[...]
Still, the Google researchers note the many ethical challenges posed by a system like MusicLM, including a tendency to incorporate copyrighted material from training data into the generated songs.
[...]
Test cases (in particular the Holy Blood Holy Grail case - link) show that copyright doesn't hold if the new artwork is similar to, but not exactly the same as, the old artwork (I personally thought it was a scandal that Dan Brown's "Da Vinci Code" took so much from the original book - but I accept that this is how the law works).
Therefore, generative AI needs two steps:
Step 1: generate the new art
Step 2: check that the new art isn't too similar to any old art
Here's an alternative plan: in a world where the generation of art no longer has any value, scrap copyright laws altogether!
Here's a question: assuming that it doesn't infringe on existing copyrights, who owns the copyright for all the new text that Chat GPT is producing at a phenomenal rate?
The simple reveals itself after the complex has been exhausted.
towforce wrote: ↑Sat Jan 28, 2023 6:32 pm
Here's an alternative plan: in a world where the generation of art no longer has any value, scrap copyright laws altogether!
Times are a changing
towforce wrote: ↑Sat Jan 28, 2023 6:32 pm
Here's a question: assuming that it doesn't infringe on existing copyrights, who owns the copyright for all the new text that Chat GPT is producing at a phenomenal rate?
AFAIK, up to now all patents, copyright (and also scientific paper authorship) was dismissed for AI generated stuff.
...I know, this was Sci-Fi topic in the past, but if the AI gains meanwhile some kind of sentience, it might be considered as a juristic person some day in future, then it can claim the same rights as author of...
towforce wrote: ↑Sat Jan 28, 2023 6:32 pm
Here's a question: assuming that it doesn't infringe on existing copyrights, who owns the copyright for all the new text that Chat GPT is producing at a phenomenal rate?
AFAIK, up to now all patents, copyright (and also scientific paper authorship) was dismissed for AI generated stuff.
...I know, this was Sci-Fi topic in the past, but if the AI gains meanwhile some kind of sentience, it might be considered as a juristic person some day in future, then it can claim the same rights as author of...
--
Srdja
Early in this thread (seems like a long time ago now!), I linked an article by a legal expert who said that if somebody commissioned a work of art, the copyright belongs to the person who commissioned it, not the artist who created it. The expert said that he thought that this would apply to AI generated art: the copyright would belong to the person who commissioned it - not the company that produced the AI software. IMO asking Chat GPT to generate text is directly equivalent to commissioning a piece of written art.
Put it another way: a newspaper owner asks a reporter to write an article, which is then published. The copyright for that article belongs to the newspaper, not the reporter.
The simple reveals itself after the complex has been exhausted.
towforce wrote: ↑Sat Jan 28, 2023 7:08 pm
{...]
Put it another way: a newspaper owner asks a reporter to write an article, which is then published. The copyright for that article belongs to the newspaper, not the reporter.
Yea, cos the reporter has a copyright in the first place.
[...]
Still, the Google researchers note the many ethical challenges posed by a system like MusicLM, including a tendency to incorporate copyrighted material from training data into the generated songs.
[...]
Maybe the datasets for NN training are copyright protected? Also in chess?
a) Human hand-picked* collection of games?
b) Labeled game positions with an evulation score from an copyright protected chess engine?
* if you select by certain criteria, not only functional, by playing style for example.
Indeed a)* could mean there is a copyright on the collection of games used as training material (so not on the games themselves). But this will not seep through into the NN, let alone into the moves played.
It is a bit disappointing if an AI music generator copies material from training data into the generated songs. But this may be the same problem faced by human composers: hard to be sure for a composer that a fraction of a tune that he thinks is his original work is not "too similar" to something he heard 20 years earlier.
towforce wrote: ↑Sat Jan 28, 2023 7:08 pm
{...]
Put it another way: a newspaper owner asks a reporter to write an article, which is then published. The copyright for that article belongs to the newspaper, not the reporter.
Yea, cos the reporter has a copyright in the first place.
Exactly. There has to be an expression of some human's creativity, and it will be the reporter's even if the copyright belongs to the employer.
The UK copyright act has a provision assigning the copyright on a work that has no human author to "the person by whom the
arrangements necessary for the creation of the work are undertaken". But I believe it has not yet been decided whether this means that there can be a copyright without human creativity. (Before Brexit almost certainly not in view of EU copyright law.)
The reporter writes the report on the word processor which does some of the work:
* correcting grammar, spelling and other mistakes
* sometimes prompting the reporter what to write
There is still some human artistic input though, so it is copyrightable.
IMO the same applies to Chat GPT: in most cases, the user has an idea what's going to come back from the prompt, if not the exact words. There is human artistic input in forming the prompt (or prompts). So text generated by Chat GPT ought to be copyrightable by the user who prompted the creative writing.
Of course, under the economic laws of supply and demand, when "good" art can be generated at almost no cost, copyright loses its value anyway.
Today, you can ask for a text story in the style of your favourite film star. It seems likely to me that tomorrow you'll be able to ask for a film in the style of your favourite film star.
The simple reveals itself after the complex has been exhausted.
towforce wrote: ↑Sat Jan 28, 2023 8:46 pm
[...]
Today, you can ask for a text story in the style of your favourite film star. It seems likely to me that tomorrow you'll be able to ask for a film in the style of your favourite film star.
Yea, it is really spooky, generative AIs, text, source code, images, audio, video, 3D models...everyone wonders what GPT-4 or 5 will produce.
syzygy wrote: ↑Sat Jan 28, 2023 8:20 pm
Exactly. There has to be an expression of some human's creativity, and it will be the reporter's even if the copyright belongs to the employer.
I do not know why I want to equate this with the unhealthy obsession we see on here (IMHO...) where people 'play with their engine' others and their engines and then have to balls to say "look at this game I won/played"...only there is not much 'human creativity' involved.
