EU enforces labeling AI generated content

Posted by nucatus 3 hours ago

Comments

Comment by nucatus 17 minutes ago

The actual EU regulation on labeling AI content:

https://digital-strategy.ec.europa.eu/en/policies/code-pract...

Comment by account42 1 hour ago

I wonder if this will just end with "may contain AI content" labels on everything instead of achieving something useful.

Comment by boobsbr 15 minutes ago

Like everything in California may cause cancer.

Comment by hananova 15 minutes ago

"May contain X" labels are usually not allowed here with some exceptions. This has some downsides (like deliberately adding allergens to food because it's impossible to make sure there are absolutely none) but in general you can't just hedge your way past regulations.

Comment by Xelbair 2 hours ago

Isn't the whole problem that it is impossible to determine if content is AI or not? Doesn't this make it impossible to prove under 'innocent untill proven guilty' unless you're really blatant about it?

And isn't this easy to sidestep via plausible deniability and using middleman outside of EU borders?

So bad actors, especially ones outside of EU, can just continue to operate as is and ones within can just outsource the 'bad' work outside.

AI disclosure makes sense, but it should be a short list in standardized label format.ie AI used for code, assets, text, etc.

The problem is what we(as society )try to sidestep: is to have content creators(including major traditional media) be held responsible for content they provide.

Comment by broken-kebab 1 hour ago

I'd say that for now LLM-generated content is detectable, still at least. But anyway, "proven guilty" doesn't work as many people seem to believe. Often, if not most of the times, courts deal with cases with no full direct information which 100% prove guilt or its absence. If enough pieces of general picture suggest something, it usually considered proven.

In the same way a suspect may be declared guilty when e.g. there were reasonable expectation to double check a published image, a possibility to do it, obvious benefits from spreading fakes. Even if that image is genuinely looking convincing.

Probably, eventually somebody will try to use it to silence real whistleblowing. And it's unclear how this will work at all with expected advancements of generation capabilities. But I believe there's a public demand for such law.

I expect pretense of uncertainty to be used to get around: "we got this video from unknown source and we are not sure if it's AI-generated, but look like how this politician f*cks a pig"

Comment by inigyou 2 hours ago

It's impossible to determine whether a killing is a murder or not. We still do it. And it's easy to sidestep by doing the murder outside the country.

Comment by NorthSouthNorth 29 minutes ago

Honestly don't even know how to feel about this regulation. I think in principle I like it. If one uses AI to generate content on a topic and is not actually checking its accuracy, why not force them to label it.

What practical benefits this might have in the long term I do not know. Perhaps an implicit trust in orgs/individuals who actively tag AI-generated content.

If one is known to tag AI-generated things as AI-generated, the un-tagged things have more weight as being original. To be honest I don't think this is a likely outcome.

Comment by dzhiurgis 1 hour ago

What for?

Comment by brettermeier 13 minutes ago

What not for? That people can't post AI fake stuff and pretend it's true.

Comment by perarneng 3 minutes ago

What if you find ex images that is not labeled should you assume its not AI? This will make everything not labeled as AI get more legitimate even though it's still at a similar level of trustworthiness. this a big problem.