Indonesian Political, Business & Finance News

AI Hallucinations and the Threat to Scientific Integrity

| Source: Antara (Indonesian) Translated from Indonesian | Technology
AI Hallucinations and the Threat to Scientific Integrity
Image: Antara (Indonesian)

AI is a useful tool, but it is not a source of truth that should be accepted without checking.

Jakarta (ANTARA) – A student is preparing a research proposal with the help of artificial intelligence (AI). Within seconds, AI produces ten references complete with authors’ names, publication years, article titles, journal names and digital object identifier (DOI) numbers.

The student feels that the work has become easier. But when several references are checked on publishers’ websites, the articles in question cannot be found. Some DOIs also lead to articles with different titles and authors.

How can AI produce apparently complete scholarly references that were never actually published?

This phenomenon is known as AI hallucination: when an AI system generates incorrect or unsupported information but presents it in convincing language.

In academia, such errors can affect literature searches, proposal preparation, data interpretation and the writing of scientific manuscripts.

Large language models (LLMs), such as ChatGPT, are trained on enormous collections of text. They learn language patterns, relationships between concepts and information contained in their training data.

When given a question, AI generates an answer based on the patterns it has learnt and the context provided by the user. This enables it to explain complex concepts and produce well-structured writing.

The problem is that the ability to produce plausible sentences is not always accompanied by checks on the accuracy of the information. AI can combine correct facts with inaccurate information, or provide an answer when there is not enough evidence.

Kalai and colleagues (2025), in their study Why Language Models Hallucinate, explain that hallucinations are linked to the training and evaluation of language models. Assessment systems that reward correct answers more than admissions of uncertainty can encourage a model to guess when it does not know the answer.

This is similar to a student facing an exam question without knowing the answer. Some choose not to respond, while others take a guess. The difference is that AI’s guesses are often delivered fluently and with technical terms that sound scholarly.

Non-existent references

References are an important foundation of research. Through previous publications, researchers learn about developments in their field, the methods used and questions that remain unanswered.

For example, a student researching volcanic soils might ask AI to find publications on the relationship between non-crystalline minerals and phosphate retention in Andisols. AI may produce a list of articles complete with journal names and DOIs.

Some of those references may be genuine, while others may combine authors’ names, titles and bibliographic details from different publications.

I have encountered a similar case when checking the reference lists of students’ research proposals and undergraduate theses. My name was listed as the author of several scientific articles, even though I had never written or published them.

That experience shows why references must be checked for authenticity before they are included in scholarly work. Bibliographic errors do not always originate with AI, but the technology can increase the risk when its answers are accepted without verification.

AI-generated citations have also attracted the attention of the international scientific community, including in discussions published by the journal Nature in 2025 and 2026.

The presence of a DOI does not automatically guarantee that a reference is correct. The identifier should be checked through Crossref or the publisher’s website to confirm that the title, author names, year and publication details match.

Data interpretation

More from 6 November 2026 →

Cite this article
Source: jawawa.id (https://jawawa.id/newsitem/ai-hallucinations-and-the-threat-to-scientific-integrity-1793965900)
Originally published by Antara (Indonesian) (https://www.antaranews.com/berita/5836831/halusinasi-ai-dan-ancaman-integritas-ilmiah), translated from Indonesian
Print