Stop Using AI-Generated Images
Day 11 of National Blog Posting Month #NaBloPoMo
Recently I’ve been doing some front end development work for a lovely project that showcases original short stories submitted by writers and climate change activists. Most of the stories are accompanied by an illustration specially created by an artist. One story did not have an accompanying image, so the author suggested using an illustration produced by an AI image generator.
No doubt you won’t have failed to notice how AI (artificial intelligence) is proliferating in almost every area of our lives. At FF Conf yesterday I spoke with at least 4 people (myself included) who work for companies that specialise in AI. There has been a lot of discussion (and soul-searching) about how LLMs (Large Language Models) like Chat-GPT will impact the internet, with many warning of the dangers of rampant misinformation. Plenty of writers are rightly worried about their livelihoods. The Hollywood writers’ strike is, in part, triggered by the concerns of writers about how AI will affect their jobs. Maggie Appleton explored the topic in depth in her talk at FF Conference, and what some possible futures might look like.
AI image generators are seemingly of less concern. As we can see in the example at the beginning of this article, many authors who are concerned about AI subsuming their own work have few qualms about using AI image generators, which threaten to displace illustrators in the same way. I don’t believe this is conscious, for the most part. I think it’s likely that they view the creative process and output of illustrators differently from their own, where they consider it at all.
How AI image generators are destroying illustrators’ livelihoods
As a former illustrator, I confess to having some skin in the game here, and perhaps have a little more insight than most into the potential impact of AI on the industry. Just like LLMs, AI image generators are trained on image data scraped from the internet, much of it copyrighted, and almost all of it without the artists’ permission. Artists train for years, sometimes decades, to hone their craft in a precarious and low-paid industry. Without their skill, which produces the training data, these models simply wouldn’t be possible. And yet the artists are not compensated or credited in any way.
A common way for illustrators to break into the industry is through editorial commissions that might pay a couple of hundred pounds for a few days’ work, if they’re lucky. Illustrators need to work fast if they want to earn a decent living. Highly detailed, rich, photorealistic illustrations like the ones frequently produced by image generators can’t realistically be produced by most human illustrators in a short space of time, and so most editorial commission tend to feature styles that can be produced quickly, which tend towards simple and blocky. (Make no mistake though, communicating complex concepts in a simple, minimal style is a skill in itself.)
By choosing AI-generated images over those made by a human, editors are taking away one of the only sources of income available to those entering the field. This means that fewer people will choose to train in the profession. It will disproportionately affect those from low-income backgrounds, unable to justify the high cost of an art education with little prospect of earning a living. Illustration will become the preserve of the already well-off.
AI image generators suffer from a similar set of problems as LLMs like Chat-GPT, in that they can fill the internet with false information, so that you can no longer believe what you see. Add to that the ability to generate deepfakes and even child pornography (which there is already evidence they are being used for).
AI tends to sand away the unusual
Ben Myers explores this from his perspective in his insightful article I’m a Spotless Giraffe.
Removing the human
When I look at a piece of art, I’m looking at all the ways the human experience has shaped what I’m seeing. I’m admiring the skill, sure, but I’m also seeing the invisible human hand, the mind that has made the creative leaps. I feel connected to the artist, because they are telling me something about themselves.
AI art, by comparison, has no soul. AI can mimic. It can look impressive. It can’t feel the depth of human emotion, or distill the experience of being alive in a particular time and place into something more than the sum of its parts. I have yet to meet anyone who wants to hang AI art on their walls (although I fully expect to see it in hotel chains).
AI art will eat itself
It’s pretty easy to spot AI-generated art at the moment. A lot of it has an eerie, hyper-real aesthetic. Many details in the background or foreground can appear hazy and dream-like. Often you can more or less guess the exact prompt that was fed into the model in order to generate the image.
Perhaps AI will get better at mimicking actual human creativity. The flip-side is that as the web becomes awash with these kind of images, more models will be trained on them. The AI starts to consume its own output. I’m not sure what that future looks like.
Artists are resisting
The problems with AI image generators are explored in greater detail in a fascinating episode of the podcast Tech Won’t Save Us, featuring illustrator Molly Crabapple.
The law hasn’t yet caught up, but artists are finding ways to resist. There have been lawsuits filed against AI companies, including stock photography company Getty Images against Stable Diffusion. A new tool has been developed that invisibly alters how images will be interpreted by AI models when fed in as training data.
We have a choice
AI can be used by artists to augment their work and explore an exciting, new, creative frontier. But the vast majority of “AI art” I’ve seen is not that. It’s theft. You don’t have a right to “free” art, in the same way you don’t have a right to have your car fixed for free.
The proliferation of AI content on the web may feel all but inevitable, and perhaps it is. But we can choose whether to participate in that harm or not. We can ask ourselves, “Is this the web we want”?
Webmentions for this pageAbout webmentions