Hatena
Has the Drop in Inquiries About 'Non-Existent Papers' Really Been Solved? A Report on the 'Academic Information Distribution and Library Services in the AI Era' Session at Library Fair 2026
AIFirst seen 41 points
A library trade show forum in Japan looked at how generative AI is changing academic information. Speakers asked whether fewer fake-paper questions means the problem is solved, or just harder to see.
What happened
At the 28th Library Fair 2026 in Yokohama, INFOSTA (a Japanese information science group) joined a forum called "Academic Information Distribution and Library Services in the AI Era." Three speakers from research, publishing, and university libraries described what generative AI has changed. The forum centered on one question: AI looks smarter than ever, so has academic information become more fragile?
Key numbers and facts
- Almost all students now use generative AI for assignments.
- More than 80% of researchers use AI somewhere in their work.
- About 10% of students start research at a library or database. About 40% start with an AI chat or AI research tool.
- At Niigata University, AI-written explainers were published for 850 Japanese papers and 350 English papers.
- Making a fake citation takes seconds. Correcting it can take about 550 days (median).
Background: the "fake paper" problem
Generative AI sometimes invents papers that do not exist, called hallucinations. Libraries saw many requests to check whether such papers were in their collections. Publishers also saw a new pattern called the "crocodile effect": search results are shown more often, but fewer people click through to the original paper. The two lines spread apart like a crocodile's open jaws.
Why it matters
The speakers said the problem has not gone away. It has moved. AI hallucinations may be going down, but fake citations hidden in real papers are rising. Speakers said the real issue is the system that checks and fixes errors, not only the AI models. Librarians and information workers need to teach new skills, like judging whether AI answers are reliable, and to help people check sources.
What to watch
- Results from the "Taming the Crocodile" research project, which studies how people find information through AI.
- Whether publishers and libraries develop new ways to measure use, beyond downloads, such as citations.
- Whether more Japanese libraries adopt AI licenses that let them build research tools in closed, university-only AI systems.
- How fast journals remove papers that contain fake citations.