“You can no longer believe in ANYTHING that comes from a digital device. Contrast everything”, says journalist Robert Scoble in a tweet that accompanies an alleged image of Elon Musk holding the hand of US politician Alexandria Ocasio-Cortez, known as AOC. It is one of the many alert messages that have been seen in a few days full of fakes In Internet. Images of Donald Trump arrested and his eventual escape from prison and Pope Francis in a sophisticated Balenciaga feather coat, the false and hyper-realistic images that have circulated on the networks since last week have shown that information is entering a new era.
This is your regular warning. You can not believe ANYTHING anymore that is coming from a digital device.
Validate everything. https://t.co/lF0y4vbKZw
—Robert Scoble (@Scobleizer) March 26, 2023
Generative artificial intelligence (AI) is already sufficiently developed to make us fall, at least to the naked eye, into hoaxes that even almost nothing required complex technologies and expert users in digital manipulation. Today, preventing us from being fooled by screens will be increasingly difficult, according to experts. The blockchain is presented as a complex and long-term solution, but for now there is only one remedy of a lifetime: check the sources, pay attention to details and doubt everything.
It’s not the first time that images generated by the Midjourney AI tool have gone viral, but none have been as widespread as Balenciaga’s Pope. First, because it seems much more plausible than the other assumptions of people who had an artificial appearance, as if they were from a video game or with many filters. Also because the context that surrounds the pontiff can suppose that he would perhaps dress in such a way.
However, when doing zoom, you can notice the size of the ear, the severed hand that is not quite grasping the coffee cup, the deformity of the glasses or the crucifix where you cannot see Jesus Christ carved and without part of the chain that supports it. These details indicate that it is not a real photograph, but rather qualify the flaws of the AI: a tool that knows the surface of reality well, but not how physical objects interact or all the characteristics of the human body. At least for now.
Even so, they are factors that go unnoticed, especially if you are scrolling fast and in the mobile format. In some images of the false arrest of Trump, which circulated on the networks a few days before, the evidence that it is a fake they were more noticeable. The first aspect is also contextual: the arrest of the former US president would be covered in the mainstream media, and it was not. Second, as they had already affirmed some artistsgenerative tools lack the ability to render details of the human body, especially the hands. In some images of Trump, the proportions of his body can look distorted or even melted and there is a blurry appearance. In others, distorted texts are observed.
But these tools are improving and it is a matter of time before they stop failing in these details and start creating false images that seem extremely real, according to specialists. The new one midjourney version it is already capable of generating realistic human hands, eliminating what until now was the easiest way to identify an artificial image.
The speed at which improvements occur is what raises concerns. Elena Verdú, a member of the Artificial Intelligence and Robotics research group at the International University of La Rioja, explained to Maldita.es that just one month is enough for some of these recommendations to identify false images to become obsolete. We are facing a collective concern about the power that these tools exercise over society. So much so, that artificial intelligence experts demanded in an open letter on Wednesday that the “race without control” of the ChatGPT, which also uses generative AI, be stopped for six months.
Journalist Robert Scoble explains in the tweet Other ways to identify deepfakes. The first thing is to check with multiple reputable sources and verify their credibility. Also value the context. “Use critical thinking: Analyze the information you receive and consider the context, consistency, and logic behind it,” suggests the author.
When the content goes viral, it is also useful to check what can be seen on the internet. Google, for example, provides a reverse image search tool, where it is possible to upload a photo and check where it has already been shared and what people are saying about it. If a photo supposedly taken by a photojournalist was first posted online by an anonymous stranger, that’s a big reason to be suspicious.
On Twitter there is also the option to classify content as false, although it is necessary for many people to do so for there to be an alert in the publication. According to collect magazine Timesthere is a range of software in the market that claims to be able to detect the deepfakes. However, there are few or no reliable free tools available to the end user. One of the detectors, which is hosted on the Hugging Face AI platform, was able to say “with 69% certainty” that the pope’s image was AI-generated. But when presented with the alleged image of Musk with OAC, the tool failed to answer that the photo was 54% true.
Other experts talk about how the blockchain could help combat them. Known for being the technology behind cryptocurrencies like Bitcoin, the World Economic Forum noted its ability to provide validation of authenticity and a clear chain of custody that “makes it potentially effective” for tracking and checking all kinds of content, not just financial. The key is that this technology has a mechanism that does not allow the alteration of the message, nor the moment of publication or the origin. But it is not a solution for the present: from theory to practice, there is still a long way to go.
The article points out that the blockchain also has its limitations. While it is capable of verifying the existence of a document, it cannot prove intellectual property, for example. Furthermore, to be truly effective for the end user, it will need to be integrated into the chips that power smartphones and computers. Something that also depends on a cohesion between international communities, governments, companies and civil society to shape a governance model for the consumption of digital content.