The Supreme Court will again tackle Fair Use thanks to Andy Warhol

[TwinIon]

 

The Atlantic has a great write up on the case before the court and why it's important.

 

The short version is that the photographer that took the photo that Warhol based his Prince Series on is suing the Andy Warhol Foundation (AWF) for copyright infringement. The case only started in 2017 when Prince died and the photographer became aware that her photo had been used. Originally the case was decided for the AWF, but last year the 2nd Circuit reversed that ruling. Importantly, they didn't think Warhol's work was sufficiently transformative, and was instead derivative (legally speaking) because the photograph "remains the recognizable foundation" of the image.

 

While there is a lot more in that ruling, and the 2nd Circuit affirms that Fair Use is not something that adheres well to bright line rules, it is the kind of thing SCOTUS could decide to make into a broader rule, which would dramatically change how we understand Fair Use today. The case will be argued on Oct 12.

 

 

[marioandsonic]

Oh, great.  With this plus the Amendment 230 ruling, this is going to kill the Internet.

[unogueen]

I LOVE when freedom of expression defeats freedom of expression.

[SuperSpreader]

What's going to be interesting is how this ends up playing with ai generated art/games. Ai needs examples to generate something, but it doesn't own the examples and the people generating Ai art don't own the Ai software (maybe they can license the software). It's like that random Ai death metal algorithm on YouTube wasn't so special if you know the band Car Bomb. 

[unogueen]

You bitches can't delete us. You can't even take the brush, warhol himself was a challenge to corporate fit, the iconography of money. AI art has a fundamental flaw that cannot reconcile the velocity of history. Try pulling 'minimalism' off any generator.

[SuperSpreader]

1 hour ago, unogueen said:

You bitches can't delete us. You can't even take the brush, warhol himself was a challenge to corporate fit, the iconography of money. AI art has a fundamental flaw that cannot reconcile the velocity of history. Try pulling 'minimalism' off any generator.

 

Oh I agree. I think it's going to end up a brainstorming tool and nothing more. How can it possibly think outside the box? It can't. 

[legend]

On 10/3/2022 at 8:21 PM, SuperSpreader said:

What's going to be interesting is how this ends up playing with ai generated art/games. Ai needs examples to generate something, but it doesn't own the examples and the people generating Ai art don't own the Ai software (maybe they can license the software). It's like that random Ai death metal algorithm on YouTube wasn't so special if you know the band Car Bomb. 

I think AI art built from unlicensed datasets is a problem that needs addressing. But I also think it’s relatively easy to address. Make some legal regulation about requiring license for content to be used in a dataset. Then build websites where artists can submit their art with some standard licenses for use in datasets and pay people according that that license for the data. This is beneficial to everyone because it actually gives people an incentive to help build good datasets and distributes the wealth. 

[b_m_b_m_b_m]

12 minutes ago, legend said:

I think AI art built from unlicensed datasets is a problem that needs addressing. But I also think it’s relatively easy to address. Make some legal regulation about requiring license for content to be used in a dataset. Then build websites where artists can submit their art with some standard licenses for use in datasets and pay people according that that license for the data. This is beneficial to everyone because it actually gives people an incentive to help build good datasets and distributes the wealth. 

This is the opening pitch for nfts

[legend]

31 minutes ago, b_m_b_m_b_m said:

This is the opening pitch for nfts

 

No, it's really not. NFTs offer this approach absolutely nothing that regular databases don't also offer. In fact, the lack of traceability of NFTs is an inherent weakness, unless you setup a central system to validate identities at which point why are you using NFTs? (You can also ask "why are you using NFTs" for basically anything  )

[b_m_b_m_b_m]

1 minute ago, legend said:

 

No, it's really not. NFTs offer this approach absolutely nothing that regular databases don't also offer. In fact, the lack of traceability of NFTs is an inherent weakness, unless you setup a central system to validate identities at which point why are you using NFTs. (You can also ask "why are you using NFTs" for basically anything  )

It was the opening pitch not the whole deck

 

 AI gotta update it’s sarcasm sensors

[TwinIon]

1 hour ago, legend said:

I think AI art built from unlicensed datasets is a problem that needs addressing. But I also think it’s relatively easy to address. Make some legal regulation about requiring license for content to be used in a dataset. Then build websites where artists can submit their art with some standard licenses for use in datasets and pay people according that that license for the data. This is beneficial to everyone because it actually gives people an incentive to help build good datasets and distributes the wealth. 

