Hey Guys, Andy here. The topic of AI generated art has been gaining speed recently and, in true tech form, generated more questions than answers as those it affects try to foresee where the path will take us. A pressing concern is copyright though when is copyright and its protection never a concern?
For those unfamiliar with the topic there exists a slew of apps, developed through machine learning, that generate digital images based from text prompts. While it can be argued that a more accurate term is procedural generation, such apps are more commonly called AI art generators.
Such work recently and controversially won first place in the digital category at the Colorado State Fair. The controversy comes from the fact the art was created via a language prompt and had little to do with artistic talent. The artist claims his input was key to the award and that the work should be evaluated equally next to normal art. The definition of art notwithstanding, many in the creative industry are wondering how long their jobs will continue to exist.
While the destination of AI art is unclear the origin of the art is less so, although it is equally controversial. The apps scrap, or extract, data from existing sources on the internet. At time of writing this scrapping has been done without permission of the artists and there does not appear to be an opt out. That is to say the apps and work produced from them are being monetised without any financial recompense to the underlying creators, from which the machines have learned, which has led to concern from the creative industry.
This unease has led to the creation of the reverse search engine Have I Been Trained, which will check several of the AI apps if your work has been used in the learning. A notable return of the search has come in the form of the Polish artist Greg Rutowski, known for his distinctive fantasy landscape style. His name has been used in almost 10% of the 10 million prompts on the Stable Diffusion app. The artist initially believed this would be a novel way of reaching a new audience however the work returned had his name attached yet was not, in fact, his, rather it had been generated in his style. While there is no universal standard in regards to rights of work produced by AI art generators, the default seems to be you have rights over the works while the apps also claim rights and the right to allow others to remix your work. There does not, yet, seem to be any convention in regards to credit.
In addition to AI art potentially swamping genuine SEO returns there is also the issue of the data set that has been used in the learning. According to Arthur Holland Michel, a wider dive on Have I Been Trained shows an additional issue; “datasets are plagued with problems including curation biases, inclusion of problematic content in the images, as well as contributing to the gradual erosion of privacy.” He goes on to add, as an example;
“If you search “Michael,” you get a ton of memes, stock pictures, and headshots of people presumably called Michael. Nothing pornographic whatsoever. But when you search “Michaela,” you get—no joke—90% porny results.”
Such issues are deeply troubling to creatives and beg the questions; if creators were being paid for machine learning, would any accept if they knew there was a potential of their work disappearing under a tsunami of generated work. And, would you want your work sitting along side NSFW images?
What are the options? Many creatives use the internet to reach their audiences and gain sales. And yet, by being present on the world wide web, there are risks of infringement by multiple attack vectors. It is a case of damned if you do and damned if you do not. Level of attention and care is probably proportional to the impact you think it will have on you and whether you see an opportunity. Are you willing to pivot again? Or will you accept some level impact if it means you can remain a photographer and pays the bills? In short, what level of risk and infringement are you comfortable with?
Abuse has already come full circle with AI artists protective of their rights in regards to prompts; oh the irony of being protective of text prompts when the underlying works have been infringed upon.
Why does copyright and protection matter? A tech friend regularly berates me for, what he sees as, introducing friction to my workflow; he believes such friction hinders me from gaining work. Of course he is right, it does. Who wants to deal with IP when you will be able to find someone to waive rights for free?
However, I argue, where my friend sees friction I see structure. I regard terms and conditions, including licensing and protection of work, as goalposts from which to establish a relationship and trust. If the goalposts move, then work becomes difficult to achieve and trust becomes challenging to build. Work without trust or a solid foundation is exploitation.
Legislation is already having trouble to keep up with the pace of technology and brand new situations which fall outside the remit of existing laws. It is a magic trick pulled by the tech industry; profit off copyright infringement, keep the public in the dark about their creative rights and then kick the can down the road. The issues of rights is therefore never addressed. As a creative, I doubt the path of AI art is going to end well.
Accompanying image generated by MidJourney using the prompt ‘close_up_world_war_2_Spitfire_fighter_plane_black_and_white’