Artists often put hours into each piece they create, not wanting to share it with the world’s eyes until every last detail and brushstroke they put in is at the exact place it needs to be. Now they are seeing AI art generators who can churn out hundreds of pieces within a minute.
Artificial Intelligence (AI) has recently started gaining fame in mainstream media, with all sorts of new programs being able to do things independently, like converse with the user or create art pieces. Bots, AI art generating programs like DALL-E, NightCafe, and filters on TikTok and Instagram are leading some to question what the overall effect on human production will ultimately be.
8forty interviewed two high school students, both artists who’ve spoken about this topic on their social media platforms, and knowledgeable AI users who are interested in tech and AI. They told 8forty how they feel about AI art. All names in this article are nicknames or online personas.
The way AI art generators are trained is similar to a really large internet library that collects data. It takes millions of images and links each one to a human description; the human description is fed into one side of a neural network, and the images are compared on the other side. The computer makes random changes to the neural network to try and match the expected image. What this means is that the AI isn’t necessarily taking art directly, rather it’s learning a specific aesthetic through analysing thousands of images, then generating images based on the aesthetics and patterns it’s learned. When you put in a prompt the AI’s never seen before, it’ll generate an image with a combination of patterns it has seen before, and generate a weird mashup of images.
Some AI art generators like DALL-E have other OpenAI models analysing these images for it. DALL-E uses a program called CLIP, which stands for Contrastive Language-Image Pre-training, and this OpenAI is what’s being trained on the millions of images and the text prompts that are connected. DALL-E’s entire functionality is dependent on CLIP’s training, and its performance depends on how well CLIP is trained.
Some artists believe that the AI’s use of images on the internet constitutes copyright infringement.
Aosia, a young artist, believes although both artists and AI use other’s art to learn, there is a large difference between artists using pre-existing art as a reference and AI art generators using the same pieces.
“A reference image is what you’re drawing based on, [but] AI will indirectly use millions of images, but not directly,” Aosia told us. “It’s technically still indirect because it doesn’t ‘rip directly’ from images but it uses those images to create a concept based on the image.”
For many artists, a large part of art is originality. What Aosia talks about is how AI art always generates the same sort of images when given a prompt. “Apples will always be circular, bumpy, etc. but a million times more precise.”
Even when artists aren’t creating original art, they credit the reference pieces they use. Artists generally respect each other and their creations, as they’re all coming from similar places. AI, however, takes these pieces and sort of mashes them together, and doesn’t credit anyone.
There’s even been many instances where people have found what appear to be artists’ signatures in the AI art pieces, though they may simply be the result of the AI creating signature since that is a feature of many of the images it is trained on.
There’s also been a recent movement where artists post a sign with the letters “AI” in bold inside a large red circle. This movement started on Twitter, when an artist with the username Artofinca tweeted about this issue with an image he created. The image sparked a trend among artists, using this picture to show support, with either the image they created posted with art, or artists draw variations of this image.
Surprisingly, the other student we interviewed had views that were very different from Aosia’s.
“Looking at the current art-generation APIs out there on the internet, I don’t believe that the artists the machines were trained off of need to be paid,” Bread told us, speaking on the issue of AI ‘copying’ people’s art directly. “This is because the machine doesn’t ‘save’ the images it trains itself off of – instead, it looks for patterns in it’s dataset… [and] other than using the images to compare its current idea of what good ‘art’ looks like, the AI does not interact with the images in its dataset.”
What Bread believes is that what the AI does with our art is very similar to what humans do when using others’ art to create their own, and at this point in AI’s development and uses, artists shouldn’t worry about AIs overtaking human artists.
Bread also has a rather optimistic view on what this means for hobby artists, telling us how they believe people shouldn’t lose confidence in their art even if they deem AI art to be better than their own creations.
“AI generated images serve an extremely different demographic than human-made art,” they told 8forty. “People shouldn’t worry! At this stage, AI can’t express emotion the way humans do… because of that, human art stands out by the masses from AI.”
Bread thinks that even if art generated by AI appears to be impressive, you should still continue and persevere. “Why stop, even if you deem that art ‘better’? Refusing to participate because you won’t come in first in your head isn’t the way to grow in any situation whatsoever,” Bread says. “Even if you do like the AI’s pretty pictures better than your own, learn from it! It’s giving you unlimited references just by typing a single prompt.”
“Don’t look at it like a chokehold for opportunities, rather, see it as a gate that opens up to unlimited free resources to learn off of yourself.”