StudyAI: How Generative AI Is Eroding Trust in Online Text, Voices and Video

Generative AI is quickly changing not only how content is produced, but also the basic sense of what can be believed online at all. It used to be that crude fakes and doctored footage were seen as the main threat. Now the problem runs deeper: text, voice and video increasingly look plausible by default, which means the internet itself is losing its status as a place where you can see proof with your own eyes.

The author suggests treating this not as a narrow deepfake problem but as a continuation of an older media logic, one Marshall McLuhan described through the influence of a medium on how a message is perceived. The internet made the spread of information instant, emotional and hard to control. Against that backdrop, traditional authorities have weakened, and any piece of content can easily be pulled out of context and built into a new story.

Even before generative models took off, the web was already fertile ground for disinformation, and with accessible AI the scale and speed of the problem have grown by an order of magnitude. A good example is recontextualization: a real video, photo or quote is moved into an unrelated situation and made to serve a new, false interpretation. Formally, the material may be authentic, but its meaning has been forged.

One of the key effects of the new era is the so-called "liar's dividend." The more realistic synthetic media becomes, the easier it is for a person to reject even genuine incriminating material, a recording or testimony by claiming it is a neural-network fake. The other side of the same problem is "truth apathy."

When users know that almost anything can be faked, their motivation to look into details drops. Instead of checking facts, a defensive mode kicks in: take nothing seriously, scroll on, and don't spend energy distinguishing truth from imitation. This is dangerous not only for news but also for law, reputation, political communication and public trust in general.

The paradox is that the more sophisticated generation tools become, the cheaper it gets not only to produce lies but also to refuse to acknowledge the truth. Text looks especially vulnerable. Video and audio can still be checked through biometric signs of a living person: microscopic changes in skin color linked to breathing and blood flow, or vibrations of the vocal apparatus that are hard to model correctly.

Such methods are not perfect, but they at least point to a direction for technical defense. Text is harder: if a model writes coherently, confidently and in the right style, a reader has almost nothing to rely on except outside context, publication history and the author's reputation. That is why the text environment may be the first to enter a phase where telling human from machine without additional metadata becomes practically impossible.

Hence the growing risk for education, expertise and public debate: synthetic articles, reviews, academic papers and comments will increasingly be taken as ordinary background noise. The piece, however, does not boil down to pessimism. Technological progress creates both the threat and the tools to respond to it.

The logic is simple: fighting AI fakes will probably also require AI and related verification systems. This is not a magic button but a constant race between attack and defense. Some will improve generation; others will improve detection, content provenance verification, biometric markers, cryptographic signatures and trust infrastructure.

There will be no absolute protection, but the article does not predict a complete collapse of reality either: society has adapted to new media environments more than once, changing its information consumption habits and its criteria for credibility. Users will probably have to relearn digital hygiene, and platforms will have to build content provenance checks in as a basic function rather than an option. The main conclusion is that the problem with AI is not only that it can create fakes, but that it blurs the very idea of digital evidence.

In the coming years, what will matter is less the words, images or recordings themselves than the confirmed context of their appearance: who published it, where it was created, whether the chain of provenance can be verified, and whether there are independent signs of authenticity. Otherwise the internet risks turning into an environment of total doubt, where truth technically exists but stops working socially.