It's an interesting problem, but my initial reaction is very much that use in such a dataset to create AI art easily falls under the idea of a "transformative work."

 

Looking through the Dall-E paper, they mention on page 23 that they used datasets of 650M images. I don't know what kind of licensing fee would seem reasonable, but it's pretty obvious that if copyright holders made even a single dollar that such a huge dataset would be fiscally impossible. If you gave them the Spotify rate, of $0.0005, the whole data set would only cost a bit over $3M, but I don't know who that's supposed to make whole.

 

Once the algorithm is trained, you'd know better than I if there would be any way to specify what works in the dataset were really used in the creation of a specific image, but my guess is that even if you could narrow it down (say, if a user asks for a Warhol like image), it's hard to imagine finding a model that generates much revenue for the creator.

[legend]

46 minutes ago, b_m_b_m_b_m said:

It was the opening pitch not the whole deck

 

 AI gotta update it’s sarcasm sensors

 

If you also agree it's a bad pitch, then no problem here  

 

15 minutes ago, TwinIon said:

It's an interesting problem, but my initial reaction is very much that use in such a dataset to create AI art easily falls under the idea of a "transformative work."

 

Looking through the Dall-E paper, they mention on page 23 that they used datasets of 650M images. I don't know what kind of licensing fee would seem reasonable, but it's pretty obvious that if copyright holders made even a single dollar that such a huge dataset would be fiscally impossible. If you gave them the Spotify rate, of $0.0005, the whole data set would only cost a bit over $3M, but I don't know who that's supposed to make whole.

 

Once the algorithm is trained, you'd know better than I if there would be any way to specify what works in the dataset were really used in the creation of a specific image, but my guess is that even if you could narrow it down (say, if a user asks for a Warhol like image), it's hard to imagine finding a model that generates much revenue for the creator.

 

I'm not really concerned about how it matches our current legal definition of "transformative work," because those can and often should change. I'm concerned with the impacts it has and what that means about the society in which we'll be living. The future we're heading toward if we don't change compensation and licensing is one where corporations can scrape the internet for artists work, generate "new art," use it in unbounded ways that profits them, and artists never get any kind of compensation. If your tool is valuable, you should compensate the people who made it possible instead of benefiting off their work for free, otherwise you're just further enriching the wealthy. (Unless they already agreed for their work to be license-free.)

 

Ultimately, I'm not really concerned with mankind's ability to figure out a way to monetize something like this either. We're kind of good at that  But if you asked me now, I probably wouldn't bother trying to figure out which art was associated with which individually generated image. I'd just tie it to the use of the generative art tool and the dataset on which it was trained.

 

[SuperSpreader]

5 hours ago, TwinIon said:

It's an interesting problem, but my initial reaction is very much that use in such a dataset to create AI art easily falls under the idea of a "transformative work."

 

Looking through the Dall-E paper, they mention on page 23 that they used datasets of 650M images. I don't know what kind of licensing fee would seem reasonable, but it's pretty obvious that if copyright holders made even a single dollar that such a huge dataset would be fiscally impossible. If you gave them the Spotify rate, of $0.0005, the whole data set would only cost a bit over $3M, but I don't know who that's supposed to make whole.

 

Once the algorithm is trained, you'd know better than I if there would be any way to specify what works in the dataset were really used in the creation of a specific image, but my guess is that even if you could narrow it down (say, if a user asks for a Warhol like image), it's hard to imagine finding a model that generates much revenue for the creator.

 

You'd also need to get permission to use the images.

 

Additionally you can point the software to images, so what if you use some that isn't yours? How does that work? 

[legend]

On 10/6/2022 at 2:49 PM, legend said:

I think AI art built from unlicensed datasets is a problem that needs addressing. But I also think it’s relatively easy to address. Make some legal regulation about requiring license for content to be used in a dataset. Then build websites where artists can submit their art with some standard licenses for use in datasets and pay people according that that license for the data. This is beneficial to everyone because it actually gives people an incentive to help build good datasets and distributes the wealth. 

 

 

Looks like Shutterstock is already starting to move in the direction of the kind of compensation model I was suggesting here. I really do think humanity will be able to figure out a compensation strategy because montization is what we're good at, and we should. We just need the law to catch up to further encourage it.

 

Shutterstock will start selling AI-generated stock imagery with help from OpenAI - The Verge

WWW.THEVERGE.COM

---

The stock image company is also promising to reimburse creators whose work is used to train AI art models with a new Contributors Fund.