Building a validity scorecard in the age of AI

A German artist made waves last week after he revealed that his winning submission in this year’s Sony World Photography Awards was actually generated by AI.

The stunt, the artist said, was intended to start a conversation about realism in the age of AI, and was the latest example of just how easy it has become for media spawned by AI to deceive even the most discerning viewers.

Andy Parsons, senior director at the Adobe-led Content Authenticity Initiative (CAI), sees incidents like these as a “wake-up call.” The CAI is a consortium of around 1,000 members, including companies ranging from big-name media outlets and camera makers to tech startups, collectively working to craft a set of standards that they claim could guarantee the authenticity of digital content.

Adobe first formed the group in 2019, but its mandate has become more urgent as new AI tools have made the technology more sophisticated and accessible in the past year or so. That has resulted in viral fake news videos and photos, like one of Donald Trump being arrested in Manhattan, and led Getty Images (which is a CAI member) to ban AI-generated content from being uploaded to its platform.

Detecting deepfakes is a "constant cat-and-mouse game," Parsons said. Only a few months ago, experts would advise savvy media consumers to look at the ears or hands of a photo subject as a way to potentially detect if they were AI creations—those body parts tended to be trouble spots for the technology—but techniques like that became outdated quickly, often in a matter of weeks, he said.

“We’re about to cross the chasm in terms of photorealism and audio realism where [detection] will become effectively impossible, certainly in terms of social media scale,” Parsons said.

That’s why the CAI is instead focused on a standard that it claims would certify the authenticity of a photo or video from the source, using “cryptographic asset hashing to provide verifiable, tamper-evident signatures,” which would then be amended to reflect any alterations. The group compares the standard to a “nutritional label” for digital content.

“You can buy the sugary food if you want it and use it in moderation,” Parsons said.

Of course, the standard is only as effective as the number of companies that agree to take part in it, and that’s where Adobe’s position as one of the leading creative software suites is important. Camera companies like Nikon and Leica have already agreed to build the tech into some of their cameras, and the CAI hopes that member news businesses like the Associated Press, the Washington Post, and Gannett will eventually implement the standard in their content as well.

“At the end of the day, we’re most interested in making sure that there’s one standard way to do this provenance thing, not 15 different competing standards,” Parsons said.

Projects like the CAI have arisen as AI’s development has so far outpaced the ability of regulators to keep up with the fast-changing space.

While politicians in the US have expressed concerns about the pace and direction of AI technology in the past several weeks, there has yet to be much concrete legislation comprehensively governing AI content. One exception is China, which earlier this year enacted a far-reaching deepfake law that requires disclosure of tech alteration, mandates consent of deepfake subjects, and bans deepfake use in news, among other provisions.

“If there is legislation that eventually enforces things like nutrition labels like the FDA does, that’s great. We’ll be ready for that,” Parsons said.

The CAI views the looming 2024 presidential election cycle as a deadline of sorts for rolling out its technology at scale. By then, the group aims to give its member organizations the means to implement the standard, as well as educate consumers on what the icon means when they encounter it online.

“We think about the US election in 2024 as kind of a milestone moment where it’s very critically important that provenance technology and CAI be relevant,” Parsons said.