WChicken concept artist and painter RJ Palmer He witnessed for the first time the accurate, realistic image of the compositions produced by AI image creator Dall-E 2, and he felt uncomfortable. The tool, released by artificial intelligence research firm OpenAI, showed a significant improvement over Dall-E for 2021, and competitors such as Stable Diffusion and Midjourney soon followed suit. Write Any Surreal Wave, by Kermit the Frog Edvard Munch styleto Gollum from the Lord of the Rings Feeding a slice of watermelon, these gadgets will bring back an amazingly accurate shot moments later.
The Internet has enjoyed memes-making opportunities, with a Twitter account documenting “weird Dall-E generations” who have amassed over a million followers. worldwide shake the world The first magazine cover created by artificial intelligence, tech investors have signed themselves up to wave into the new era of “generative artificial intelligence.” Image creation capabilities in video have already proliferated, with the release of Imagen Video by Google and Meta Make-A-Video.
But the new ingenuity of artificial intelligence has not been well received by some of its creators. “The main concern for me is what this will do to the future of not just my industry, but the creative human industries in general,” Palmer says.
By ingesting large data sets to analyze patterns and build predictive models, AI has long proven to be superior to humans at some tasks. It’s this amazing number that led AI to overtake world champion Go Back in 2016Quick computing of the most useful game strategies, and not afraid to perform moves that would have been ridiculous if they had come from a person. But until recently, producing original output, especially creative work, was considered a clearly humane endeavor.
Improvements in artificial intelligence have transformed telephony. AI image generators can now not only convert written phrases into new images, but great strides have been made in AI speech generation as well: large language models like GPT-3 Have reached a level of convincing fluency At least one researcher was recently fired from Google From the feel of the machine. Plug in Bach’s works, and AI can improvise music In the same style more or less With the caveat that it is often impossible for a human orchestra to play.
This class of technology is known as generative AI, and it works through a process known as diffusion. Basically, huge data sets are put together to train the AI, and through a technical process, the AI is able to create new content that is similar to the training data but not identical. Once he sees millions of photos of dogs tagged with the word “dog,” he’s able to pixelate a brand-new puppy that resembles the dataset closely enough that we have no problem classifying it as a dog. It’s not perfect – AI photo tools still struggle with rendering hands that look human, body proportions can be off, and they have a habit of producing nonsense writing.
While netizens have embraced this superior creative potential — armed with the properly polished router, even beginners can now create engaging digital paintings — some artists have dismissed the new technology’s ability to imitate. Among the claims entered into the image generators Stable Diffusion and Midjourney, several tags indicate an artist’s name in order to ensure a more aesthetic style of the resulting image. Something as plain as a bowl of oranges can become eye-catching if presented in the style of Picasso, for example. Because the AI has been trained on billions of images, some of which are copyrighted by living artists, it can generally create a very honest approximation.
Some are angry at what they see as theft of their art. Greg Rutkowski, concept artist and illustrator well known for his epic fantasy scenes with golden light, is mentioned in hundreds of thousands of claims in use across Midjourney and Stable Diffusion. “It’s only been a month. How about a year? I probably won’t be able to find my work there because of [the internet] You will be filled with the art of artificial intelligence,” Rutkowski said MIT Technology Review. “This is worrying.”
Dall-E 2 is a black box, where OpenAI refuses to release code or share data on which tools have been trained. But Stable Diffusion chose to open its source code and share details of the image database used to train its model.
Spawning, an artist group, has built a tool called Was I trained? To help artists discover if their artwork is among the 5.8 billion images used for Stable Diffusion training, and to opt in or out of future training sets. The company behind Stable Diffusion, Stability AI, said it is open to working with the tool. Of the 1,800 artists who have signed up to use the tool already, Matthew Dreyhurst, an academic and member of Spawning, says it’s split 60/40 in favor of opting out.
but the concept art association The CAA maintains that the damage was done this time around, because the tools had already been trained on the artists’ work without their consent. “It’s like someone actually robbed you saying, ‘Do you want to pull out of my robbery for you?'” says Carla Ortiz, an illustrator and board member at CAA. “
Emad Mostaque of Stability AI says that while the data used to train Stable Diffusion did not offer the option to opt out, it “was a largely test model, not significantly improved in online snapshots.” He says new models are usually trained on new data sets and that’s when the company takes artists’ requests into account.
And it’s not just works of art: Analysis of Stable Diffusion’s training database shows it’s also sucked into privacy Medical ImagingPictures of members of the public (sometimes alongside their full names) and pornographic material.
