AI‑generated imagery ushers in a new era of digital deception

When AI image and video generators became widely available, their creations largely became the subject of jokes and ridicule. Indeed, the early attempts looked like the drunken dreams of a robot that did not fully understand what a human should look like. Strange, tiny teeth, often too many, or an inappropriate number of fingers, sometimes even extra hands or legs, were just some of the flaws in early AI-generated graphics.

A video depicting actor Will Smith eating pasta became famous. It looked like something we might see in a horror movie. However, a year later, AI has no problem generating reasonably realistic-looking people who are eating:

The rapid development of artificial intelligence is best seen in the images that recently circulated on the internet. Until now, even if a person was rendered well, small details allowed us to recognise AI – mistakes in the background, clothing, or eyes. The pictures generated using Lora AI software amazed people because, at first glance, it was virtually impossible to distinguish them from photographs. Even with closer analysis, it's hard to definitively point out something that indicates they were not taken by a photographer and edited:

Videos soon followed, and the limitations of AI became more visible. While very realistic, there is still something unnatural in the facial expressions. But would someone immediately recognise that a video was AI-generated if it included voice and subtitles and was watched on a small smartphone screen? Not.

It only took several months for artificial intelligence to cross the Rubicon of realism—and there's no turning back. Observing this development, I have two feelings: on one hand, I am amazed and admiring the creators because the pace of development of these tools is incredible. On the other hand, I am apprehensive because we cannot guarantee the ethical use of AI.

Thousands of social media profiles use generated graphics to gain popularity, but often also to spread misinformation. The same goes for videos. The more realistic the effects, the easier it will be to manipulate public opinion, especially in politics. Moreover, widely known figures and those who influence public opinion, such as Elon Musk or some politicians, spread AI-generated fake videos.

The problem is that they often don't label them, so less informed people or those without time to verify might believe in the presented content and form an opinion based on it. Elon Musk was sharply criticised for sharing a video pretending to be a campaign spot for Democratic presidential candidate Kamala Harris.

Just last Thursday, Donald Trump posted on his profile on the platform a video showing him and Elon Musk dancing to a Bee Gees song. It's clearly AI-generated work that isn't very well done, but at this pace of development, in six months, we'll likely see the same, but already looking completely realistic.

It's worth noting that some of these tools are truly within anyone's reach. Sure, creating a realistic image – like the ones shown a few paragraphs above – still takes effort. You have to pay attention to details. Sometimes, you need to fine-tune what the AI generates. But a few days ago, Elon's platform released version 2.0 of its AI called Grok. Using it, I generated the following images – it took me 3 minutes. It didn't require effort; I just instructed Grok to create a "realistic photograph."

At first glance, it's hard to distinguish them from stock photos, especially the first graphic. The only detail I think reveals AI involvement is the man's imperfect thumb – something isn't right. But unless someone looks closely, there aren't significant differences. In the second case, it's a bit easier because although the graphic looks realistic, there are at least some strange solutions for human cities in the details – like a tunnel leading nowhere in the foreground.

This shows how quickly the boundary of realism shifts with AI-generated images – moving or still. There are countless scenarios where someone uses fake, AI-created videos to damage a political rival's career or a business competitor. There's also been much talk about so-called "revenge porn", where real people's faces are inserted into pornographic films. The cheaper and better the technology, the easier it is for anyone to take revenge, threaten, or ruin someone.

Can we defend against this? Yes, in one crucial way: not believing everything we see online, especially on social media. Try to verify and seek sources, especially for content depicting politicians or situations with a political context. This is where manipulators looking for applause or manipulating public opinion harvest the most. Social media is the most effective place for spreading disinformation because videos and photos spread across Facebook, Instagram, or TikTok at lightning speed.

The European Union recently adopted the AI Act, a regulation designed to mitigate the risks of artificial intelligence. However, I don't believe they will effectively protect us from AI-generated "fakes." The problem encompasses not only images but also audio, and people can already encounter scams where someone's voice is generated using AI to sound like their loved ones. Regulations can't keep up with the pace of AI development and will be very hard to enforce.

I must emphasise that I am not against technological development, nor do I believe the world will collapse soon because of artificial intelligence. People once feared electricity, then the internet; today, these fears can be related to AI. Like with many technologies, we need to educate and hope to learn to use this technology responsibly and with perspective.

For now, we must realise how rapid the progress is. We must be mentally prepared for the imminent inundation of images and videos that are not real and whose sole purpose is to deceive and manipulate us – pushing us towards one politician or another or to extort money or data from us. This is a threat we must learn to live with.