Google take a step towards transparency in the Gen AI era

Google is taking steps to address one of the main concerns in the AI era. The use of manipulated images for malicious purposes has been on the radar of companies for a while now. Now, Google is taking a new step towards transparency in the use of generative AI.

There are still people who are hesitant about using artificial intelligence to generate images. For instance, some artists believe that AI-powered services steal their work from the Internet without their consent, while others worry about the potential misuse of manipulated images. Therefore, the big names in the segment have a challenging task ahead of them in convincing the public.

Google’s improved “About this image” to bring more transparency about AI-generated pictures

In line with this, Google is implementing tools for greater transparency regarding the origin of images, including those generated by AI services. Earlier this year, the company joined the Coalition for Content Provenance and Authenticity (C2PA) committee. Since then, they’ve been helping develop technologies that help identify the “trail” of images on the internet. Now, a new blog post from the company provides updates.

In the coming months, Google Search results will get an improved “about this image” feature. The company will be implementing C2PA metadata to make the feature more powerful. C2PA metadata can indicate whether an image originated from AI, underwent editing, or originated directly from a camera. The “About this image” feature was already capable of displaying AI-generated image labels. However, the integration with C2PA metadata will enable it to show more information about the image’s origin.

Improvements powered by C2PA’s standard

If you’re not aware, C2PA’s standard includes a series of technical data that creates a trail of an image’s origin. Big names like Google, Amazon, OpenAI, Microsoft, Adobe, Intel, and Arm have all endorsed the standard. However, implementation into their main tools has been a bit slow. Google’s move could inspire other companies to take action in this regard.

There are still some technical limitations to the mass implementation of C2PA metadata. On the camera hardware side, only a few Leica and Sony models currently support it. Nikon and Canon cameras are expected to be compatible in the future as well. Furthermore, there is still no news about support on Android and iOS devices. On the software side, important editing tools remain unsupported. For example, Adobe’s Photoshop and Lightroom support C2PA standards, while others such as Affinity Photo or Gimp do not yet. Plus, not all social platforms can display AI-generated labels yet.

Overall, it is only a matter of time before mass adoption of standards and legislation for AI takes place. LLMs took the world by storm, developing at a faster pace than legislators and standards developers. Indeed, early August 2024 saw the establishment of the first EU AI Act.