Fact check: Spotting AI-generated ‘historical’ images

In Italy’s Pompeii, archaeologists have, for the first time, used artificial intelligence to reconstruct the image of a victim of the volcanic eruption in 79 C.E.

AI is increasingly being used to generate “historical” images as the technology has become widely available, mostly without the consultation of experts. Roland Meyer, a media scholar at the University of Zurich, thinks the trend is no coincidence.  

“Through AI, it seems possible to recombine images of the past, generate synthetic images that do not exist but perhaps could have existed, and thus fill perceived or actual gaps in the archive and in the historical record,” he said. “To me, however, this seems like a deceptive promise.”

AI generators have been known to make mistakes and “hallucinate.” DW Fact check explains how to spot AI-created “historical” videos.

#1: Could this video even exist?  

Many viral AI‑generated “historical” videos that claim to show the past blur the line between fact and entertainment. They can often be identified by simply asking whether the video could have been made. 

“Are we talking about a time when videos already existed, or are we talking about a time when they didn’t?” asked historian and AI-specialist Sebastian Kubon.

This video supposedly depicting Pompeii can be discredited on that basis alone.  

It also shows clear signs of AI generation. People move in a zombie-like manner, and the woman’s appearance looks far more modern than what we know of ancient Rome.  

The rise of AI tools has made it easier than ever to create such “historical” images and videos, but experts warn that AI is trained on existing data and will repeat its biases. 

“The images of the past generated by AI primarily reproduce the typical patterns, cliches and stereotypes with which that past has been depicted thus far,” said Meyer.

“Above all, these are images of our expectations: not a window into the past, but a mirror of our present.” 

#2: Look for visual clues and check the source 

Secondly, a close look at the images themselves and their sources is helpful.  

In this video, the camera glides in closer to a group of soldiers holding the US flag, showing their faces as they stare into the distance. It looks more like a Hollywood film than authentic documentary footage.  

One soldier in the background glitches unnaturally. The watermark also gives it away: this video was produced by Veo, a generative AI tool.  

The video mentioned above is based on the iconic 1945 photograph of US soldiers on Iwo Jima. There is no authentic video footage of this specific moment.

All the experts DW spoke with emphasized the importance of critical source analysis. Where does the image come from, who presents it, for whom and why?

The video’s account promises to explore “patriotic stories” and “educational and entertaining content that brings America’s story to life.” It reveals a handful of similar AI-generated videos. 

Images and videos always present a particular perspective, Kubon pointed out. 

“They involve a selection of image fragments and audio sources; in essence, they present one perspective on the story. We’re too quick to assume that just because something was recorded, it’s somehow the truth.”

Jürgen Zimmerer, an Africa history professor at the University of Hamburg, warned that visual sources still carry a great deal of authority and have been problematic even before AI.

“Images have always been highly controversial sources. This is evident from the fact that dictatorial regimes, whether Nazi Germany or Stalinist Russia, have also retouched images,” he said.  

The difference today is the scale at which AI can generate altered and completely fabricated scenes. 

#3: Compare with verified historical sources  

Look to archives, museums, reputable media or academic experts.  

A video showing alleged CCTV footage of the Chernobyl nuclear disaster in 1986 is also AI-generated. Whereas we can find the authentic photograph of Iwo Jima in the US National Archives, no authentic footage exists from this moment in Chernobyl. 

Such AI creations raise the question of responsibility.  

In comparison to the use of fictional scenes in documentaries, where a director or production team decide how to portray the past, AI pulls from large datasets without understanding of historical context.  

“It produces something that looks plausible at first glance, not because it is historically accurate, but because it matches our expectations,” Meyer explained.

This is most dangerous when authentic footage does exist, and the video could be real.  

Whereas most viewers understand that videos from ancient Rome are fake, large numbers of AI-generated images threaten to replace real documentation of times the technology was present.

Zimmerer emphasized that history is always an approximation. “We approach historical events when there is no original film footage, and when you pretend there is, you make something seem clear that is not actually clear,” he said.

So, don’t fall for AI-generated historical slop. Dig deeper instead and explore the real past.