The Will Smith Concert video highlights concerns about the AI ​​counterfeit crowd: NPR

Will Smith Concert video I've recently tore the internet – not for his performance, but for the crowd. Viewers of Eagle's eyes noticed strange fingers and faces suspected of AI manipulation in the audience, amid a visual glitches.

Cloud scenes present certain technical challenges in AI image creation tools, especially videos. (Smith's team has not publicly commented on requests about NPR about how to make videos, or has not responded to requests for NPR.) “You're managing so many complicated details,” says San Francisco-based visual artist and researcher Kyt Janae, who manages AI image creation expert Kyt Janae. “You have individual humans in the crowd. They all move independently and have unique features like hair, faces, hats, phones, shirts, etc.”

However, the latest AI video generation models like Google VEO 3 And Openai's Sora 2 It's become quite good. “We're moving into a world where reality boundaries are really blurred with estimates of generous time for a year,” Yanae said. “And verifying what is realistic and what is authentic is something that has to be almost like practice.”

Why Cloud Image Is Important

This observation can have serious consequences in society where images of large crowds at public events such as rock concerts, protests, political gatherings and other public events have major currency. “We want visual metrics, a way to determine if someone is successful.” GADO imagesa company that uses AI to help manage visual archives. “And crowd size is often a good indicator of that.”

a Report Global consulting firm Capgemini shows that nearly three-quarters of images shared on social media in 2023 were generated using AI. With technology becoming increasingly adept at creating compelling crowd scenes, manipulating visuals has never been easier. This brings both creative opportunities and social dangers. “AI is a great way to chew and inflate the size of a crowd,” Smith said.

He added that there is also a behind-the-scenes side to this phenomenon. “If you have a real image that surfaces something that is politically inconvenient or harmful, you're likely to say, “No, that's a fake AI.” ”

One example of this occurred in August 2024. Spreading false claims The team of its democratic rival, Kamala Harris, used AI to create images of a large number of supporters.

Chapman University lecturer Charlie Fink writes about AI and other emerging technologies. Forbessaid it's easy to trick people by believing that the fake crowd scenes are real, the real crowd scenes are fake, or the real crowd scenes are fake, due to the way the images are delivered. “The challenge is that most people are looking at content on a small screen, and most people are not so critical of what they see or hear,” Fink said. “If it looks real, it's real.”

Balancing creativity with public safety

For the technology companies behind AI image generators and social media platforms where AI-generated stills and videos have been landed, there is a delicate balance between allowing users to create more realistic and reliable content, including detailed crowd scenes, and allowing them to mitigate potential harm.

“The more you can create results, the more realistic and incredible you can give people more choices for creative expression,” says Oliver Wang, a leading scientist at Google Deepmind, co-leading the company's Image Generation efforts. “But the misinformation is something we take very seriously. So we stamp every image we generate with visible and invisible watermarks.”

However, the visible, i.e., public surfaces that are currently displayed in videos created using Google's VEO3 are hidden in the corner of the screen and are easy to miss and easy to miss. (Invisible watermarks like Google Synthidinvisible to the eyes of normal users. They help tech companies monitor behind the scenes AI content. )

Additionally, AI labeling systems are still applied unevenly across platforms. The companies that NPR spoke to for this story said they had motivation to develop them, but there is no industry-wide standard yet.

Instagram's parent company, Meta, the current The label uploaded content generated to AI when the user discloses it or when the system was detected. Google videos created using their own generation AI tools on YouTube have labels automatically in the description. that ask People who use other tools to create media and self-disclose when using AI. Tiktok need Creators who source or label content that shows scenes and people that appear realistic or have heavily edited. Unsigned content may be removed, restricted or labeled by your team in response to any harm that may cause.

Meanwhile, Will Smith has been enjoying AI more since the controversial concert video was released. He posted a playful Follow up The camera pans from footage of the singer playing energetically on stage, revealing an audience packed with fist pump cats. Smith included comments: “The crowd was Poppin' Tonite!!!”