Palestinians in the rubble of destroyed buildings hit by Israeli missiles in the center of Khan Younis, southern Gaza, on Monday, Oct. 30, 2023. Authorities in Hamas-run Gaza said the death toll since the war erupted had surpassed 8,000, while US President Joe Biden and his Egyptian counterpart agreed Palestinians shouldn’t be displaced to Egypt. (Ahmad Salem/Bloomberg)
(Bloomberg) -- An image of a bloodied corpse. A photo of a crying baby edited to look as if it were lying in rubble. Shots of what at first appears to be an entire neighborhood leveled in Gaza.
There’s little doubt that deepfakes - images and videos digitally altered and spread spread to form false narratives - have been deployed by both sides in the war between Israel and Hamas. How profoundly they’ve worked to convince the masses on social media, shape public opinion, influenced decisions and proliferate, as many feared, with the use of generative artificial intelligence tools remains less clear.
The majority of the fake photos that have emerged on platforms including X and Facebook since Hamas attacked Israel two weeks ago are what Henry Ajder, an expert on deepfakes and generative artificial intelligence, call “shock and awe” images.
The creators intended for them “very emotive, very sensational” and to elicit immediate emotional responses, he said.
For instance, one image on X - identified as fake by a deepfake detection tool created by a professor at the University of California at Berkeley - purports to show two children hugging each other amid the rubble, the young girl wrapped in a Palestinian flag.
“Not necessarily a long talk or a really nuanced picture being painted, but a bloodied corpse, an injured baby, destroyed buildings,” Ajder said. “Things that can spread easily that are instantly recognizable that you don’t have to actually listen to or engage with that much.” Those types of deepfakes surprised Ajder and other experts, who were used to seeing deepfakes being deployed of top officials uttering statements they never made, as has been the case in US politics and the war in Ukraine.
Once crude and easy to detect, deepfakes have gotten substantially more realistic and easier to make with the advent of generative AI tools. But at a time when false claims on social media have become the norm in political campaigns, wars and other major events, a proliferation of hard-to-detect deepfake images threatens to make it even more difficult for consumers to form opinions based on actual facts.
“Poorly contextualized, unreliable deep detection” is “more harmful than if there were no detection at all,” Ajder said. “It gives the illusion of there being a path to truth and, and a decisive assessment when actually, as we know it is not the case.” The “liar’s dividend is arguably more pernicious,” Ajder added.Five accounts that belong to a network aligned with Hamas on X have regularly claimed real footage from Israel was AI-generated as a means to discredit it, according to an analysis by AI company Accrete for Bloomberg News.
In one recent example, a startup called Optic labeled a picture shared by the Israeli government as a fake that showed a torched baby’s remains, leading to widespread skepticism online. Hany Farid, the UC Berkeley professor, however, maintains that the picture is real.
Optic didn’t respond to a message seeking comment.
“The impact of generative AI goes beyond just the creation of fake content, it is also casting a long shadow on content that is real,” Farid said.
Andy Carvin, a senior fellow at the Atlantic Council’s Digital Forensic Research Lab, said it’s not unusual to go to the comments on Facebook and X and, under real images, see people challenging their authenticity or the context. “Others will just say, ‘Deepfake!’ and that’s it. End of conversation,” he said.
Technology created by Berkeley’s Farid has detected deepfakes on X of numerous images of children amid damaged buildings. He said it was hard to quantify how many are circulating but he is receiving requests to identify whether an image is real or not daily, adding, “There have been a lot. I see this issue less as a volume issue and more as an information pollution issue.”
Carvin, from the Atlantic Council, said deepfake images are being posted online by accounts linked to Israelis and Palestinians. “We’ve seen examples of people producing images to rally around a particular cause, and these often include children,” he said. Generative AI tools usually have restrictions on creating violent content, Carvin said, but he explained that it’s simple to ask one of them to create an image of children standing in the middle of rubble, after an airstrike.
Only one of the five images on X identified as deepfakes by Bloomberg News, using Farid’s technology, had a label warning that it wasn’t real, despite the social media network’s policy that bans users from sharing synthetic and manipulated media that may confuse people and “lead to harm.”
Deepfakes relating to the Israel and Hamas war are appearing in Arabic speaking news outlets, in addition to popping up on social media, according to a Middle East-focused analyst at the Atlantic Council’s DFRL who asked not to be named for security reasons. One bogus image - purporting to show a refugee camp for Israelis in occupied Palestinian territories - has been shared multiple times in Arabic-speaking news outlets, including one headquartered in Yemen and another headquartered in Dubai. The image had tell-tale signs of AI, with the Israeli flag showing a double Star of David, and some of the tents appearing misshapen, according to the analyst.