The end of 'who made this'?
Generative AI and ownership
“Good artists copy, great artists steal” gets attributed to Pablo Picasso quite a lot. Mainly because Steve Jobs credited it to him when talking to a newspaper about copyright. The quote may have originated with TS Elliot’s “Immature poets imitate; mature poets steal. Bad poets deface what they take, and good poets make it into something better, or at least something different.” And Elliot may have been inspired by the writings of W. H. Davenport Adams. This is all to say, the history of art and identifying and crediting the inspiration for art is complicated and full of both blatant dishonesty and inspired reimaginings of older ideas.
That constant building on older foundations, or stealing outright of existing blueprints, has gone into overdrive in the Internet Age. If you chat with any artists, whatever the format, ideas about ownership and crediting will often be shaped by just how online they are. A creator on TikTok will be aware a viral moment could come from a stranger downloading their video and posting it on a different platform. It’s at once precarious and hopeful and depressing.
This brings us to the Coalition for Content Provenance and Authenticity (C2PA). The coalition counts Adobe, Microsoft, BBC, Sony and Publicis Group among its steering committee. In short, it aims to introduce an industry standard for labeling content by the original creators. At present it’s focused on visual and audio, but it’s not inconceivable that text-based content could feature in future iterations.
What’s this got to do with generative AI? Everything. A creator uploads their original work with C2PA running and they can attach a name, a date and a location that is permanently locked onto that piece of content. If someone edits that work in the future their information will be added. It’s like an edit history of a shared document, one that allows a creator to track when and how their work is being used and when they have a payment to pursue. And that would include some of the myriad ways their work could be repurposed through generative AI. Shutterstock have recently committed to integrating C2PA into its AI capabilities, including its DALL-E powered AI image generator.
At first glance this is manna from heaven for creators. In a past life I wrote a piece for VICE on the difficulties online creators have with getting credit for their work and the frustrations that come with unsourced ‘who made this'?’ reshares. The C2PA labeling, sometimes likened to a ‘nutrition label’, would be easily found by viewers. At present C2PA suggests it could be a small i icon on the piece of media. Beyond crediting creators, C2PA is also getting plenty of good press for its potential to combat disinformation.
So that’s it? Game over, we’ve solved the problem? Not quite. Epic Games CEO Tim Sweeney recently posted, “Adobe’s C2PA proposals are abhorrent not only to artist rights but to a free society as well.” when sharing a clip from a July 12 Senate subcommittee on Artificial Intelligence and Intellectual Property. He wrote an interesting thread outlining his issues with the proposal, focusing on the potential for government surveillance and censorship, as well as barriers it could impose on creators, it’s worth a read.
But outside the C-suite, among creators, the debate about C2PA hasn’t really caught fire yet. The tweet below sums up the view of many Very Online creators. If existing art online is not just the inspiration for, but the actual ingredients of, your art, then this C2PA is a clear enemy. That includes a lot of creative people! We’re a couple of decades into meme culture where art can be made of ripped-up and repurposed existing art. That’s not going away with AI, it’s probably becoming more even pronounced.
But that view doesn’t represent all online creators. Many would like a way to pay their bills through their occasional viral moments and C2PA represents a possible solution. None of this is clear-cut, many of those same creators would also very much agree with the above tweet and feel naturally wary of a solution offered by major corporations. As creator Karpi told us in last Thursday’s edition, “the dangers of generative AI are the dangers of capitalism.” C2PA may offer as many dangers as opportunities.
A small style note: on a rewrite of today’s newsletter draft I replaced the word ‘content’ with ‘art’ in a few instances. It’s a dilemma, ‘art’ can sound pretentious, particularly when referencing new mediums or commercial creative work. But ‘content’ can read like this newsletter is written from a trading floor on Wall Street. The same goes for ‘artist’ versus ‘creator’. For the moment, we’re using all options.
When building a basic website for Explainable I wanted something AI-generated on the main page and on my initial test post. So that meant 30 minutes of throwing prompts into DALL-E and searching for names of different art styles. Everything looked a bit crap, like the image above, until I added “in the style of” and started thinking of artists I liked. Suddenly the results looked pretty decent. Or at least vaguely interesting. There was one problem though: the artist wasn’t getting a dime for the value that their work was adding.
Advice on fair use, depending on the lawyers you speak to, or whether you even have access to lawyers, can range from “don’t try anything, ever. In fact, never mention fair use again, it’s not a thing” all the way to “do whatever you want because; fair use”. And that level of vagueness is from a pre-generative AI age!
I toyed with adding a link to a page where the artist sells their work (in the example I briefly used, it was Stanley Donwood of Radiohead album cover fame) before deciding that was a little like the mega-successful Instagram scraper accounts who throw original creators a barely chewable bone with a credit in the comments. So down came the “in the style of” generated images.
This does not seem to be a problem that something like C2PA could fix, at least in its current proposed form. If a specific artwork was referenced, then C2PA could come into play, but the entire body of work of an artist? That could involve a continuation of the current vague whatever-you-think-best guidance. But maybe it will result in a mechanism for “in the style of” prompts to offer royalty payments. But that’s a big maybe.
A little look at copyright
With plenty of countries still scrambling to catch up with regulations around generative AI, it’s worth looking at policy published in March by Shira Perlmutter, the Director for the US Copyright Office.
The public guidance issued is pretty accessible, as documents on copyright law go, and based on case-by-case inquiries. A work described as “autonomously created by a computer algorithm running on a machine” was denied an application for copyright because it lacked “creative contribution by a human actor”. While a human-authored, Midjourney-illustrated, graphic novel was considered copyrightable as a whole, but not when images were considered individually.
In short, providing a single prompt is not enough to secure copyright of a work that is then generated by AI. A degree of human authorship needs to be shown within the process. To paraphrase an example from the document, if an artist uses generative AI like a visual artist uses Adobe Photoshop or a musician uses a guitar pedal, then copyright will likely be granted.
Deeper dives this week