Ortiz particularly objects to Stability AI, which is marketing part of its business – DreamStudio, which provides customers with custom templates and improved ease of use. “These companies have now set a precedent for you to use your private, copyrighted data without anyone choosing it,” she says. “Then they say: We can’t do anything about it, the genie is out of the bottle!”
What can be done about it other than relying on the benevolence of the companies behind these tools remains in question.
The Civil Aviation Authority cites worrying legislation in the UK that could allow AI companies more freedom to suck up copyrighted creative work to train tools that can be deployed commercially. In the US, the organization has met with government officials to talk about copyright law, and is currently in talks with lobbyists in Washington to discuss how to roll back this as an industry.
In addition to imitation, there is a bigger problem that Palmer identified: Do these tools put an entire class of creators at risk? In some cases, artificial intelligence can be used in place of stored images – the Shutterstock photo library recently struck a deal with OpenAI to integrate Dall-E into its product. But Palmer argues that artwork such as illustration for articles, books or album covers may soon face competition from artificial intelligence, undermining a thriving area of commercial art.
Owners of AI image generators tend to say that these tools, on the contrary, democratize art. “A lot of the world struggles with constipation creatively,” Stability AI founder Imad Mostaqi, He said at a recent event To celebrate their $101 million fundraising round, “And we’re going to do it so they can bring out rainbows.” But if everyone can harness AI to create technically ingenious images, what does it say about the essence of creativity?
Anna RiddlerD., an artist known for her work with artificial intelligence, says that although the Dall-E 2 feels “magic” the first time she uses it, she has so far not felt a bit inspired in her experiments with the tool. She prefers working with another type of AI called Generative Adversarial Networks (GANs). GANs act as an exchange between two networks, one of which creates new images, and the other that decides how well the image meets a specific goal. The goal of a technical GAN might be to create something as different as possible from its training data without leaving the category of what humans would consider visual art.
These issues have intensified the debate about the extent to which we can rely on AI for creativity. according to Marcos de SautoyOxford University mathematician and author The Law of Creativity: How Artificial Intelligence Learns to Write, Draw and ThinkAnd the Dall-E and other image generators are likely to be as close as possible to replicating some kind of “harmonic” creativity, because algorithms are taught to create new images in the same manner as the millions of others in the training data. GANs of the kind that Riddler works with, he says, are closer to “transformational” creativity – creating something in a completely new style.
Riddler objects to such an approach to defining creativity. “It settles it down to thinking of art as an interesting background, not something that tries to express ideas and search for truth,” she says. As a conceptual artist, she is well aware of the shortcomings of AI. “Artificial intelligence cannot handle concepts: moments of collapse in time, memory, thoughts, emotions – all this is a real human skill, making a piece of art rather than something visually beautiful.”
AI image tools show some of these shortcomings. Whereas the phrase “astronaut riding a horse” will return an accurate display, “astronaut riding a horse” will display images that look very similar – suggesting that the AI doesn’t really understand Causal relationships between different actors in the world.
Dreyhurst and Riddler claim that the idea of ”replacing the artist” stems from an underestimation of the artistic process. Dreyhurst regrets what he sees when the media teases disturbing accounts, highlighting a talk The New York Times Article – Commodity About an artist who used Midjourney to win the digital category for the Colorado State Fair’s annual art competition. Dryhurst points out that the state fair isn’t exactly a prestigious forum. “They were giving out prizes for canned fruit,” he says. “What bothers me is that there seems to be this kind of thirst to intimidate artists.”
“Art is dead, dude.” Said the winner of the state fair.
It is possible that the hype about these tools as disruptive forces outweighs the reality. Mostaque says AI image generators are part of what he calls “smart media”, which presents a “trillion dollar opportunity”, citing Disney’s content budget of more than $10 billion (£8.7 billion), and the value of the entire game industry at more than $170 billion. . “Every bit of content from BBC to Disney these models will make it interactive,” he says.
Emerging applications right now are more realistic including mood boards for design consultancy, billboards for films, and models for interior design, and Mark Pique, an analyst in Omdia’s AI division, doubts the $1 trillion figure. “What are the killer use cases here?” He says. “It doesn’t make sense. What problem are you solving with this?” An analyst from consultancy Accenture says the tools could one day be used to create content to train machine learning algorithms, such as self-driving vehicles, and speed up game creation. It remains to be seen if it will come up with anything profitable like AI image generators and backers.