Images from ancient Pompeii, animated scenes from World War II or alleged surveillance recordings from the Chernobyl nuclear disaster: artificial intelligence (AI) is increasingly used to generate supposedly historical images and videos.

For media researcher Roland Meyer, this trend is no coincidence. “It seems possible, through AI, to reconstruct images of the past, generate synthetic images that do not exist but perhaps could have existed, and thus fill supposed or real ‘gaps’ in the archive, in the historical record. It seems to me a misleading promise.”

DW Verifies shows how to recognize “historical” AI-generated videos, how they perpetuate clichés, and how they appeal to our emotions.

Many signs of AI involvement

On the one hand, AI-generated content is becoming easier to identify: through watermarks, references on social media platforms or community notes on X.

On the other hand, AI tools are constantly evolving, AI-generated videos are becoming more sophisticated, and watermarks indicating their use are often removed. Therefore, critical image analysis remains essential.

This TikTok video, showing Pompeii before, during and after the eruption of Vesuvius that destroyed the city, is labeled as AI-generated on YouTube. During the eruption of Vesuvius in 79 AD. C., the Roman city was buried under volcanic ash and lava.

The video presents typical AI characteristics: people move like zombies and the protagonist has a seductive appearance that does not correspond to the typical image of women in Rome at that time.

Great moments, great images

This video from Xwhich supposedly shows surveillance footage from the 1986 Chernobyl nuclear disaster, was also AI-generated. Although it does not have a watermark, it was labeled as AI-generated content in the annotations below in the Community Notes section.

Another example: this YouTube video was produced by Veo, a generative AI tool. The camera zooms in on a group of soldiers holding the American flag, showing their faces as they look into the distance.

The video is based on the iconic photograph of the flag-raising ceremony on the Japanese island of Iwo Jima. The photo, taken by Joe Rosenthal on February 23, 1945, became an iconic symbol of the US Army’s victory over Japan and one of the most recognizable images of World War II.

The authentic photograph of Iwo Jima is located in the United States National Archives. However, there is no authentic video of that moment.

The YouTube account that posted the AI-generated video promises to feature “patriotic stories” as well as “informative and entertaining content that brings American history to life.”

Are AI fakes crowding out real sources?

The video is based on a real historical event, but adds details and perspectives that are not historically documented and presents them in a patriotic way. Therefore, it is useful to carefully examine the images and their sources. Where does the image come from, who presents it, for whom and why?

Many AI-generated viral videos that purport to show the past sit in a space between reality and entertainment. While these videos may spark interest in historical events. Experts warn that they give the past a certainty that often did not exist. “With each representation of a specific scene, other possible interpretations that also exist are erased,” says historian Jürgen Zimmerer.

For viewers, therefore, it is important to recognize that this is an AI-generated “historical” video and then analyze the representation with the necessary critical distance.

“With images of ancient Rome, we know they are fictional; however, with images from the Nazi era, the large number of AI-generated forgeries threatens to displace authentic documents, especially since AI-generated images adapt to our expectations and viewing habits,” media researcher Meyer also warns.

Clichés and stereotypes

AI models use both authentic sources and staged representations to create these images. The data sets used to train the AI ​​include historical photographs, but also, for example, digitized paintings, movie scenes and video game images.

“The images of the past generated by AI therefore mainly reproduce the typical patterns, clichés and stereotypes with which this past has been represented until now. They are, above all, images of our expectations: not a window to the past, but a mirror of our present,” notes Meyer.

These AI creations also raise the question of liability. Unlike the use of recreated scenes in documentaries, where, for example, directors or producers consciously decide how the past should be represented, AI relies on large data sets.

And it does so without understanding the historical context. “It produces something that seems plausible at first glance, not because it is historically accurate, but because it meets our expectations,” explains Meyer.

AI renderings in collaboration with experts

The use of AI to create “historical” images is therefore controversial. However, there are also examples where AI representations are created in collaboration with experts.

For example, in Pompeii, Italy, archaeologists used artificial intelligence for the first time to reconstruct the appearance of a victim of the 79 AD volcanic eruption. C. and thus reconstruct the period around the year 800 AD. c.

The representation is based on archaeological findings and aims to make the research results more vivid; However, it is understood as a scientific reconstruction and not as an authentic representation.

(rml/ms)