A trickle of essays generated through artificial intelligence were entered in the School Library Association’s contest last year, but educators fear such cheating could become a deluge with no effective countermeasures in place.

For the nationwide contest, elementary, junior high and high school students entered more than 2.6 million essays about books they read during the summer vacation.

Teachers noticed something odd with more than 10 of the essays, and the students who submitted them admitted to inappropriately using AI chatbots. Their essays were subsequently dropped from the competition.

The next essay contest is scheduled for autumn, and officials fear more AI-generated works may escape detection.

Students apply for the contest through their schools, and teachers read the essays before they can advance to the next stage of the selection process.

The association does not impose a blanket ban on AI tools for the contest.

“Our stance is that generative AI can be educationally beneficial, depending on how the technology is used,” an association official said.

The association’s guidelines for applicants forbid plagiarism and inappropriate citations.

For example, essays that include content copied and pasted directly from AI-generated text constitute an “inappropriate” use of the technology, according to the association.

As AI chatbots have become readily available to anyone, the association since last fiscal year has warned teachers about potentially problematic use of the tools.

The association has conducted experiments to spot artificially generated texts, but an official at its secretariat said AI technology remains ahead of detection.

“Each time we have AI write essays, it produces different content,” the official said. “That made it difficult for us to distinguish hallmarks of material made by AI. Besides, the generative technology has advanced dramatically since last year.”

The association said teachers who are familiar with the writing abilities of their students are the most reliable party to check the authenticity of essays.

“If teachers cannot find cheats, I am afraid that nobody can,” the official said.

Another essay contest for high school students faces similar challenges.

More than 20,000 essays were submitted to the competition last year. In a questionnaire after the contest, some members of the judging panel said they suspected several entries were manufactured by AI, according to the organizer.

The organizer said it could not conclude whether the essays were made with AI because it could not confirm the suspicions with the authors. But the judges did not allow such essays to move to the next stage.

In last year’s contest, the organizer did not spell out its policy concerning computer-generated texts. It believes that generative AI is useful in terms of brainstorming and proofreading.

For the contest this year, the organizer will touch on “inappropriate use” of AI models.

“Fairness is critical,” an official with the organizer said. “To get ahead, some students emphasize their prize-winning essays during the admissions process for high school or college.”

However, many organizers agree that current technology to distinguish between text composed by humans from those generated by AI remains unreliable.

In guidelines published last year, the education ministry said children from elementary to high school level should become aware of the importance of learning how to use AI tools.

But it also pointed out that it would be unacceptable for students to submit contest entries that have been “outsourced” to AI.

ARTWORKS COPIED

The use of AI art generators is also a source of concern.

In fiscal 2023, Trident College, a technical school in Nagoya that teaches skills for such endeavors as creating computer games, graphics and illustrations, accepted entries to its competition for 2-D and 3-D illustrations for high school students around Japan.

It forbade the use of AI models.

The school is affiliated with Kawaijuku Educational Institution, a major cram school chain.

Trident College students well versed in the field of illustrations were enlisted to weed out suspicious entries from the initial screening.

Experts say works created by AI generators tend to mimic existing art and specific works.

None of the entries showed telltale signs of AI-generated art, according to the school.

But school officials said the visual checks by humans may not be effective much longer.

“If a user abuses an AI tool and trains it to create works that do not look like specific art pieces, we may not be able to recognize AI-generated content,” said Yuji Kojima, a lecturer at the school.

SOLUTIONS SOUGHT

Businesses are working on programs to determine if something was made with AI or by humans.

Kawaijuku in March said it has developed a program to detect computer-generated text. It is part of a larger system that has been offered to universities since fiscal 2018 mainly for handling admissions processes.

Kawaijuku said the new program has the potential for wider applications, including screening entries for essay contests and evaluating reports produced by college students.

When used to evaluate essay questions in the admissions process, the program is designed to issue a warning when it detects suspicious texts.

“We feel there is a great need out there for a tool that recognizes artificially generated texts,” a Kawaijuku official said.

The official quoted a university staffer as saying that AI may have been used to generate material required for admissions.

“Our program is just a reference, but it may be useful as part of the overall assessment of applicants,” the official said.

Kazunori Sato, associate professor specializing in media literacy and information education at Shinshu University, stressed the need for adults to employ AI tools to understand such creations.

“Essays using AI models tend to have generic content and lack originality,” he said. “Adults, including educators, should routinely experiment with AI to learn patterns of computer-generated texts.”

He noted the use of generative AI has been permitted in academic circles.

Sato said the rule requiring researchers to report their AI use and for which parts of their studies has been widely accepted so that reviewers can carefully check for potential copyright issues.

The same approach can be applied to essays and other contests, he said.

Some contest organizers invite works created by generative AI.

Last year, Gunma Prefecture allowed elementary, junior high and high school students to use AI-created works in a manga and anime event as well as in a contest of computer graphics and games.

While permitting the use of AI models, the prefecture cautions students against potential copyright infringements.
Entrants are obliged to declare which AI tool was used in which part of their works and in what manner.

Nineteen entries used AI, but none won prizes, according to the prefecture.

“We believe that generative AI will grow to be an indispensable tool in the future,” a prefectural official in charge of the events said. “We hope that participating in these events will nurture students’ ability to use AI as we work to minimize the risk of copyright issues.”

(This article was written by Kohei Kano and Yukihito Takahama.)