Here is an article from Tristan Péloquin from LaPresse
The art of copying without paying
https://www.lapresse.ca/actualites/enqu ... -payer.php
Google and free translation.
----------------
In this article, the author says that 'If a picture is worth a thousand words, with artificial intelligence, it is now enough to write a short sentence in a dialog box to generate endless stunning images.'
Algorithms are being used to generate images that look like the 'real artist'. And applied to astrology, tomorrow a software will become an astrologer able to master the most complex systems like Indian astrology.
When the AI swallows the works without the knowledge of the artists
“The operation of the software is confusingly simple. Imagine a scene, mundane, fantastic or surreal, and type its description in a dialog box. Add the name of the great painter you want to imitate, then wait a few seconds.
If one of the four images that appear on the screen does not satisfy you, simply reformulate and start again. Again and again.???
Here is an example of 2,421 results of images of Marylin Monroe generated by Stable Diffusion, from the British company Stability AI and posted on lexica.art
https://lexica.art/?q=marylin+monroe
“For a little over a month, the result of this effortless operation, repeated hundreds of thousands of times by Internet users from all over the planet, has given birth to what is certainly one of the largest image galleries in art that humanity has been able to see to date. The directory, published on the Lexica.art site, already contains hundreds of improbable portraits of Justin Trudeau posing with an assault rifle, thousands of fantastic urban landscapes inspired by the real world, and gigabytes of images reconstructing as many exploits of phantasmagorical hockey players, illustrations of historic battles from the Vietnam War or drawings of Celine Dion as the heroine of the series The Walking Dead.???
“The artificial intelligence matrix behind this feat, Stable Diffusion, from the British company Stability AI, is not the only one to compete for the nascent market of art created by natural language (“ text-to-image art???) ) . Its competitors, Midjourney, launched in March, and Dall-E, which appeared in April, are also establishing themselves with funding rounds raising hundreds of millions in venture capital. Google promises a replica that will be called Imagen.
But there is a hitch.???
In order for them to be able to generate art on demand, the developers of this software first had to make them swallow millions of descriptive photos and existing works of art, in order to train their "neural network" to “understand??? and “interpret??? the world.
Stable Diffusion has revealed that the 2.3 million images that were used for its training come from a gigantic scrape passed on the web thanks to indexing robots ( web crawlers). The group of university researchers who carried out the operation never asked the slightest authorization from the tens of thousands of artists who hold the rights to these works.
“People thought the images came from public databases, but that's not the case. I thought it should be known to the population and the artists affected, ???says Andy Baio, influential technology blogger and former administrator of the crowdfunding site Kickstarter, who put online, in early September, a search engine allowing you to search in the database of images thus gobbled up by Stable Diffusion. And they didn't ask for permission.
Nevertheless, the result is pleasing to the eye.
After a few tests with the software, you can create a piece of art that looks like Dali.
New "artists" dialog box
The Montreal Institute for Ethics in Artificial Intelligence, which is taking a close interest in the question, recognizes that the unprecedented “scale and pace??? at which these machines are capable of producing will shake up certain artistic professions.
“But that doesn't mean humans aren't involved. There is a new emerging field related to AI, which is called prompt engineering , which is in itself an art form,??? says Abhishek Gupta, Principal Investigator and Director of the Institute.
On the Reddit and Discord forums, these “natural language art artists??? share tips and key phrases for “creating??? ever more striking works with AI. “It will become a skill that can be learned and improved with practice,??? predicts Mr. Gupta. Eventually there will be a class of people who can be employable with these skills. »
But until then, it will first be necessary to settle the thorny issue of copyright, he recognizes.
Already, the major photo and creative image agencies Getty Images and Shutterstock, whose millions of protected images have been swallowed by Stable Diffusion, have announced that they are banning everything created by artificial intelligence from their catalogs. . Andy Baio expects one of the two companies to take legal action against Stable Diffusion. “They are the only ones with the resources to do it,??? he believes.
Too late to “unlearn???
For artists, on the other hand, legal recourse, which involves attacking a company established abroad, promises to be more difficult, estimates Karim Benyekhlef, director of the Cyberjustice Laboratory at the University of Montreal. “Whether their works end up in a database of 6 million, 600 million or 6 billion images has no bearing on the seriousness of the violation of intellectual property,??? he stresses.
But we are here in the Californian logic [of Silicon Valley]: these companies go for it, break the rules, and it is part of their calculation to tell themselves that in the end, there is only one hundreds of artists who will be strong enough to pursue them, so the game is worth the effort.
And anyway, the genie is already out of the bottle. "Even if artists manage to have their artwork removed from the database, at this point, unless you recreate the machine's artificial intelligence from scratch, it cannot 'unlearn' what she has already learned,??? assures Andy Baio.???