Video of Chinese paraglider Peng Yujiang likely manipulated with AI, experts say

RNZ News

2 days ago

By
Max Walden
and
Gillian Aeria
, ABC
Chinese paraglider Peng Yujiang says he "definitely won't fly for a while" after surviving high altitude accident.
Photo:
CCTV via Reuters via CNN Newsource
Video of a Chinese paraglider widely published by news media outlets around the world was at least partly manipulated by generative artificial intelligence tools, experts said.
Remarkable footage of 55-year-old paraglider Peng Yujiang, face encrusted with ice after purportedly being sucked into the upper atmosphere, was widely covered by Chinese and international media outlets last week including the ABC.
The video had been sourced from Chinese state media outlet CCTV via Reuters.
Reuters distanced itself from the content in a statement.
"The content is clearly labelled as third-party content and is not verified or endorsed by Reuters," the company said.
"When we became aware of a piece of content that likely contained AI-generated elements on the Reuters Connect platform, we investigated and took it down because the material does not comply with our partner content policy."
The ABC also removed the original story and published an editor's note clarifying that Peng's claim that he was lifted 8598 metres could not be independently verified.
Associate Professor Abhinav Dhall from Monash University's Department of Data Science and AI said there were several factors that meant the video was difficult to identify as manipulated, including the low quality of the footage.
"It's difficult to figure out if it's real or fake," he told the ABC.
"[But] if we closely observe the starting say 3 or 4 seconds of this video we can see that the clouds in the background do not really look real. They look like as if, you know, it's a 2D picture."
Dhall said it was difficult even for experts to detect "subtle manipulations" to images or video.
RMIT generative AI and journalism researcher TJ Thomson agreed.
"It didn't seem overly dodgy or suspicious at first glance especially looking at it on a small smartphone screen with our attention frayed," he said.
"But if you look at it more closely, I think things start to come to the surface. You can pick up little things - the colour of the helmet, for example, changing colour.
"Looks like the colour of cord shifts over time, looks like the ground's visible and then not a few seconds later."
'Generational change' forced by generative AI
It is plausible that Peng did indeed reach the height he claimed, despite the video being doctored.
German paraglider Ewa Wisnierska was sucked nearly 10,000 metres into the air above NSW by a storm front in 2007 and lived to tell the tale.
"It could be that they're splicing together multiple clips of actual real footage or that AI has been used here to generate or edit some content," Dr Thomson said.
Still, he said the case demonstrated the risks of generative AI for shrinking newsrooms, which have become ever-more reliant on crowd-sourced material.
"We see 728,000 hours of video being uploaded online every day and more than three billion images being uploaded online every day," he said.
"It can be helpful to integrate that crowd-sourced footage into our reporting, but it's the scale and speed by which that content travels is really hard for journalists to fact-check."
Australia's journalist union, the Media, Entertainment and Arts Alliance (MEAA), told the ABC it had been calling on the federal government to regulate AI.
"Our members have been telling us that they are concerned about misinformation and disinformation and the potential erosion of public trust in journalism and the media," a spokesperson said.
"That's why we have been calling on government to introduce new AI legislation that will govern the use of generative AI, protect creative works from AI theft and increase transparency."
MEAA chief executive Erin Madeley said last year that use of AI generation would "drive the erosion of our news and information to the point where the community cannot tell fact from fiction".
Monash's Dr Dhall said there was a "big, generational change" happening regarding the creative ways in which users could manipulate data - and that there was a role for both industry and government.
"I reckon it will take some time for systems - automatic systems and human observers - to get on page and quickly realise that something is fake or not," he said.
"But there is hope. There is research going on within academia and industry, developing tools that will be able to identify [AI manipulation], along with a human being."
-
ABC