Claude's watermarks and buttons to flag AI content

I'm clearly in favour of identifying AI content. But there's a bigger reason.

modelos
ética-e-limites
dados

Ilustração de uma sirene de alarme vermelha com as letras IA

I'm clearly in favour of identifying AI content. And in the past few days there's been news of big tech decisions in that direction:

Anthropic (Claude) is going to add "watermarks" to the text it creates.
LinkedIn already lets people report content that "looks like AI slop".
Meta labels AI content on Facebook, Instagram and Threads.
TikTok uses labels, metadata and watermarks.
YouTube started applying labels automatically to some videos.
Pinterest flags images as "AI modified".
Reddit is more focused on detecting bots, automation and AI content used to deceive.

At first glance, this looks like a response to users who don't want their feeds full of generic, artificial "fast food" content.

But there's a bigger reason.

Models have already been trained on a huge share of the public text available on the internet.

Now, if the internet starts filling up with AI-made content, a loop can form:
- A model produces content.
- That content gets published.
- It gets collected to train the next models.
- Models start learning, in part, from other models' output.

Researchers call this "model collapse".

When generative models are trained indiscriminately on synthetic content, they can start losing diversity, nuance and rare cases. They become more uniform. And errors can repeat and amplify.

And if we already think AI content all looks the same, imagine if nothing is done about it. And who's better placed than human users to judge whether content looks like AI to them or not?

Right now, human content — original, contextualised and verifiable — can become an increasingly valuable resource.

That's why labels, watermarks and "slop" reports don't only exist to protect readers. In fact, I'm always suspicious when one of these big tech companies suddenly discovers a "human" concern like this. And on that front, the European Union has been very prudent, creating laws that, to me, make complete sense.

The image for this post was made by me with AI.

Sources