I couldn't believe when I saw AI generated prints at Gagosian...

I believe creativity, concept and emotion are the last bastions against the robots ... Good art needs all. So i feel it will be used as a tool more than anything

Thanks for your reply and opinions alberti, to others as well. So input images/content can be traced, interesting, but how, isnt everything online technically in the public domain anyway or could be deemed fair usage? I could easily see Getty etc fighting a losing battle with this one. If this AI is so powerful and evolves surely it could just branch off and create new sites that do similar and better things and who would be responsible for those then from a legal stand point.
Commercial illustrators should be very worried then as this has just ended their careers and I suppose we should all be worried really. Its not the now we should be questioning but how this changes things 10 years down the line and beyond.
Scary times indeed.

“isnt everything online technically in the public domain anyway or could be deemed fair usage?”

Absolutely not the case - copyright is a huge issue, as anyone who has used a Getty Images licensed piece without paying them will attest. I have some skin in the game here as my current PhD thesis work is on an associated topic that also requires large data sets. What these AI companies have done is scrape copyrighted material off the web and used it to train their models. Note that there is no database that images or text can be removed from. The image requires training data, but it does not store it like a traditional database; instead each image or piece of text updates a huge set of parameters that the model uses to generate content when asked. The “Un-training” of models is incredibly hard, as it’s essentially a one-way process, unless you also keep model checkpoints to roll back to, which is incredibly resource intensive.

Imagine you are an AI company, and Getty get in touch and say, “you’ve used material I hold the copyright for in your model”. You check and the image in question was used sometime like a month ago. In the interim period you have done another training run and another 4bn images have been used to train your model. You either have to find some way of unlearning a specific data point, and more importantly PROVABLY removing it, or you have to roll back 4bn training points and start again. Won’t happen unless the courts specifically require it. Machine learning has come of age now, and it is too big to put back in the box. The outcome will be that large copyright holders such as Getty are either financially compensated, or they’ll take shares instead. Small copyright holders will get screwed as they don’t have the technical nous or the money to fight to prove they’ve been impacted. It’s shit, but that’s tech.

Thanks! As for the programmers, it’s mathematics getting in the way rather than a lack of talent. Plus most of what happens is essentially a “black box” - no one gets to see how the sausage is made, so to speak. We are not at a point of true artificial intelligence, but we don’t need that to see massive societal problems. Large Language Models like ChatGPT are prone to hallucinations, and image generating neural networks are just weird full stop; try using made-up words when asked for a description of what you want to draw and see what happens! The next 10 years are going to massively change how we interact with the Internet, and the results won’t be pretty. 

Yeah, US law is, as for many other topics, the outlier. EU law allows training for non-commercial purposes, and explicitly states that datasets should not be shared, and should be appropriately secured. My field, cybersecurity, has a similar issue with finding training sets large enough to prove useful, to the point that the creation of training data is publication-worthy in itself.

OpenAI made a submission to the US Patent Office (link) that states they believe that their training constitutes fair use as it is “transformative”. There is some cited case law that on the surface looks persuasive, but ultimately boils down to, “we want to”. Their statements to the effect that other remedies are available, like robots.txt files on websites to stop bots from scraping, suggests that the onus should be on the copyright owner, and that if the material is ingested it’s because their locks aren’t strong enough.

Where this will fall down, or at least where it should, is on the test of depriving an artist of livelihood. It is simple to perform extra training of a model against a particular artist and have it produce near identically styled images that are “new”. It’s hard not to argue that this is impactful to the original artist.

Academic use of data comes with significant guardrails to stop misuse. I don’t think fair use would ever impact this type of research. But if you are creating a commercial product, with the potential to impact the livelihood of someone who has provided training data without consent, then you are on shakier ground. Unfortunately it’s unlikely that the US courts will feel the same way, so it then becomes a battle between them and the EU, as with so much other tech.

I am literally impressed all day long just seeing the results of some of the Instagram artists I follow. Many are creating stories into their prices so I see a lot more graphic novels coming out of it. Too many to name and none stand out because they all look so good. Though there is a 'samey' feel to a lot of it. Not sure if that's because of the ones I gravitate to or limitations of what can be created.


I'm also seeing shades of animation in some of the pieces so get ready! Things are gonna get spicy!

Thanks for the link!!  Hmmm, robots.txt?  The onus should be on the scrapper/bots themselves, not the owners.  The only thing owners is to track bots activity and detect if they are bad bots or good, and possibly block the bad bots. We know that robotx.txt is not the best way to protect content and that blocking individuals is not possible.  Any one can create a scrapper with two lines of code (e.g. curl). And we know very well content that AI scrapped comes from bad bots from third parties companies or subcompanies. This claim is not vali

As long as work is non commercial, private and doesn't compete with other copyright owner in a similar market for commercial stuff, work is fine.  This said, as you pointed it out, any commercial artwork produced to replicate the style or the art of an artist may be seen a depriving artist from income and thus direct competition in a similar market.       But, fair use and all is really at the discretion of the judge and they are quite a lot of factors to consider; we have seen several judgments for similar cases, sometimes difficult to understand.   

Here are some more AI generated pieces by 'Magnus Gjoen',They look like old masters at first until you zoom in, quite interesting I think.

'Painted by Artificial Intelligence'

www.magnusgjoenart.com/ays

I'm not sure where anyone 'plucks there values from' these days! but they are all 'Unique' pieces, not additions, and are somewhat price-wise in alignment with his other works.
A 70x70cm Unique AI print from MG for £1000 or a 70x70cm Unique AI print on canvass from DH for £3500.

Yes, the lenticulars are great, I own quite a few of his works, His 'Skulls' being a particular favourite!......No AI pieces though btw 

Correction, these do in fact appear to be painted by Artificail Intelligence. The latest breakthrough.

Does this mean we can all give up on NFTs now . . .? Asking for a friend.