They say a picture speaks a thousand words. And they also say that you should always trust your eyes. But, in front of those eyes today, thousands of pictures and videos are shown daily in the media and on social networks. Insignificant content is mixed with valuable content, mimes with scenes from cult films, authentic footage with those generated by artificial intelligence.
So, should we still believe our eyes? Sandra Ristovska (39) does not have a simple answer to that question. Ristovska, born in Macedonia, is today an associate professor of media at the University of Colorado in Boulder, near Denver. She founded the Visual Evidence Laboratory there and that is her main topic - how we see the same footage differently and under what conditions footage can be valuable evidence in court.
Ristovska received her doctorate in the field of communications, before that she studied film. At the beginning of the conversation with "Vreme", he says that all those references and extensive experience are no longer enough to recognize which video is "real" and which was generated by artificial intelligence (AI).
"Before, I could do it. Because there were clues that indicated that something was strange," she says.
"WEATHER" Nowadays, it is more difficult to notice traces in the generated video?
SANDRA RISTOVSKA: Yes. They haven't completely disappeared, but it's hard to tell when you see something on social media. Partly because today much of the content on the networks is visual rather than textual. We usually scroll quickly and if something looks good and realistic, we usually take it to be real, to be authentic. Today we should stop and take a closer look. Admittedly, there is no golden rule to recognize a fake. That's why I don't even believe in my ability to always recognize VI's work.
Ristovska recalls an example from the end of last year. Jan Lekun, chief engineer at artificial intelligence company Meta, shared a video on social media of a New York City police officer angrily chasing away Immigration Service agents. The video turned out to be generated by artificial intelligence. Lekun deleted the post and did not comment on it.
And if you have all the time in the world and you can use various software... can you then recognize the work VI?
Not always either. Deepfakes were weaker before. But technology is advancing rapidly, so the detection tools we use today may not work tomorrow.
Ristovska gives an example from Myanmar. After the military coup, 2021. a confusing video appeared. It shows how Pyo Min Tein, the then arrested first man of the Yangon region, in custody he confesses and claims that he bribed the ousted politician and Nobel laureate Aung San Suu Kyi. In Myanmar, many doubted that he had seen the work of VI.
Many experts set out to verify this. In court, it would have to be proven beyond a reasonable doubt - well, they couldn't prove it so convincingly. So maybe the video was authentic and maybe it wasn't. It cannot be said. Therein lies the complexity.
I am interested in how we believe images and how we derive meaning from them. For that, the technical characteristics of the image are important - but that's where things just begin. Because the context in which images are shared and how they are discussed is also important. In the example of Myanmar, even if the video itself is authentic, it does not mean that the arrested politician was giving an honest testimony. Maybe he was tricked into telling it, maybe he's lying from his account. So a video can be authentically inauthentic.
When you talk about how hard it is to identify fakes, it sounds to me like the average user of social networks has no chance.
We are all in the same boat – ordinary people, experts, judges, even the police force. That's why the institutions that deal with establishing the facts - the legal system, and even journalists - have an increasingly important role to responsibly deal with any event that is documented with pictures and videos.
Ristovska moves on to fresh examples from Minneapolis. There, in January, officers of the Immigration Service killed two citizens, Renee Goode and Alex Pretty. Goode was shot in her car, through the windshield, when she tried to walk away from the officer. The assailant was shot in the back when he had already been overpowered and the gun had already been taken from him, who was wearing a seat belt all the time and for which he had a license. Both murders were captured on multiple cell phone cameras, but fake videos produced by artificial intelligence were soon circulating.
In Renee Goode's case, we had video of her hitting a police officer with her car, which didn't happen at all. In the case of Alex Pretty, people were trying to use VI to "unmask" police officers who were in action and had phantoms. Of course, it is impossible to "take off" their mask like that. Artificial intelligence will then "guess" and produce a face, but it will not be the authentic face of the specific police officer. That's why experts had to explain in the media that "removing" the mask is impossible and dangerous - that way, a person whose face "resembles" the one created by artificial intelligence could be targeted.
