AI Training is Copyright Infringement (under European law) | Hacker News Reader

AI Training is Copyright Infringement (under European law)

urheber.info

27 pointslayer844 comments

> Axel Voss, MEP and host of today's event in the European Parliament, expressly thanks the scientists Dornis and Stober and is pleased that

This guy is pleased, PLEASED about the state of the laws. He s not gravely concerned and devastated that all across the EU training is illegal. He 's not asking for laws to be rewritten and exceptions being introduced, but instead he s jerking off to his own image in the mirror.

I wonder how do German and Dutch people (the biggest EU contributors) feel about funding this level of idiocy. And what about the rest of us, the level of technophobia in the EU is giving us psychological trauma. Can we at least get compensation for our wounds

Let's just say that I wasn't surprised at all when I saw the name, as I remember it coming up in the context of laws that are at odds with what the typical internet freedom proponents ("netizens", pirate parties, ...) consider sensible.

“The composer and spokesperson for the Copyright Initiative, Matthias Hornschuh, comments: […]All the better that the authors of our tandem study…“

Seems quite a bit biased, no?

Europe is never going to be able to catch up in AI, much less innovate if it keeps cranking out useless and damaging regulation such as this.

As Elon pointed out, when regulators crank out an endless stream of laws, they never ever think about:

    a) unintended consequences

    b) the fact that once a law is on the books, the effort to cancel it is such that it almost never goes away, however irrelevant or damaging it becomes. The result is mathematically an inextricable operating environment that stifles innovation, and more generally speaking economic output.

It should be compulsory for every voted law and regulation to come with an automatic expiry date, only to be given a new lease on life by democratic process.

There is no regulation being cranked out, in the context of TFA. It is the existing copyright law in Europe that is lacking any exemptions for AI use cases.

Neither does the US for that matter. It’s not yet known if the Fair Use positive defense will apply.

This is not a new EU regulation. This is an assessment of model training under pre-existing regulation.

If legal risk exist, the issue can be discussed and regulation can evolve accordingly to people's decision.

If this is used in the EU successfully, it just means AI will continue to be developed and used everywhere else. I doubt the EU can afford to reject the kind of competitive advantage that could be enabled by it, particularly in the future when more advanced models are launched. We also can't afford to not be involved in developing them, but it already looks like we're hopelessly behind.

From the article:

> parts of the training data can be memorized in whole or in part by current generative models

Humans do the same thing...

>can therefore be generated again with suitable prompts by end users and thus reproduced

Personal copies (even verbatim) are not disallowed in most countries AFAIK, the concern is regarding distribution which is a separate issue.

I would have thought it would be fair use: https://www.youtube.com/watch?v=X9RYuvPCQUA&t=41m44s

Since the derived work is merely inspired by the original, just like a human being uses inspiration to create new things, whether consciously or not (hence the video "everything is a remix"). The only debate IMO is just how much is being "copied" exactly, and how obvious it is.

Fair use is a concept from US law.

The article is about German and European law, which differ significantly from US law in that regard.

Yes I think fair use should exist in the EU as well.

We don't even know in the US whether fair use applies. There are court cases out there which are trying to test those bounds.

Personally, I don't think it is. Significant portions of copyrighted works can be re-created, and these models are being leveraged to make money. That makes them unlikely to be fair use. But, I'm not a lawyer nor judge.

I don't think there can be a blanket definition that applies to everything. Just like "inspiration" humans use to create all kinds of art already (even subconsciously), it's a case-by-case analysis, and one judge's subjective opinion, over how significant is too significant, for each creation that is distributed.

Sure. But how many failed cases does it take for OpenAI or Diffusion close their doors? A failed copyright defense can result in a halt on distribution (shut it down while damages are worked out) in addition to fines and lawyer fees.

My understanding is that those companies are not distributing their own generated content, and that user-submitted ones (if such things are even posted on their own website? I haven't seen much) would fall under Section 230 protections in the US.

If the U.S. operated like the EU, we’d be far less advanced technologically. In a world run by EU-level bureaucracy, there would be no Tesla, Apple, SpaceX, OpenAI, Meta, Amazon. Europe seems to carry this inflated sense of self-importance, but what have they really contributed to the global tech landscape that justifies this ego? I live in Europe, but let’s be honest—people here rely on U.S. products daily. Aside from Spotify, how many truly global tech innovations have come from Europe? The EU is not only a stagnating economy, it’s also dependant on U.S. not only for military aid but also products and services like Google search, phones, Instagram and ChatGPT.

> how many truly global tech innovations have come from Europe?

Way more than you think. You are mostly listing businesses and not actual technologies or inventions that shaped the world.

True. WWW came out of Zurich as I recall. On the other hand, if someone doesn’t productize a technology and get it into peoples’ hands, does it matter that it was researched?

Yes it matters. Even in the US. Otherwise you wouldn't have copyright laws, royalties, patent...

