Meta Thinks Everything Is AI Now

Meta Thinks Everything Is AI Now

Did you use AI to make that product image? No? Then why is Meta telling everyone you did?

Meta Applying “AI” Label to More Than You Think

Well, I didn’t see this on my “AI takes over everything” Bingo card.

Meta is labelling images as having been “Made with AI” — images which were not, in fact, made with AI. And Techcrunch reports today this has been going on since last month.

One example they showed was of a team winning a cricket league tournament. It’s a real photo, shot with a real camera, held by a real human — but the “Made with AI” label shows up anyway. Interestingly, it’s only visible on mobile apps and not the web.

It’s not a bug

And lest you think this is a bug, no — because Meta lives in a parallel universe, this is by design.

Meta’s system seems to be applying the label to any image where AI has even just touched the image — to add a small element with generative fill, for instance.

Pete Souza, the former White House photographer, said all he did was flatten an image into a JPG and that’s all it took for Meta to apply the label.

Industry standards at fault?

Also, Souza did this photographic work not on Meta’s own platform, but using an Adobe tool. Still, Meta applied the tag.

That’s because Meta is honouring some new technical standards — C2PA and some from the IPTC. This lets software add metadata identifying that some AI took place. This means that Meta can identify when images touch AI from all the major sites: Google, OpenAI, Microsoft, Adobe, Midjourney, and Shutterstock.

This means even if all you do is remove a tree branch from the background using Adobe’s Generative Fill, Adobe will dutifully slip some metadata in, Meta will pick that up, and report that (well, imply that) the whole image was conjured out of thin air.

New language needed?

Obviously this is a language issue, and the industry still hasn’t figured out how to handle granular use of image AI. There’s a big difference between “Made with AI” and “Touched up by AI.”

But do we really even need that?

Removing items from backgrounds has been something photographers have done since the invention of the medium. And even recent technology, like Pixelmator’s Repair tool, can do this and doesn’t use any of the new generation AI.

To be fair, once you tap Meta’s label, it does explain things a bit better. It reads “Generative AI may have been used to create or edit content in this post” — but nobody taps that.

The marketing implications

As for marketers, this might actually become a bigger issue as the months go on. There are national elections happening this year all over the world: France, Ghana, Mozambique, the U.K., Austria… and I feel like I’m missing another one. It’ll come to me.

Elections, of course, are a breeding ground for misinformation, and we may very well see consumers form worsening opinions about generative AI.

Marketers could very well end up having the same labels applied to their images as some big scandals — and that’s no good no matter how hard you or Meta spin it.

Oh…. Venezuela. That’s the one I forgot. July 28th.