Adobe releases open source toolkit to counter visual misinformation – TechCrunch
Adobe envisions an Internet dotted with photos and videos that contain additional data about their origin. The company’s central goal is to mitigate the spread of visual misinformation, but the system could also be a boon for content creators keen to keep their name attached to their work.
First announced in 2019, Adobe’s Content Authenticity Initiative (CAI) project has since released a white paper on a technology to do just that, introduced the system in its own software, and partnered with newsrooms and hardware manufacturers that could help universalize its vision.
Picture credits: Adobe
As EXIF data stores aperture and shutter speed details, the new standard also records information about a file’s creation, including how it was created and edited. And if the company’s shared vision comes to fruition, that metadata, which Adobe calls “content credentials,” would then be widely visible on social media platforms, image editors, and content sites. news.
“Detection [for misinformation] It’s going to be an arms race, and, you know, frankly, the good guys are going to lose,” Andy Parsons, senior director of Adobe’s Content Authenticity Initiative (CAI), told TechCrunch. “We’ve been more focused on doubling down on the authenticity of the content, which is this idea of proving what’s real, how something was done, where it makes sense, who did it.”
The underlying standard, C2PA, is the result of a collaboration between Adobe’s CAI and partners including Microsoft, Sony, Intel, Twitter and the BBC. The Wall Street Journal, Nikon, and The Associated Press also recently joined Adobe’s commitment to mainstream content authentication.
CAI’s main goal is to counter visual misinformation online – think old recirculated images misrepresenting the war in Ukraine or Nancy Pelosi’s infamous “cheapfake”. But a digital chain of custody could also benefit content creators who have their work stolen or offered for sale, a phenomenon that has plagued visual artists for years and is now causing NFT market headaches.
Parsons noted that CAI is also seeing surprising interest from companies producing synthetic images and videos. By embedding origin metadata into the kind of AI creations we see on models like DALL-E, companies can be sure that these generative images aren’t easily mistaken for reality.
While the C2PA standard can be compared to something like EXIF, Adobe says the new content attribution standard is much less “fragile” in terms of tampering with or breaking attached information. Through a verification website Adobe launched last year, anyone can drag and drop an image with content credentials to verify that they are legitimate. And even if someone breaks the embedded data, it can be reconnected through image fingerprinting techniques.
Adobe’s vision for authenticating online content is broad, but it remains realistic about the limits of the project. Bad people will always find a way to mislead others, but the company hopes that a good portion of average Internet users will be willing to receive more information about what content they can trust online.
The company also hopes that the more the standard is adopted, the more high-profile social media platforms will be stimulated to implement it, even if they are reluctant at first. Flickr, for example, has long displayed EXIF data next to each image, but Instagram and most other modern image-based social networks remove metadata, even though some of them invite users to add location tags.
Social platform integration starts with convincing these companies to keep content credentials intact. “Once apps and websites and platforms start preserving them, you know, downstream consumers who might download something from the Instagram website or might share something on WhatsApp will know that the data exists,” Parsons said. “And then we think it’s a matter of time, like a matter of the snowball effect, for people to show it. There’s really no downside.
The company is investing in education around content authenticity and visual misinformation, but it’s moving quickly to lay the groundwork for widespread adoption — and that’s something Adobe has been doing for a long time with formats. of ubiquitous files like XMP and PDF and its industry standard image editing software. .
“It’s one thing to create a specification, have it ratified by industry leaders like our partners, Microsoft, Intel, ARM, Nikon and others. It’s another thing to see it actually adopted,” Parsons said. “…The catalyst for adoption has been [always] are going to be tools that help people implement things and make sure they can do it for free. So, you know, no intellectual property, no license fees, no royalties, just real free open source code.