AI Art Generators are Here and They're Poised to Change the Art World Forever but an Increasingly Bizarre and Strange Range of Reactions from the General Public Threaten AI Before it Really Blossoms
I think this article glosses over a few really key points.
You say that ai art is better when humans are using it too, but that's just because it needs more data and learning, right? That's the whole point of it, so surely it will continue to grow and the impact the human artist would have would be negligible. What you mean to say is that human + ai is better *for now*.
You also suggest artists will have this as a tool in a tool belt, but again the real life implications are missing. An ai thay can work instantly at any time anywhere, constantly improves and encorporates ideas with a click, without need of a break, without needing compensation....and more points I am probably overlooking, and this will without a doubt cause many people to lose work. It will become a case of point and click and done. Why would work streams increase to the source artist when I can impersonate them with a fully crafted image before lunch?
Your analogy of technology evolving and people not knowing the opportunity yet again really misses the point. In a world where ai systems mass produce, how do you give space in the industry for human art? If you can't safeguard an area, how does a person's skills and talents have any room to flourish and develop? If you can't do that, how many aspiring creatives will you lose, because the employment opportunities dry up? Sure hobbyists and people who solely create for the pleasure of it will exist, but that doesn't put food on the table. Lack of opportunity for work = people look elsewhere for work = not spending as much time practicing and developing = full potential never reached.
If the suggested solution would be to train them in how to use the software, that would be like telling a bricklayer to go learn plumbing. It's still related to construction afterall.
One of biggest misconceptions I see every time this discussion arises is treating art as consumable, which it is not. No matter what tool I am using, AI or a brush the main goal of what I'm doing is to communicate. Also on the other side many times reading a book, watching a film or looking at an image I seek the meaning that the author wants to transfer through that. So as long as you treat your art as a consumable, as a pretty image to catch the attention and eventually the be bought: yes you might be doomed with AI development. But once you create to deliver some meaning and to communicate with others: AI is your biggest friend, cause it allows you to create this meaning easier, faster and often more accurately. And I definitely vote for this second role of art - freeing deep communication from the product-market logic it is bound to right now.
Regarding copyright, you might be wrong. None of the artists licensed their works to openAI or any other corporation doing image generators. Academic use of web scraping for datasets should fall under fair use, but this does not mean that openAI can release dallE into the wild and expect James Jean to sit on his arse and not sue the shit out of them. There is a clear predecessor for this in the case of google images which falls under fair use under very strict agreements (see here, for instance: https://www.quora.com/How-is-Google-Images-not-a-copyright-infringement-by-Google). As I said, none of the artists used in the dataset agreed to anything regarding dallE or Imagen or Parti or Stable Diffusion or whatever and this is one hell of a copyright lawsuit just waiting to happen.
Additionally, you regularily generate images that clearly show Shutterstock- or other stockphoto watermarks, albeit distorted. If i were Adobe or Shutterstock or Unsplash, i would sit on a mountain of evidence against openAI and Google right now, maybe wait a little more, and then sue the shit out of them. I mean, for them its either this or death of the business model. We know how these things go.
awesome research and cool outcomes. what will be the future direction all these datasets will be into . do you see any further consolidation as market driven approaches have curtailed reach to some extent the open-AI innovations. How can this take the art and design world to better creative enterprise coupled with human intellect and zapping of algorithms!
How come nobody is concerned about what future training sets are going to look like when some massive portion of the data will be AI generated. Will it degrade? Will there be 2nd generation and 3rd generation effects? Will AI generated output someday outnumber human generated content?
We'll only have this "virgin" human produced training set once and soon it will be gone.
I disagree that there isn't a potential ethical question about the use of training data. I myself have tuned AI on copywritten images as has OpenAI etc. Early commercial versions of AIDungeon even outputted lines from copywritten text. It happens, and you cannot handwave the actual issues away.
But a well trained AI is definitely not regurgitating images it trained on. The trained model is MANY orders of magnitude smaller than the dataset it trained on, and AIs do already create images far outside of anything that existed in the training data.
So we have to find a path between 'yes, this thing was trained for profit on information that wasn't paid for' and it isn't 'oh humans learn, too'.
> After getting crushed by the machine, Kasparov got back up and came up with the new idea of pairing supercomputers with grandmasters that he called freestyle chess. That evolved further over the years and by 2005 he created a full scale Centaur tournament, where humans, AI, or human-AI teams could enter to compete. The winner wasn't Hydra, the best chess program at the time and it wasn't a team of grandmasters and computers either. It was a couple of amateur chess players and some ordinary computers.
