But it doesn’t answer any tough questions either, like: if an AI model outputs something very close to a training input, does the result infringe on the copyrights of the input work?
But it doesn’t answer any tough questions either, like: if an AI model outputs something very close to a training input, does the result infringe on the copyrights of the input work?
>The Office recognizes that AI-generated works implicate other copyright issues not addressed in this statement. It has launched an agency-wide initiative to delve into a wide range of these issues. Among other things, the Office intends to publish a notice of inquiry later this year seeking public input on additional legal and policy topics, including how the law should apply to the use of copyrighted works in AI training and the resulting treatment of outputs.
There are also laws to consider. I know this is US law and the US doesn't have those laws, but copyright harmonisation is a thing and the EU, UK, Japan etc all have laws very explicitly stating that TDM is an exemption to copyright - some bodies would presumably consider that.
And finally, there's national security. The US's current view is that it's in an existential great power competition with Chinese society, and needs to compete on every level to succeed. One of those levels is very explicitly AI, it's considered a main "battlefront" in the new Cold War. If the US significantly restricts TDM, that not only makes it less competitive compared to jurisdictions that don't like the UK, EU, Japan, and China, but it specifically improves the effectiveness of the Thousand Talents program which the US IC and analyst community view as a very serious threat. Academics want to be able to do their research, AI is critically important to US national security, and currently the US is statistically the best place to be an AI researcher. The government won't want to jeopardize that when China is already advancing in AI very rapidly and actively trying to tempt scientists to come over with better pay and conditions.
A Deeper Look into the EU Text and Data Mining Exceptions: Harmonisation, Data Ownership, and the Future of Technology, Journal of European and International IP Law, 2022
https://academic.oup.com/grurint/article/71/8/685/6650009
"This paper focuses on the two exceptions for text and data mining (TDM) introduced in the Directive on Copyright in the Digital Single Market (CDSM).
While both are mandatory for Member States, Art. 3 is also imperative and finds application in cases of text and data mining for the purpose of scientific research by research and cultural institutions; Art. 4, on the other hand, permits text and data mining by anyone but with rightholders able to ‘contract-out’ (Art. 4).
We trace the context of using the lever of copyright law to enable emerging technologies such as AI and the support innovation. Within the EU copyright intervention, elements that may underpin a transparent legal framework for AI are identified, such as the possibility of retention of permanent copies for further verification.
On the other hand, we identify several pitfalls, including an excessively broad definition of TDM which makes the entire field of data-driven AI development dependent on an exception. We analyse the implications of limiting the scope of the exceptions to the right of reproduction; we argue that the limitation of Art. 3 to certain beneficiaries remains problematic; and that the requirement of lawful access is difficult to operationalize.
In conclusion, we argue that there should be no need for a TDM exception for the act of extracting informational value from protected works. The EU’s CDSM provisions paradoxically may favour the development of biased AI systems due to price and accessibility conditions for training data that offer the wrong incentives. To avoid licensing, it may be economically attractive for EU-based developers to train their algorithms on older, less accurate, biased data, or import AI models already trained abroad on unverifiable data." [reformatted]
Fair use is entirely dependent on the specific use. While those decisions may protect (for instance) OpenAI, they’re not helpful in determining if anyone can use the outputs from such a model.
Yes Google can index an image and serve it in search results. But I cannot copy a copyrighted image from Google and publish it in my book.
Questions like “is the output that ChatGPT just gave me encumbered by copyright?” are still extremely difficult to answer.
I agree that we’re losing our technological edge. This is why I wish we had better answers for these questions. If we want to lead with the rule of law, then the rule of law needs to be able to lead.
eg: “only human works are subject to copyright”, “the work must be an original work of authorship”, etc.
It’s just disappointing that they haven’t produced anything helpful yet. I suspect that the result will end up being permissive because there will be a big industry built up around people just kicking the can down the road by the time anyone authority issues a more specific opinion.
In that case, the consequence would be: If you don’t validate that your AI model is producing something sufficiently different than all of the inputs, you’ve created a copyright infringement Russian roulette.
You can also copy and paste from an existing copyrighted work.
In either case, it is you, the human - with agency to do so - who should be liable for the content that you publish.
I intend to make an e-ink panel that displays woodcut style images generated by stable diffusion (get a list of a 30 or so different descriptions of scenes, and have them slowly get generated over the day and display a different one each hour).
So, here's the question - are any of those woodcut images derivative of some other work? Don't know - and it likely doesn't matter as it would be something hanging on a wall in a room.
On the other hand, if I was to collect them and publish them as a book, then I, as the human who is publishing them, would need to do sufficient diligence to see if any of them are derivative works. They might be - but I am the one publishing it then - not Stable Diffusion.
Likewise, if I was creating a collection of epic rap battles between historical figures as generated by GPT... then I should search to make sure that none of the phrases that are used in there are lifted directly from some other source. Again, it is me as the human with agency that is publishing it, not GPT.
If you were to go to Fiverr and request 100 pictures of woodcuttings or people to write epic rap battles between Julius Caesar and Ghengis Khan - would it matter if it was a human lifted the text or GPT? It is still upon you, the person publishing it to do the check.
Is that sarcasm?