Have you seen pictures of the Pope in the Vatican in a loose white coat on social media platforms these days, or pictures of former US President Donald Trump and the police trying to arrest him dramatically, or pictures of French presidents? Emmanuel Macron between riot police and some protesters surrounded by smoke? Or a picture of Elon Musk walking with the CEO of General Motors (Tesla's competitor)!
— No Context French 🇫🇷 (@nocontextfre) March 21, 2023
But none of these events actually happened, and the images circulating across social media platforms and deceiving users were fake and generated by generative AI tools.
It turns out that many social media users thought the photos were real. Celebrities tweeted that they believed the Pope's soft coat was real because there is nothing in the picture that contradicts logic, many social media users may have thought of a similar coat, and strangely enough, the Pope in the Vatican also has a coat!
The photos also made headlines, and news organizations were quick to debunk the fake photos, particularly Trump, who was indicted but not arrested by a grand jury in Manhattan.
These pictures are definitely fake. But this leads us to a new problem:
Circumstances indicate that we have a big problem on the Internet. A new suite of AI tools is making it cheaper and easier than ever to create photo-realistic images, as well as audio and video, and those images are likely to appear more often on social media.
While these AI tools may lead to new ways of expressing creativity, the proliferation of fake media they generate also has the potential to spread more misinformation online.
This may increase the challenges for users, news outlets and social media platforms to check what is true, as years of fighting disinformation online reveal less sophisticated pictures.
There are also concerns that fake AI images could be used to deceive and harass users of social media platforms, confuse netizens and provoke certain behaviors or incite false hatred between communities.
Claire Leibovitz, head of artificial intelligence at the nonprofit Partnership on Artificial Intelligence, told CNN, "Compared to AI-generated text, images have also gained in popularity recently — thanks to tools like ChatGPT ... — perhaps especially powerful."
Social Media Disclaimer:
Social media platforms have been preparing for this moment for years. They have detailed warnings about fake videos, they know that anyone with video editing software can add politicians' faces to controversial fake photos, and they have policies that restrict or prohibit sharing of manipulated media that could mislead users.
But the recent proliferation of imaging tools based on generative AI models has put platforms like Twitter, Facebook, and TikTok in an unprecedented position.
Images that once would have taken half an hour or an hour to create in Photoshop and similar programs can now be created in five minutes or less with tools like Midjourney, which can create your first 25 images for free, or Stable Diffusion, which is completely free to use.
Rapidly evolving generative AI tools:
AI technology has come a long way recently as some tools have been around for a while eg: the stable diffusion which was used to create images last year but does not produce convincing images like we see in this image today.
Image creation tools have come a long way in the past six months, and the latest version of Midjourney can generate both real and fake images based on short text prompts.
"The latest version of Midjourney is only available to select paying customers," Midjourney CEO David Holz told CNN in an email.
According to a post on Discord (David Holz), Midjourney also suspended the free trial of its previous version this week due to unusual demand and abuse.
According to a post on Discord (David Holz), Midjourney also suspended the free trial of its previous version this week due to unusual demand and abuse.
For the most part, the creators of modern fakes don't seem to be acting maliciously; The image of Trump's arrest was created by Eliot Higgins, co-founder of investigative journalism organization Bellingcat, who explained that when it was posted to Twitter, the images were AI-generated images. "Create pictures of Trump's arrest while we await his arrest," he said.
Are there any solutions against the spread of fake AI images?
Making pictures of Trump getting arrested while waiting for Trump's arrest. pic.twitter.com/4D2QQfUpLZ
— Eliot Higgins (@EliotHiggins) March 20, 2023
Technical solutions are being developed to ensure that content is generated by AI, including: using watermarks in images or embedding transparent tags in image metadata so that anyone viewing them on the Internet knows AI is making them.
The nonprofit (Partnerships in Artificial Intelligence) has developed a number of practices with partners like OpenAI, TikTok, and Adobe, including recommendations on how to spot fake AI images and how companies can share data about them.
Besides these efforts; An open letter signed by Elon Musk and dozens of tech leaders has surfaced, calling for a halt to development of artificial intelligence models stronger than GPT-4 for at least six months until we make sure the exit initiative is full of confidence and its risks under control.
The letter has persisted ever since, urging the government to intervene if all AI labs fail to comply with the moratorium. And it appears that the government is ready to take such a step as the Italian government has decided to prevent ChatGPT from investigating the legality of collecting voice sample data during its creation phase.