That was 17 years ago. How are centaurs doing now?
Great piece! Thanks.
I think this article glosses over a few really key points.
You say that ai art is better when humans are using it too, but that's just because it needs more data and learning, right? That's the whole point of it, so surely it will continue to grow and the impact the human artist would have would be negligible. What you mean to say is that human + ai is better *for now*.
You also suggest artists will have this as a tool in a tool belt, but again the real life implications are missing. An ai thay can work instantly at any time anywhere, constantly improves and encorporates ideas with a click, without need of a break, without needing compensation....and more points I am probably overlooking, and this will without a doubt cause many people to lose work. It will become a case of point and click and done. Why would work streams increase to the source artist when I can impersonate them with a fully crafted image before lunch?
Your analogy of technology evolving and people not knowing the opportunity yet again really misses the point. In a world where ai systems mass produce, how do you give space in the industry for human art? If you can't safeguard an area, how does a person's skills and talents have any room to flourish and develop? If you can't do that, how many aspiring creatives will you lose, because the employment opportunities dry up? Sure hobbyists and people who solely create for the pleasure of it will exist, but that doesn't put food on the table. Lack of opportunity for work = people look elsewhere for work = not spending as much time practicing and developing = full potential never reached.
If the suggested solution would be to train them in how to use the software, that would be like telling a bricklayer to go learn plumbing. It's still related to construction afterall.
One of biggest misconceptions I see every time this discussion arises is treating art as consumable, which it is not. No matter what tool I am using, AI or a brush the main goal of what I'm doing is to communicate. Also on the other side many times reading a book, watching a film or looking at an image I seek the meaning that the author wants to transfer through that. So as long as you treat your art as a consumable, as a pretty image to catch the attention and eventually the be bought: yes you might be doomed with AI development. But once you create to deliver some meaning and to communicate with others: AI is your biggest friend, cause it allows you to create this meaning easier, faster and often more accurately. And I definitely vote for this second role of art - freeing deep communication from the product-market logic it is bound to right now.
Regarding copyright, you might be wrong. None of the artists licensed their works to openAI or any other corporation doing image generators. Academic use of web scraping for datasets should fall under fair use, but this does not mean that openAI can release dallE into the wild and expect James Jean to sit on his arse and not sue the shit out of them. There is a clear predecessor for this in the case of google images which falls under fair use under very strict agreements (see here, for instance: https://www.quora.com/How-is-Google-Images-not-a-copyright-infringement-by-Google). As I said, none of the artists used in the dataset agreed to anything regarding dallE or Imagen or Parti or Stable Diffusion or whatever and this is one hell of a copyright lawsuit just waiting to happen.
Additionally, you regularily generate images that clearly show Shutterstock- or other stockphoto watermarks, albeit distorted. If i were Adobe or Shutterstock or Unsplash, i would sit on a mountain of evidence against openAI and Google right now, maybe wait a little more, and then sue the shit out of them. I mean, for them its either this or death of the business model. We know how these things go.
May you live in interesting times, they say ;)
Deeply researched and illuminating. Thanks for sharing!
awesome research and cool outcomes. what will be the future direction all these datasets will be into . do you see any further consolidation as market driven approaches have curtailed reach to some extent the open-AI innovations. How can this take the art and design world to better creative enterprise coupled with human intellect and zapping of algorithms!
How come nobody is concerned about what future training sets are going to look like when some massive portion of the data will be AI generated. Will it degrade? Will there be 2nd generation and 3rd generation effects? Will AI generated output someday outnumber human generated content?
We'll only have this "virgin" human produced training set once and soon it will be gone.
I disagree that there isn't a potential ethical question about the use of training data. I myself have tuned AI on copywritten images as has OpenAI etc. Early commercial versions of AIDungeon even outputted lines from copywritten text. It happens, and you cannot handwave the actual issues away.
But a well trained AI is definitely not regurgitating images it trained on. The trained model is MANY orders of magnitude smaller than the dataset it trained on, and AIs do already create images far outside of anything that existed in the training data.
So we have to find a path between 'yes, this thing was trained for profit on information that wasn't paid for' and it isn't 'oh humans learn, too'.
> After getting crushed by the machine, Kasparov got back up and came up with the new idea of pairing supercomputers with grandmasters that he called freestyle chess. That evolved further over the years and by 2005 he created a full scale Centaur tournament, where humans, AI, or human-AI teams could enter to compete. The winner wasn't Hydra, the best chess program at the time and it wasn't a team of grandmasters and computers either. It was a couple of amateur chess players and some ordinary computers.
That was 17 years ago. How are centaurs doing now?