AI companies aren't seriously arguing that copyright shouldn't apply to them because "it's bad for business". The main argument is that they qualify for fair use because their work is transformative which is one of the major criteria for fair use. Fair use is the same doctrine that allows a school to play a movie for educational purposes without acquiring a license for the public performance of that movie. The original works don't have model weights and can't answer questions or interact with a user so the output is substantially different from the input.
> Fair use is the same doctrine that allows a school to play a movie for educational purposes without acquiring a license for the public performance of that movie. This is a pretty bad example, since fair use has been ruled to NOT allow this.
It is a bad example, but not for that reason. Instead, it's a bad example because Federal copyright law has a specific carve out for school educational purposes:
https://www.copyright.gov/title17/92chap1.html#110 "Notwithstanding the provisions of section 106, the following are not infringements of copyright:
(1) performance or display of a work by instructors or pupils in the course of face-to-face teaching activities of a nonprofit educational institution, in a classroom or similar place devoted to instruction, unless, in the case of a motion picture or other audiovisual work, the performance, or the display of individual images, is given by means of a copy that was not lawfully made under this title, and that the person responsible for the performance knew or had reason to believe was not lawfully made;"
That is why it is not a good comparison with the broader Fair Use Four Factors test (defined in section 107: https://www.copyright.gov/title17/92chap1.html#107) because it doesn't need to even get to that analysis, it is exempted from copyright.
If AI companies don’t want the court headaches they should instead preemptively negotiate with rights holders and get agreements in place for the sharing of data.
Feels like bad faith to say that knowing full well that
1. This would also be a massive legal headache,
2. It would become impossibly expensive
3. We obviously wouldn't have the AI we have today, which is an incredible technology (if immature) if this happened. Instead the growth of AI would have been strangled by rights holders wanting infinity money because they know once their data is in that model, they aren't getting it back, ever—it's a one-time sale.
I'm of the opinion that AI is and will continue to be a net positive for society. So I see this as essentially saying "let's go an remove this and delay the development of it by 10-20 years and ensure people can't train and run their own models feasibly for a lot longer because only big companies can afford real training datasets."
Why not simply make your counterargument rather than accusing GP of being in bad faith? Your argument seems to be that it's fine to break the law if the net outcome for society is positive. It's not "bad faith" to disagree with that.
But they didn't break the law. The NYT articles were not algorithms/AI.
It's bad faith because they are saying "well, they should have done [unreasonable thing]". I explored their version of things from my perspective (it's not possible) and from a conciliatory perspective (okay, let's say they somehow try to navigate that hurdle anyways, is society better off? Why do I think it's infeasible?)
If they didn't break the law, your pragmatic point about outcomes is irrelevant. Open AI is in the clear regardless of whether they are doing something great or something useless. So I don't honestly know what you're trying to say. I'm not sure why getting licenses to IP you want to use is unreasonable, it happens all the time.
Edit: Authors Guild, Inc. v. Google, Inc. is a great example of a case where a tech giant tried to legally get the rights to use a whole bunch of copyrighted content (~all books ever published), but failed. The net result was they had to completely shut off access to most of the Google Books corpus, even though it would have been (IMO) a net benefit to society if they had been able to do what they wanted.
> Your argument seems to be that it's fine to break the law if the net outcome for society is positive.
In any other context, this would be known as "civil disobediance". It's generally considered something to applaud.
For what it's worth, I haven't made up my mind about the current state of AI. I haven't yet seen an ability for the systems to perform abstract reasoning, to _actually_ learn. (Show me an AI that has been fed with nothing but examples in languages A and B. Then demonstrate, conclusively, that it can apply the lessons it has learned in language M, which happens to be nothing like the first two.)
> In any other context, this would be known as "civil disobediance". It's generally considered something to applaud.
No, civil disobedience is when you break the law expecting to be punished, to force society to confront the evil of the law. The point is that you get publicly arrested, possibly get beaten, get thrown in jail. This is not at all like what Open AI is doing.
>I'm of the opinion that AI is and will continue to be a net positive for society. So I see this as essentially saying "let's go an remove this and delay the development of it by 10-20 years and ensure people can't train and run their own models feasibly for a lot longer because only big companies can afford real training datasets."
Absolutely. Which, presumably, means that you're fine with the argument that your DNA (and that of each member of your family) could provide huge benefits to medicine and potentially save millions of lives.
But significant research will be required to make that happen. As such, we will be requiring (with no opt outs allowed) you and your whole family to provide blood, sperm and ova samples weekly until that research pays off. You will receive no compensation or other considerations other than the knowledge that you're moving the technology forward.
May we assume you're fine with that?
Yeah, and the online radio providers argued that they don’t do anything shady, their service was basically just a very long antenna.
Anyway, the laws were not written with this type of processing in mind. In fact the whole idea of intellectual property breaks down now. Just like the early days of the internet.
https://www.copyright.gov/title17/92chap1.html#110 seems to this non-lawyer to be a specific carve out allowing movies to be shown, face-to-face, in non-profit educational contexts without any sort of license. The Fair Use Four Factors test (https://www.copyright.gov/title17/92chap1.html#107) isn't even necessary in this example.
Absent a special legal carve-out, you need to get judges to do the Fair Use Four Factors test, and decide on how AI should be treated. To my very much engineer and not legal eye, AI does great on point 3, but loses on points 1, 2, and 4, so it is something that will need to be decided by the judges, how to balance those four factors defined in the law.
> AI companies aren't seriously arguing that copyright shouldn't apply to them because "it's bad for business".
AI companies have, in fact, said that the law shouldn't apply to them or they won't make money. That is literally the argument Nick Clegg is using to ague that copyright protection should be removed from authors and musicians in the UK.
That's not entirely true. A lot of their briefing refers to how impractical and expensive it would be to license all the content they need for the models.
> allows a school to play a movie
No, it doesn’t. Play 10% of a movie for the purpose of critiquing it, perhaps.
https://fairuse.stanford.edu/overview/fair-use/four-factors/
Fair Use is not an a priori exemption or exception; Fair Use is an “affirmative defense” so once you have your day in court and the judge asks your attorney why you needed to play 10% of Priscilla, Queen of the Desert for your Gender Studies class, then you can run down those Four Factors enumerated by the Stanford article.
Particularly “amount and substantiality”.
Teachers and churches get tripped up by this all the time. But I’ve also been blessed with teachers who were very careful academically and sought to impart the same caution on all students about using copyrighted materials. It is not easy when fonts have entered the chat!
The same reason you or your professor cannot show/perform 100% of an unlicensed film under any circumstance, is the same basis that creators are telling the scrapers that they cannot consume 100% of copyrighted works on that end. And if the risks may involve reproducing 87% of the same work in their outputs, that’s beyond the standard thresholds.