By TAIMOOR SOBHAN and KELVIN CHAN, Associated Press
One viral video shows what appears to be four sharks swimming in a Jamaican hotel’s pool as floodwaters allegedly brought on by Hurricane Melissa swamp the area. Another purportedly depicts Jamaica’s Kingston airport completely ravaged by the storm. But neither of these events happened, it’s just AI-generated misinformation circulating on social media as the storm churned across the Caribbean this week.
These videos and others have racked up millions of views on social media platforms, including X, TikTok and Instagram.
Some of the clips appear to be spliced together or based on footage of old disasters. Others appear to be created entirely by AI video generators.
“I am in so many WhatsApp groups and I see all of these videos coming. Many of them are fake,” said Jamaica’s Education Minister Dana Morris Dixon on Monday. “And so we urge you to please listen to the official channels.”
Although it’s common for hoax photos, videos and misinformation to surface during natural disasters, they’re usually debunked quickly. But videos generated by new artificial intelligence tools have taken the problem to a new level by making it easy to create and spread realistic clips.
In this case, the content has been showing up in social media feeds alongside genuine footage shot by local residents and news organizations, sowing confusion among social media users.
Here are a few steps you can take to reduce your chances of getting fooled.
Check for watermarks
Look for a watermark logo indicating that the video was generated by Sora, a text-to-video tool launched by ChatGPT-maker OpenAI, or other AI video generators. These will usually appear in one of the corners of a video or photo.
It is quite easy to remove these logos using third-party tools, so you can also check for blurs, pixelation or discoloration where a watermark should be.
Take a closer look
Look more closely at videos for unclear details. While the sharks-in-pool video appears realistic at first glance, it looks less believable upon closer examination because one of the sharks has a strange shape.
You might see objects that blend together, or details such as lettering on a sign that are garbled, which are telltale signs of AI-generated imagery. Branding is also something to look out for as many platforms are cautious about reproducing specific company logos.
Experts say it’s going to get increasingly harder to tell the difference between reality and deepfakes as the technology improves.
Experts noted that Melissa is the first big natural disaster since OpenAI launched the latest version of its video generation tool Sora last month.
“Now, with the rise of easily accessible and powerful tools like Sora, it has become even easier for bad actors to create and distribute highly convincing synthetic videos,” said Sofia Rubinson, a senior editor at NewsGuard, which analyzes online misinformation.
“In the past, people could often identify fakes through telltale signs like unnatural motion, distorted text, or missing fingers. But as these systems improve, many of those flaws are disappearing, making it increasingly difficult for the average viewer to distinguish AI-generated content from authentic footage.”
Why create deepfakes around a crisis?
AI expert Henry Ajder said most of the hurricane deepfakes he’s seen aren’t inherently political. He suspects it’s “much closer to more traditional kind of click-based content, which is to try and get engagement, to try and get clicks.”
On X, users can get paid based on the amount of engagement their posts get. YouTubers can earn money from ads.
A video that racks up millions of views could earn the creator a few thousand dollars, Ajder said, not bad for the amount of effort needed.
Social media accounts also use videos to expand their follower base in order to promote projects, products or services, Ajder said.
So check who’s posting the video. If the account has a track record of clickbait-style content, be skeptical.
But keep in mind that the people behind deepfake videos aren’t always trying to hide.
“Some creators are just trying to do interesting things using AI that they think are going to get people’s attention,” he said.
So who is behind the account?
While it’s unclear who exactly created the pool shark video, one version found on Instagram carries the watermark for a TikTok account, Yulian_Studios. That account’s TikTok profile describes itself, in Spanish, as a “Content creator with AI visual effects in the Dominican Republic.”
The shark video can’t be found on the account’s page, but it does have another AI-generated clip of an obese man clinging to a palm tree as hurricane winds blow in Jamaica.
Trust your gut
Context matters. Take a beat to consider whether what you’re seeing is plausible. The Poynter journalism website advises that if you see a situation that seems “exaggerated, unrealistic or not in character,” consider that it could be a deepfake.
That includes the audio. AI videos used to come with synthetic voice-overs that had unusual cadence or tone, but newer tools can create synchronized sound that sound realistic.
And if you found it on X, make sure to check whether there’s a community note attached, which is the platform’s user-powered fact-checking tool.
One version of the shark pool video on X comes with a community note that says: “This video footage and the voice used were both created by artificial intelligence, it is not real footage of hurricane Melissa in Jamaica.”
Go to an official source
Don’t just rely on random strangers on the internet for information. The Jamaican government has been posting storm updates and so has the National Hurricane Center.
0 Comments