Copyright is one of the biggest issues in AI. Eric Wengrowski, the CEO of Steg.AI explains how digital watermarking can help.

It's fair to say the subject of copyright comes up a lot when you're talking about AI. Whether you're talking about a large language model (LLM) like the ones that power ChatGPT, or diffusion models that serve text-to-image creators like Midjourney, these generative systems suck up massive amounts of training data from the open web.

This has concerned many content creators and publishers, including The New York Times, which brought its concerns to the courts in late December. While the world waits for the law to catch up to the AI industry, the question remains: can authors, photographers, videographers and anyone else in the business of creating content do anything to ensure they stay connected and in control of the things they create?

There might be. What all these issues are circling is the concept of content provenance: ensuring the copyright holder of any piece of content is embedded within the content itself. One way to do that through digital watermarking — essentially creating an "Invisible QR code" that travels with the document, image, or video, even if it's copied and stripped of metadata.

Steg.AI is a company that specializes in digital watermarking, and The Media Copilot spoke with its CEO, Eric Wengrowski in our latest podcast. We fully explored the role of watermarking in a world where all kinds of web crawlers are constantly hoovering up data, why it's important to label synthetic content, and the incredibly important question of: can you still detect the watermark of a piece of training data in model output?

The Media Copilot is a podcast and newsletter that explores how generative AI is changing media, journalism, and the news.

⁠⁠⁠⁠Subscribe to the newsletter.⁠⁠⁠⁠

⁠⁠⁠⁠Follow on X.⁠⁠⁠⁠

Subscribe to the podcast on:

Music: ⁠⁠⁠⁠Favorite⁠⁠⁠⁠ by Alexander Nakarada, licensed under Creative Commons BY Attribution 4.0 License

© AnyWho Media 2024