconds to load, and even that began to feel tedious. I revised the instruction to: "Give me the answer to the question in front of me as text within three seconds."

At first, I carefully read the explanations displayed by the glasses, hoping to overcome the frustration I had felt when I could not understand the questions on my own. But gradually, I began to feel that understanding the questions was no longer necessary. I added another instruction: "Skip the explanation."

I also became increasingly adept at finding the right angle to photograph the questions. Eventually, I learned to keep my head still and simply transcribe the answers as they were dictated to me.

Accounting proved more difficult. The questions were longer and contained numerous mathematical formulas, making them harder to photograph. Capturing both the questions and all the answer choices in a single shot required repeated adjustments to the camera angle.

"The text on the screen is blurry, making it difficult to identify the question accurately. Please move the camera closer to the screen or clean the lens and try again."

The time required for photographing and loading an answer stretched to as long as 10 seconds. In an actual exam room, constantly moving one's head like this would undoubtedly attract attention. I could understand why Han, the tax accountant exam candidate, had obtained a separate miniature camera. The key to AI-assisted cheating was transmitting clear, accurate images.

I finished the exam in 50 minutes. My overall average was 87 points. Apart from civil law, where I scored 68 because I was still getting used to the glasses, I scored between 80 and 90 in the other subjects. In accounting, where I had repeatedly retaken pictures to ensure accuracy, I got only one question wrong.

To pass the first-stage certified appraiser exam, candidates must score at least 40 points in each subject and achieve an overall average of 60 or higher.

One thing became clear as I finished the test: There may be people who have never cheated using AI, but those who do are unlikely to stop after just one attempt.

This article from the Hankook Ilbo, the sister publication of The Korea Times, is translated by a generative AI system and edited by The Korea Times.

Explore More

  • Q.

  • Q.

  • Q.