AI as a Weapon in Media Disputes: Expert Warns of Risks from "Hallucinations" and Ideological Attacks

In the era of rapid technological transformation, accusations of using artificial intelligence (AI) are beginning to turn into a new tool for media and ideological attacks. Linguist and data expert Nikola Tulechki commented on BNR "Hristo Botev" about how these tools are changing journalism and what the real risks of their application are.

AI as a Tool for Discreditation

According to Tulechki, accusations against journalists that their texts are written by AI are often not based on facts, but serve as a "proxy" for ideological attacks. The goal is not to prove a lack of authorship, but to undermine trust in the media or a specific colleague by attacking their ability to write independently.

How to Recognize the Machine in the Text?

Although software tools for detecting AI content are often unreliable, the expert points to specific stylistic markers that give away the algorithms:

• Perfect grammar and spelling: Unlike humans, AI rarely makes typos or grammatical errors.
• Specific structure: Use of "inverse antitheses" (for example: "This is not just a text, this is the beginning of a new era").
• Convoluted sentences: Machines often use overly smooth, but sometimes artificially structured phrases.

The Risk of "Hallucinations" and Editorial Responsibility

The greatest danger in using AI is so-called "hallucinations" – a situation in which the model generates completely fabricated facts, names, dates, or sources, presenting them with a high degree of credibility. Tulechki emphasizes that this imposes a new form of editorial responsibility.

"The human becomes a kind of editor of what the AI produces. We must bear full responsibility for the factuality, because the machine does not possess context for the present and can combine data from the past in a wrong way," he notes.

The Future of Automation

Instead of fighting against technology, society must learn to use it adequately. Tulechki gives the example of Bulgarian Wikipedia, where the controlled use of AI allowed for the generation of 600 high-quality articles on biological species, while an uncontrolled flow of machine text creates unnecessary work for editors. The key lies in assessing the quality and reliability of the content, rather than the mere presence of AI in the writing process.