Why the Use of Artificial Intelligence Can Be Dangerous?

Artificial intelligence is already all around us. We use it to create content, in advertising, and in almost every field imaginable. But just how intelligent is AI? Can it mislead us, and to what extent should we trust it without question?

Believe it or not, in the modern world there is probably no one who does not use artificial intelligence on a daily basis. It makes life easier when we are overloaded or short of time, quickly finding information, analysing it and organising it for us. It even helps us in ways we may not realise, when we are not consciously aware that we are working with artificial intelligence at all.

Obreten Obretenov, a cybersecurity expert:

“Everyone has experienced it – when you’re typing a text on your smartphone and suggestions appear at the top for which word to use. Artificial intelligence isn’t some kind of electronic brain that knows everything; it’s exactly that, but on a much larger scale and done far better.”

Artificial intelligence creates systems and models capable of performing tasks. It trains itself hundreds of times faster than humans, using information that is freely available on the internet. According to experts, however, this environment is already saturated with errors.

Obretenov warns:

“A large proportion of the content published online is in fact generated by artificial intelligence. When this material is fed back into AI training models, it effectively contaminates the dataset, and the content produced afterwards contains far more inaccuracies and is, in itself, much less reliable.”

This creates a vicious circle: the more unverified, AI-generated content is published online, the more flawed data artificial intelligence ends up working with. And because AI systems are programmed to assist users, they may present so-called “hallucinations” as verified facts.

Media expert Margarita Pesheva describes the process:

“When a system like ChatGPT, or artificial intelligence in general, is trained using information created by artificial intelligence, it leads to what is known as ‘digital inbreeding’, where the AI itself starts to become less intelligent.”

To prevent this, artificial intelligence needs databases created by humans. These, however, accumulate slowly and are expensive to produce. That is why AI developers seek access to ready-made informational and scientific datasets that are not freely available online.

According to Obretenov:

“The people selecting training material for these models are interested in major publishing houses and large news organisations. There are discussions about what information can be provided, and at what price, for training and education. When models are fed verified data created by real people who provide genuinely high-quality information, the models become more accurate.”

The conclusion is clear: we should not place blind trust in artificial intelligence and should always verify the information it provides. No matter how much it knows, it cannot replicate human critical thinking – at least not yet.