DALL-E 2 was released / unveiled early april 2022 and some artists did not take well to the news of a new, "better" model. i personally didn't take well to either the hype or the panic and have been wanting to write on the subject since then. so — and take this with a grain of salt because i'm, after all, just some guy — i just want to quickly break down a few reasons why i highly doubt artists will be made obsolete and what you should be worried about instead. i'll make it short for now but i might update and expand this page later.
i should also note that when i say artists i'm mostly talking about "commercial art" rather than gallery work, even if the NFT crowd wants you to think oil-on-canvas is heading towards its grave.
first things first, DALLE-2 had a very controlled release. the website is semi-interactive, ie you click parts of prompts to see images generated and curated beforehand. images posted on twitter as proof of the program's "skills" as an artist looked very impressive on mobile, during the 10 seconds you look at an image on twitter, but further scrutiny reveals that visual indeterminacy (images that seem coherent at first, but aren't really) is still very much a thing, and an issue if you're looking to pass the turing test, as it were.
curation is a necessary process of working with AI-generation. curation is present every step of the way, from assembly of the datasets (a task on its own right) to picking what prompt to use (and how to word it, what keywords to steer the style of the image) to picking what images actually look good. in order for artificial intelligence to actually look artificial and intelligent, curation has to be obscured, the human labor (particularly the crowdsourced microtasks, such as image labeling) ignored. ML is very very far from being immune to alienation.
so, i mean two things with this section: 1) image generating networks / programs are being made to be more intelligent than they really are with careful curation and 2) if all else fails and using AI for art becomes the industry standard (and that's a really big if), artists would still have the role of curator and fixer of indeterminacy. a shitty role? yes, but even in a worst case scenario such as this one, we're not extinct (just exploited in a different way).
the output of a neural network is limited by its input, aka the dataset. you may argue that it's like this with humans too: all art is mostly a recombination of our repertoires, but where humans and algorithms differ is that our "datasets" are being constantly updated as well as comparatively unlimited. ML datasets like imagenet (itself based on wordnet) intend to be a representation of the real world, but that's simply impossible. magritte said it nearly a hundred years ago: ceci n'est pas une pipe.
GANs are based on datasets, and the datasets themselves are frozen pictures of the time they were assembled. they are biased towards realism (because most if not all images in the dataset are photographs), style transfer can only attempt to recreate other artists, since there's no thought there isn't metaphor, allegory, symbolism. and i love the aesthetics that can come from GAN art! i love the indeterminacy, i love the uncanny, i love the abstract! there's a reason i work with it and read so much about it and it's not hate. but these intrinsic limitations also mean AI is a genie when it comes to trying to get specific results, as you'd want for illustration or concept art or basically any commercial work. so isn't it much easier to talk to another human being?
we like to pretend — or rather, we are made to believe — that computers are pure cloud. the parts that make electronic devices and the infrastructure powering servers are abstracted away. as we (or rather, they) more than moore's law our way into more powerful computers, larger neural networks, higher image resolutions and so on, we simply assume it's possible to have infinite growth in the cloud, forgetting / being made to forget that this is all grounded in the physical world of natural resources and like, electricity. in fact, it's simply impossible to account for all the cogs in this global machine.
i also have to note the ethical complications that exist in every step of the way. from the exploitation of the miners and factory workers actually making electronics, to turning everything online into "data" free for the taking when constructing datasets, to exploiting the workers labeling this data / pretending to be AIs, to the absurd computational resources it takes to train the networks themselves, and then generate images (especially as we seek 4k. why is it always 4k). the biases amplified and deepened by the datasets (excavating AI talks about this mess).
i don't want to talk at length about this, because more competent people have done it and i'd just be parroting them, and also because it's a little beyond the scope of this little article. but i can't just talk about what isn't an issue and make it seem as though AI isn't ripe with problems, both built-in the tech industry at large and some brand new ones.
i believe the panic surrounding AI comes from a very emotional knee-jerk reaction. most people don't really know whay ML can actually do, and AI carries a baggage of sci-fi and malicious machines and so on. and while tech is far from neutral, our enemy isn't the tools themselves, but the people and corporations pushing for automation. i don't think "AI will replace artists" was a conscious agenda, but it does push artists towards "getting with the times" and become programmers, NFT shills (oh yeah that's a pet theory of mine), seeing ourselves as content machines, feeling more desperate about job security and as such, easier to exploit.
"AI will replace artists" paints a future of full automation, and it isn't even the sexy space communism sort, it's just automation for its own sake, automation because it sells more, new tech. it will replace us, they say, emphasis on will because this future is inevitable, no use trying to legislate, no use trying to question the ethics, that's just stalling for time, the future is a mars colony. if the future is inevitable you have no reason to fight it.
the thing is, AI isn't a future inevitability but a present matter. much like computers themselves, no one is under any obligation to use it for their art, but the world is already permeated adn shaped by algorithms. when we generate images or text through a neural network, that's a very tangible scifi-y result, but not all ML is this obvious.
if you're interested in the subject of ML, and you most definitely should be at least a little bit, because we need more tech literacy in general and AI literacy in particular, i recommend a couple of books: atlas of AI, by kate crawford and you look like a thing and i love you, by janelle shane. these are pretty good introductions to what AI is, what it isn't and what problems it brings. the good, the bad and the ugly, if you will.
sorry to be cheesy, but panic is fought with knowledge. it's important to understand what we have to be worried about and afraid of, and what is just fearmongering, otherwise the situation seems simply hopeless and we can't really construct anything.