But here's the good news - despite this deluge of content, polls show that most Americans have seen mostly authentic footage and therefore believe the police actions were unjustified. Even President Donald Trump retracted his initial statements about "domestic terrorists" - and started talking about "unfortunate incidents". I believe that Trump retorted because the public understood what happened. The truth prevailed over the lies, the authentic recordings suppressed the fake ones. So maybe VI won't stop us from believing our eyes and ears.
......
Image manipulation has a long history. Forwards, VI enables exactly that. "flooding" which you were talking about.
Yes, it is the last straw in the history of technologies that people have used to manipulate images and videos. The difference is that previously specific knowledge was required.
William Allison had been an Iowa state senator for 35 years when he was elected in 1908. won another Republican nomination to extend his term. He defeated Albert Cummins in the internal party elections. Even though Cummins' team flooded voters' mailboxes with a retouched photo of Allison. The photo has been altered to make the interim senator appear even older and very sick. Message - he cannot push through another term. Although he used a coaster, Cummins was right. - in just a few months he became a senator after allison actually died.
Today we might be able to conclude that it is a retouched photo, but we have had a hundred years to learn that. And in 1908, it was beyond the comprehension of ordinary people, so it had great power. But at that time such fraud was expensive - a dark room was needed to develop photos, a certain knowledge was needed, it was necessary to put it all in mailboxes - therefore, it required time and money.
Twenty years ago, we were afraid of Photoshop, we thought that photography as a document was no longer worth anything. That's why in 2012 in New York the company Adobe, which created Photoshop, financed a large exhibition dedicated to the manipulation of photographs throughout history. They wanted to prove that manipulation is as old as photography. And Photoshop had to be learned, you remember, not everyone could sit down at the computer and edit a photo in an instant so that it looks realistic.
Today, manipulation no longer requires skills or resources. Anyone with a smartphone or computer can select a generative VI app, type in a simple prompt—that is, a task, a command—and create a video that looks convincing but is actually there to deceive. He can then share this on social media with the potential for the fake to be seen by millions of people around the world. So, the scale is without historical precedent.
But the basic question is an old one - how do we, as a society, separate reality from fiction. Hope can be given by the fact that the problem itself is not new and that we are afraid of it for the umpteenth time in history. We were always struggling to get to the truth and facts, so we somehow got through those chaos.
And again, people like to believe what they already believe. You have done a lot of research as far as our biography, skin color or political beliefs influence what we upload to a video. Does VI help people guard against their prejudices?
Whenever the human factor comes into play, if we see technology as a panacea for underlying problems, then we only risk inflating the problem we are trying to solve. Decades of research show that our identities greatly influence how we see images. Even the speed at which the video is played has an effect, and especially social factors, whether I identify with the police, whether I'm left-wing or right-wing, that shapes us all. There are also cognitive factors such as selective attention.
In a famous experiment from 1999. people were asked to watch a short video showing people in white and black t-shirts tossing a basketball. The subjects were asked to count the passes of one team or the other, or to count how many balls are passed through the air, and how much about the floor. At one point, a person completely masked as a gorilla passes through the frame. About 50 percent of respondents did not even notice it - because they focused their attention on the ball.
Such influences, therefore, already exist in authentic videos, while artificial intelligence only deepens them. The media, courts or government agencies have not yet figured out how to treat authentic images, and now the technology that creates fake images has been grafted. So the challenges will only get bigger.
Fortunately, there are also reverse examples where VI helps a lot and makes the impossible possible. For example, with the help of technology, human rights groups were able to review thousands of videos from Syria looking for e.g. for a specific type of weapon or ammunition, to reconstruct where it came from. VI could be "trained" to recognize it, and then flesh and blood experts checked further, interviewed people, searched for information. It is one of the most positive examples.
Something else is happening. - truth becomes optional. A notorious example is the Trump administration's misuse of artificial intelligence. And when she gets caught, he says - everything is a joke, me. And nothing to anyone.
After the rise of generic VI, experts started talking about the "liar's dividend". The very existence of artificial intelligence offers a convenient excuse for liars, that is, those who want to deceive the public - if a video, for example, does not go in their favor, they can always say that maybe artificial intelligence created it. And vice versa, if they themselves share the generated video and are caught in it, then they say that everything is a joke, that authenticity is not even important. That may go well with the part of the audience that is politically eligible. But it also contributes to such a normalization of VI that people no longer know what to believe.
