Thread by @memotv

0. I dont know the nearterm solution to online content being scraped & used in training AI, to the horror of many artists who fear for their livelihood, and to the excitement of those who find joy & liberation in this new medium.

But a future solution cld be:

#AiArt #Artstation

1. ALL content uploaded online to be shared *publicly* is automatically recorded on a blockchain (& IPFS or similar). Yes like an N.F.T. But not for the purpose of selling or shilling. Instead, it's simply an immutable public proof-of-authorship.

2. This public proof-of-authorship is on an internet wide, distributed, public database (i.e. like a blockchain + IPFS - henceforth called "WWDB"), as opposed to multiple privately owned databases like ArtStation, DeviantArt, Instagram, Twitter, Facebook, Artbreeder, Tumblr etc.

3. All of those websites (ArtStation, DeviantArt, Instagram etc) can still have their own DBs, but they all either directly use, or cache, the content from the public WorldWideDB (similar to how current NFT websites work with various blockchains).

4. The creator/uploader can also set usage conditions, i.e. a license. Including allowing or not-allowing for training AI. I imagine @creativecommons licenses will be updated to provide more clarity on usage in training AI.

5. Any web scrapers, or any websites for that matter, can always trace back (through hash and/or visual search) the original image on the WWDB, and check the credits and license conditions.

6. The license could also dictate prices for being included in an AI dataset. Such that creators of AI models automatically pay the artists to use their works in TRAINING their models. Similar to stock image websites work today, but on a broader scale across entire internet/WWDB.

7. Alternatively (or as well as), royalties can be automatically paid during INFERENCE - by the service provider (e.g. Dalle, Midjourney), or the AI Image Creator, or shared between both. E.g. prompting "in the style of ABC", would send a tiny fee to the orig artist ABC.

8. Or, The AI Image Creator doesn't pay this usage fee to the original artist upon creation of a new AI image (bc they might want to try creating loads), but rather, on the monetization of the AI image - since that metadata will also exist in the WWDB regarding the new AI image.

9. Likewise prompting with an existing img as a style reference could send a royalty to the creator of the original image. Since all images are registered on this WWDB, the style image can be traced, the original creator of that image found, and credited accordingly.

10. This WWDB also solves the long standing 'inspiration / credit' issue, where there are millions of images floating on the web, on pinterest, on @archillect, on countless 'inspiration' platforms, of "found" images, where the source has been lost.

11. The key is that there is this internet wide, distributed, public database that records ALL content (provided the creator chooses this of course. If they choose not to participate, they won't get any of these benefits, but then they can't complain if their work is 'stolen' :).

12. Of course many blockchains already are "internet wide, distributed, public databases". But for the above to work, there has to be a "standard default" DB. Like TCP/IP, DNS, there has to be one standard WorldWideDB. (Or at least a standard protocol that bridges multiple DBs).

13. I'm sure there are many holes here that will be exploited by bad actors. This is just a start, and needs fleshing out. But if there is a public distributed proof-of-authorship in a WWDB, then at least we have the nessecary information, and then we can try to set policies.

14. I wish we wouldnt need something like this. I wish everyone could share everything freely, and we cld all draw inspiration from everywhere, and train AI on everything too. But we all know the profit-maximizing machine is a bulldozer that will plough thu everything it can.