Ghosts in our Machines
A thought experiment on the role of artists in the business of generative AI.
In the wee small hours of January 1st, 2023, an implausibly particular and peculiar cataclysm wiped out all human art, all digital representations of that art, all ML art training datasets, and all pre-trained generative art model checkpoints. As the head of a generative ML tech company, you see it as both a business and moral imperative to get humanity back to where it was before “The Big Deletion”. You know there are still millions of artists in the world, and you know you have the engineers, GPUs, model architectures, and training code to replace a chunk of what was lost.
What do you do? This is no chicken-and-egg problem. Without data all those GPUs are useless, and the engineers are going to burn through whatever cash you have on hand by year end. Because you’re an engineer yourself, you already understand that you don’t need a million Rembrandts, Van Goghs, and Mohrbacher’s to get started, since outliers like them account for only a small portion of the lost art/data. What you need, rather, is a decent sampling of capable and productive artists from the community as a whole—e.g., a crowdsourced work force.
Your thinking could go in a couple of directions:
“Well, artists do what they do for love, so they’ll just start again and if I wait a year or two, we’ll probably have a half-decent dataset for free. At our current burn rate we’ll be broke by December, but since the engineering team are now sitting on their hands, I guess I could lay them off and ride it out.
Of course, there’s the coming flood of copyright litigation to consider—it’s dangerous to assume that artists will be so naive about distributing their work a second time around.”
or
“If I jump-start the process by publicly incentivizing artists with a royalty, in the next few months we could have a decent dataset to start training. This would get my engineers back to work and, as a bonus, we’d have an enthusiastic community of artists helping us improve our dataset on an ongoing basis.”
Obviously the premise is beyond far-fetched, but the point is to dispel the deeply ingrained notion that art is essentially a free and common good—a pre-existent and renewable resource to be extracted for the minimal cost of scraping files from the internet. I don’t necessarily disagree with the utopian ideal that art should be free (and I say that as a working, professional composer myself), and it is undeniably true that artists will make their work regardless of financial compensation. But when we’re talking about a billion-dollar industry, and we acknowledge that artists are an indispensable part of the generative ML pipeline and are humans who need money to survive—just like every engineer on our hypothetical CEO’s team—it doesn’t take a genius to recognize that incentivizing the art community’s innate enthusiasm to create is not only the ethically sound way forward, it’s also a shrewd business move.
It’s pretty simple: Treat artists well and their inventiveness and resourcefulness will pay dividends in the long run.
