What the EU's New AI Labelling Law Means for Photographers

From August 2nd Article 50 of the EU AI Act will establish mandatory transparency obligations for AI images.

We’re already seeing labels on Facebook, Instagram and TikTok indicating if content has been created with AI. Through a partnership with the Coalition for Content Provenance and Authenticity (C2PA), LinkedIn is also flagging AI content (indicated by a small icon with the letters "cr" inside).

This may all sound like good news, but what does it really mean for our understanding of, and belief in, photographs?

Proof of the Real

Naturally, the AI Act and AI labels are designed to give us confidence, restoring trust in the photographic image.

But as AI expert Paul Melcher has pointed out, a worrying possibility is that once AI labels become the norm, the absence of a label will be understood as ‘proof of the real’. That’s a dangerous assumption. Melcher also points out that “more than 30 billion AI images are already in circulation”. That’s almost 4 AI pictures per person on the planet. A number that’s only increasing by the second.

One may start to wonder, what of images that are created and manipulated using AI outside the EU? Will they need to be labelled too? How is a regulation like this enforceable when vast swathes of AI-generated images are pouring onto the Web?

EU Artificial Intelligence Act

How Will the EU Track AI-Generated Content?

The new regulation targets content created outside the EU by stipulating in Article 2 (1)c that if the output of an AI system was intended for use or consumption in the EU then the new EU transparency rules apply regardless of where the content was created.

In an interconnected world, where everything online is accessible from anywhere, I’m not even sure how EU regulators would effectively establish intent to distribute in the EU. And how would they police the billions of images that pour onto the Internet each day?

Sure, hefty fines are a dissuasion, but the labyrinth of rules of the EU is unlikely to be looked at closely beyond the legal departments of major businesses and media organisations, for whom a transgression would destroy their reputation and literally cost millions.

With unregulated citizen journalism increasingly drawing audiences away from mainstream media, who is going to police the content uploaded by billions of individual ‘netizens’?

Getty’s Stance on AI-Manipulated Images

At LightRocket we’re partnered with Getty Images and know from first-hand experience that Getty already has a zero-tolerance policy for any AI-manipulated images in its editorial collection.

This doesn’t just mean images generated using AI, it encompasses the kinds of widely-available image enhancement apps that are seen by many photographers as innocuous tools for improving their pictures; rather than a path to deception and AI fabrication.

Background photo by Pixabay

Topaz and Luminar Neo are good examples of AI-powered photo apps that market themselves as tools for photographers to make their pictures look better without actually generating AI images. Even Photoshop, respected behemoth of digital photography, has introduced AI tools for enhancing pictures within its software.

None of these tools are acceptable to Getty, which uses automated forensic software to ferret out any signs of AI intervention on images uploaded to its editorial collection.

We have seen photographers face bans for using tools that probably seemed harmless to them. The position taken by Getty, and other news services like it, is that all its editorial imagery must come with a guarantee that it was created, managed and manipulated by a human (and ideally one imbued with the ethics of professional journalism). That being said, if AI software is being used to do what humans could previously do using a tool like Photoshop, I’m not sure what the difference is.

Any Use of AI is Labelled as Such. Even if it’s Only for Editing

Even outside the confines of professional news services like Getty and the wire services, the boundaries of AI manipulation remain problematic. I recently noticed a picture posted by a friend on Facebook that was marked as AI content. The photograph was taken in Afghanistan in the 80s, during the war. It showed a group of mujahadeen fighters. I know my friend took that picture. It wasn’t fake.

So why was it labelled as AI, prompting some of his friends and followers to raise questions about the veracity of the image?

In the ensuing comment thread, my friend revealed he’d copied his own picture (that he’d originally taken) from a publication and used AI to clean it up. In this instance, AI had been used as a tool to improve the picture without changing its content. But it still appeared with an AI label. The picture was real but the label suggested otherwise, causing confusion among its viewers.

When digital technology replaced film, we quickly understood that digital photographs were infinitely malleable and adjusted our trust and perception of photographic images accordingly.

Credibility in Favour of the Photograph Has Dissolved

In the pre-AI era of digital photography, the tools of manipulation were still relatively manual. The challenges associated with creating deep digital fakes were much more significant (lots of time fiddling around on Photoshop), so the trust problem felt marginal. The balance of credibility remained in favour of the photograph.

Today, however, AI technology is not just facilitating the manipulation of existing images, it allows anybody to create photorealistic pictures from simple verbal prompts. Increasingly (as AI improves), our eyes and brain will no longer be able to distinguish between reality and confection.

Our understanding of what a photograph means, and its value as a reliable vehicle of witness, is being challenged.

The new EU AI Act is clearly a step in the right direction. But regulations will not be enough.

Could File Traceability Make Photographs Reliable Once Again?

Just as technology creates the possibility to fabricate photorealistic pictures, innovations in technology can surely be used to provide us with an audit trail of a file’s history (C2PA aims to do this), showing us when a file was created, using what instrument of capture, and if and how the file has been altered since its inception, and so on.

Viewed from a more positive perspective, the enormous challenges AI is creating for photography may also move us towards a world where a file’s immutable history will become easily verifiable and this, combined with legally required labels (like those required by the EU) will give us greater confidence in digital photographs than ever before.

Paradoxically, AI and the extreme threat it poses to the credibility of photography may be a catalyst for a technological leap that will make photographs more reliable tools of witness than ever before.

---

Written by Yvan Cohen | Yvan has been a photojournalist for over 30 years. He's a co-founder of LightRocket and continues to shoot photo and video projects around Southeast Asia.

To read more helpful articles on photography, check out our blog page.

Join our growing photographer community at LightRocket and get powerful archive management and website building tools for free!