You are mistaking causes of all this. EU is lagging not because of regulation. EU is lagging because it doesn't doesn't have a single language and single set of laws (I mean that that laws do vary between EU members). This is THE only reason for the lack of startups, because obviously it is easier to launch a calendar app, car or rocket startup while adhering to a single set of laws instead of 27 set of laws, and don't bothering about internationalization instead of having 24 languages to work with.

Going back to regulation - there are A LOT of shitholes with low regulation around the world, and as we can see none of them are in any technologically successful. Because regulation or lack of it is not a main reason for success.

California is the most regulated US state which is where most of the companies you mention started. A much more likely and logical reason is what the other reply to your post states, the EU is a very fragmented market. There are also cultural differences certainly at play. But this very American trope that regulations are bad for business is misguided, it is fact the strong rule of law in the US and regulations that offer a great framework for startups in addition to the largest single market in the world, and the government funding a lot of research and dev through defense spending.

And from what I have seen in several sectors, companies are leaving the EU in droves due to this bureaucracy. Just try starting a small business in Germany for example... it's exhausting. California in the US is having a similar problem. Healthcare and medical companies for example not only have to deal with the federal regulatory bodies like FDA, but California also has their own independent "FDB" that also perform their own separate audits and such... and this is the case in many other industries as well.

TSMC runs on machines from ASML, which is a Dutch company. There are many “background” industries like that where European companies are global market leaders.

Companies like Apple also maintain R&D centers in Europe.

It's not Europe, it's Brussels. We don't control it, it is not accountable to us , we re just watching from afar, the same way that you do.

> I live in Europe

> I have lived in Silicon Valley for two decades

Why are you living in Europe if you such US corporation fanboy?

I'm shocked people in this thread think technological and research advancement means breaking laws and stealing copyrighted data.

I'm hoping for this article to show up again on the frontpage on a EU timezone where I can read some civilized discussion about it.

Move fast and break things has always been a popular slogan in the startup industry, and "things" is quite often a replacement word for "laws I disagree with".

Plus, many technologists believe that copyright should be outright abolished, again because they disagree with it. No matter that around 40%† of the US' GDP is generated by industries which make use of copyright protection.

† https://www.uspto.gov/ip-policy/economic-research/intellectu...

All of them can be abused but patents are the most abusable among IP protections.

> stealing copyrighted data.

As ever, I'll be a pedant and point out that "stealing copyrighted data" is not a thing.

More substantively, we don't know whether training is copyright infringement or not. The courts have yet to weigh in in any jurisdiction I'm aware of (i.e. the EU or the US).

I didn't make any statements about rulings. I just commented about what seems to be the general sentiment this time around.

As others pointed out, if LLM startups have to go broke after not being able to steal any more data then that will just be the reality of it.

Hmm...it's not "steal any more data". Stealing is taking property belonging to another, with the intention to permanently deprive.

> As ever, I'll be a pedant and point out that "stealing copyrighted data" is not a thing.

I’m sure that if you somehow got access to ChatGPT weights and started selling them, OpenAI would be happy to call it stealing.

I'm really only concerned with what the courts call it.

It's dubious as to whether LLMs are breaking the law even if this one study says they are. But it hasn't gone to court in the EU I think.

Let me get this straight. They crawl the entire web without explicit permission, only a weird opt-out process. They save a copy of all the copyrighted data on the entire web and preprocess it into a compressed form. Then in response to user requests they serve that copyrighted data directly to users, either paraphrased, quoted in small snippets, or even whole articles verbatim. Oh yes, Google Search is most definitely totally illegal.

My point in saying this isn't just that search looks very similar to AI. A bigger point is that the same people making these arguments would have, and did, make those arguments against search. Imagine the world today if their lawsuits had been successful back in 1999 and building a search index had been deemed illegal. The world would not be a better place today.

Search has already been litigated, and is allowed because it brings benefits to the original sites through links back to the original sources. AI does no such thing, and therefore is not covered under any rulings that allow search indexes. It’s a totally different thing.

Perplexity cites its sources with hyperlinks. So it's totally legal in your opinion?

Great news, protection of creators and their work should be a priority. And if text generators will go broke without stealing other's property then good riddance.

Unless I misunderstood, that's, to put a bit bluntly, just their opinion, no court or government involved?

Opinion pieces at these levels are technically just that, but in practice, they can form the legal consensus.

I don't know how common it is, but I have the impression that in at least one jurisdiction I'm aware of, publishing a journal article that argues for an interpretation of the law favorable to your client is an actual legal strategy employed by lawyers (for areas where the law is open to a lot of interpretation/there isn't much precedent yet).

Source? I have serious doubts this actually exists in any meaningful capacity in the world.

University professors specializing in copyright law and AI are involved. It’s a study whose results are being presented to the European Parliament.

The title suggests that AI Training _is_ copyright infringement, but none of the folks talking have the authority to say that. An (increasingly common) case of folks asserting opinions as fact.

And EU's descent into technological irrelevance continues at a rapid pace.

Reply on news.ycombinator.com