Pope Francis in unusual clothing, Donald Trump being arrested or Elon Musk sitting in a square in Spain. These are examples of images created with artificial intelligence (AI); At first glance, they seem like real events, so much so that they have cast doubt on thousands of users on social networks, but they are nothing more than fake photos.

“I just thought it would be fun to see the pope in a weird jacket.” Pablo Xavier, the person responsible for the images produced by artificial intelligence that went around the world last weekend, told Buzzfeed News.

Pablo used Midjourney, an artificial intelligence-powered tool that can create super-realistic images. What he did was insert a very descriptive text with the information he wanted to be graphed. “This may be the first real case of mass-level artificial intelligence misinformation,” said web culture expert Ryan Broderick, especially at a time when social media is full of false information.

How do you recognize a fake image created by artificial intelligence?

Brendan Paul Murphy, a professor of digital media at CQUniversity in Australia, has revealed that it is necessary to look critically and observe every detail of the composition of the photo. In the case of the images of Donald Trump’s arrest, you can see some objects move or overlap.

In the case of the pope’s images, one of them shows his hand carrying an object. You can see that the hand is over each other. In addition, the shadow of the glasses betrays that it is a false scene.

fake videos

In addition to photos, there are also videos with digitally created elements in circulation. This is the case of an alleged interview with Bill Gates who abruptly cut off when asked about the COVID-19 vaccines and the alleged damage they had done.

In the original interview, this clip was not seen at any point during the conversation, which did not refer to vaccines. , English subtitles were placed over the faces of the journalist and Gates. This way it was difficult to determine that the lip sync was wrong.

Key to discover untruth

Sam Gregory, disinformation expert and executive director of Witness, an organization that promotes the use of audiovisual technologies to document human rights abuses, explains that if you suspect you’re facing an AI-created image, the first thing to do is “stop, investigate the source, look for alternate cover and trace the original.”

Interviewed by Spain’s RTV, Gregory defines it as the “SIFT method”, for the acronym in English of the four concepts mentioned above. This recommendation is based on stopping researching the context of the image and checking whether it has been published “in other digital contexts” or “is basically completely new”.

It has turned out to be one of the most important elements are the hands, as artificial intelligence tools find it difficult to create this part of the human body. However, programs improve hand design.

Exaggerated facial expressions. “When you ask for expressions, Midjourney tends to exaggerate them, with very pronounced skin folds, for example when smiling,” Bellingcat founder Elliot Higgins explained to Wired magazine.

perfect faces. the images generated with AI have no imperfections like we humans have” and the brightness of the eyes can be “too exaggerated”, said Nuria Oliver of the Ellis Alicante Foundation.