Hannah Arendt was worried about a similar thing even after the Second World War even though the technology was nowhere near today's. If officials fabricate facts, the point is not that everyone believes them and falls on their faces before the lies. Rather, the point is for the public to be so skeptical of everything that fundamental consensuses, general truths that are necessary for society to function, are lost. To believe in nothing.
There is no magic wand that can make it go away. But the public has to show itself up to it. I go back to the example of the Minneapolis shooting. Public pressure still has power, we have to believe that.
In one paper, you even warn that courts should not rely lightly on video-evidence...
Courts have to use video, but I would like them to have a guide on how to use it very clearly. Of course, the video is an important piece of evidence, although there are problems with it, but there are also problems with testimonies. When we hear someone testify, I know that what we are talking about is filtered through his memory and experience, and sometimes the experiences are traumatic. But of course the testimony is important, it must be heard.
With video, people think they have a direct transmission of truth, that we are part of the scene that the video is showing. For a long time, we thought that the recordings were so objective that we would all see and load the same thing in them. But today we know that is by no means true.
Ristovska gives an example of the historic decision of the US Supreme Court from 2007. years. A young black man in Georgia drove too fast and then ran away from the police who wanted to stop him. The policeman hit the fugitive's car with his car to stop him - the young man remained a quadriplegic. The argument of the police was that driving was dangerous to the lives of other road users and that the fugitive had to be stopped, even at the risk of his life. The chase was caught on camera from a police car. Lower courts thought the case had to go to a jury. Forwards, The Supreme Court is with 8:1 decided that the case should not go to court because "there is no reasonable juror" who would, watching the video, could vote to convict the police officer. In their opinion, the fugitive was obviously driving recklessly and dangerously for innocent people on the road. The ruling is historic because it is the first time the court has decided to publish a video on the Internet. The researchers then showed it to hundreds of subjects - although most agreed that the young man was driving dangerously, there were many of those (potentialy "reasonable jurors") who saw the matter differently. And that depended a lot on skin color, origins and attitudes towards the police.
Therefore, we cannot at all assume that everyone who sees the video has the same interpretation. That's why I think the courts should be given instructions on how to understand the recordings.
So far, if a person discovered something, no matter how devastating it is, there was usually no going back, like an atomic bomb... Can artificial intelligence be reined in or regulated?
When you open Pandora's box, it's hard to put things back into it. There are many financial interests in the field of artificial intelligence. Here in the United States, both schools and universities are being pushed to implement VI. I think you should put on the brakes for a while and see where VI is really useful for something. The tendency of Silicon Valley is to develop a product, release it into circulation, and then iron out problems that users notice along the way. In the case of artificial intelligence it can be very dangerous, but that's where we are. After the invention of the automobile, it took time to invent the seat belt. Some think we'll find a seat belt for VI, but I'm not sure at all. We need to now rigorously debate what benefits society and where the risks are.
Maybe soon the news will be like, for example, craft products or food - the mass will consume what the stamp VI, and only some will care that the news was made "by hand", from man to man.
I hope we don't reach such a dystopia. Maybe naive, but I believe that truth is still important to people. People around the world are taking to the streets to protest against their governments. They vote, they believe in some democratic processes. All of us, including journalists and institutions, should insist that the reality is one and that everyone should see it.
You belong to the niche of experts who deal with this topic in the USA. How did you get such a life path??
I grew up in the former Yugoslavia, during the breakup of the country. It was there that I realized the power of images before I knew what specific images meant. Cartoons were interrupted by news from the war, refugee columns. I saw adults who wanted to protect me from those images, I knew they were "not for me", but I understood the power.
Later I wanted to make films. I was dissatisfied that films from the area of Yugoslavia were either about wars or orientalizing us... Yes, well, we had a war, but life did not stop, there are other stories as well. Our generation felt that they were not part of the war, we were young, we were growing up. And we couldn't see ourselves in art and on film. I wanted something more and that's how it started. I graduated in film, but then I wanted to do more research. I am creating exhibitions about videos and evidence with Daria Medić, a media artist and designer from Belgrade who works at the University of Denver.
When we watch a movie, it may be interesting that we interpret it differently, but it is not interesting when it comes to court proceedings. My background in film helped me understand the production of images and what was really at stake here.
