Regulation of AI-generated content: Humans are lagging behind

The business with computer-generated influencers is booming. This creates legal and ethical problems.

Soon images like this will be produced using artificial intelligence Photo: abacapress/imago

Artificially created personalities have been around for several years. They have thousands of fans online and earn a lot of money for their creators. For example, there are Sika Moon and Aitana Lopez, influencers in sexy outfits with pink hair, brought to virtual life by Spanish designer Rubén Cruz. Or Shudu, a black model who is a construct of the fashion photographer Cameron-James Wilson. Or Lil Miquela, Instagram star and musician who comes from a start-up in Los Angeles.

Artificial intelligence is revolutionizing business, especially in the beauty industry. This is because the AI ​​can now also edit images. It enhances existing photos with filters and montages into beauty shots or generates completely new looks without there ever being an original.

AI is used wherever paying customers want to be flooded with new content in front of their screens – and often secretly long for ideals that no real human can live up to. Where influencers keep their audience and sponsors happy with extraordinary images and videos; always in a good mood, always good-looking, always active and on the go, traveling, doing sports, cooking, dancing, on the sofa with the cat.

The more followers, i.e. the larger the audience, the more the advertisers pay. They make the influencer life possible. Micro-influencers with a few thousand followers earn around 50 euros per post, while mega-influencers with millions of fans get 15,000 euros.

Authenticity and illusion

Modeling with the use of artificial intelligence works like this: A good image needs a good prompt, i.e. a good instruction. Then it’s the AI’s turn. The more precisely she can define the relationship between word and image, the better the result will be. Many companies train their programs to do just that. They need tons of image data for this. They primarily use those that are publicly accessible. So the photos and clips that we show every day on Instagram and Tiktok. AI learns quickly, and the boundaries between human- and computer-generated content are blurring.

While some influencers promote products, places and events, others promote themselves. Adult-friendly content ends up on common social networks such as Instagram and Tiktok, while more explicit content ends up on Onlyfans or Fanvue. Pornographic content is also permitted there. Customers create an account and pay for the providers’ subscriptions. It costs just under ten euros a month to, for example, chat with Sika Moon or see her pictures and videos.

Everyone can decide for themselves how authentically human or illusively artificial the content should be. It is still somewhat clear where and how much AI is behind it. Where this is no longer possible, mandatory labeling could help. Camera manufacturers such as Canon and Nikon will in future add digital watermarks to the images taken by their cameras. But it will also be necessary for AI-supported programs to subsequently recognize which content was created by AI.

Because the next revolution in porn offerings is already imminent: AI-generated videos will soon be good enough to compete with “real” porn clips. AI-generated moving images have been around for a long time. But it still looks a little scary when an AI Will Smith devours AI pasta or five-legged puppies come out of other dogs.

In order for an AI to create videos that look like they were shot by a film camera, it has to take a number of factors into account: physical properties have to be right, as well as light and color, and one image has to logically follow the previous one. This requires larger amounts of data than the AI ​​needs for texts. Open AI, the company that amazed us with ChatGPT, is already on the move: its video AI Sora should be available to everyone in the next few months. The first promo videos look shockingly real.

Fear of deepfakes

Even if AI cannot yet create perfect videos, there are already problems: deepfakes that look sufficiently real. For example, faces of real people are inserted into existing images. Technically speaking, this is possible for all videos. However, according to cybersecurity firm Deeptrace, 96 percent of all deepfakes online are porn.

First, a number of actresses and singers became victims of the reputation-damaging manipulation. After Scarlett Johansson and Taylor Swift, private individuals, ex-girlfriends and colleagues were hit. With the first deepfakes from 2017, the deception was still easy to recognize. Seven years later it’s even harder.

The protection of personality applies de jure on the Internet, but not de facto. Those responsible are located abroad or there are technical hurdles in the way of enforcing laws and penalties. The EU Parliament recently passed the “AI Act”. The guidelines contained therein are intended to protect against dangers, including deceptively real fake images. Europe is reacting faster than other continents in the world. But the bureaucratic apparatus cannot be faster than artificial intelligence. She must follow the side effects of her evolution.

The big difference between man-made and AI-generated pornography will be that the people involved will, in the best case, have to agree to the resulting images, says Lori Watson from Washington University. This is not necessary for an AI without consciousness. In her text about ethics in pornography and sex work, the philosophy professor writes that AI will create a dynamic in which we get exactly the sex we ask for – faster than we would develop an ethical understanding of it.