This is a thought experiment that shines light onto limitation on our current copyright system. It shows a paradox of what constitutes an "original works". As you will see, it also relates to current debacle over AI art and legality (or originality) of it.
You wake up in a small box room. You did not remember how you got in here, or anything really. You did not remember anyone or anything outside. You don't even know what year is it, or what the outside worlds looks like.
There seem to be no way to escape it. Let's assume all amenities were provided through cabinets on the walls. There is a door with no knob, a slit big enough for a canvas, and a display. A note attached, it reads:
TO OPEN THIS DOOR, YOU MUST SUBMIT A PAINTING WITH SCORE OF AT LEAST 1000000000
Fortunately, there seem to be unlimited source of canvas and paints. So you draw a quick sketch and submit it. Some clicking sound later and the display lights up:
SCORE: 0000000005
"Huh?" you thought. No waste of time, you draw another one, this time changing a little bit. After submitting the canvas, the display says:
SCORE: 0000000013
Fast forward a bit, you climbing up the score. You don't know (or care) what the painting actually is, but there seem to be things that the door likes. For example, the door like one person in the center, from waist up. You don't know who that is, so you just drew your face with a little change. The person wears black clothes, with hand-shaped blob. The background is pretty arbitrary, as you don't know anything. "So long as it works." you thought.
You now achieved the score of 999999965. "Almost there, just a few touch up and i'll be able to escape" you said. You start painting, making sure that to be as close as previous painting. You submitted it. A lot more clacking sound later, the door plays a jingle and shows:
SCORE: 1000000023
CONGRATS!
The door creaks, slowly opening. You smiled, after all the hard work you finally can escape. The door fully opens, in front of you is someone, a researcher, holding one of your painting.
CONGRATULATIONS, YOU JUST DRAW AN ORIGINAL MONA LISA!
Let's think about it. You, without any knowledge of Mona Lisa, ended up conjure Mona Lisa, with full originality and authorship onto you. This is a concept known as Independent Creation, where two person can create similiar work without infringing copyright of either if there is no way of one to copy another work.
(Please DO NOT use this as legal advice, i am DEFINITELY NOT a laywer and NOT RESPONSIBLE for any legal action based on it)
So here is the paradox: there is no copying happened. At least that's what the law says. The researcher did not copy anything, even if they had access to Mona Lisa. You definitely did not know Mona Lisa or similiar painting. The scoring system can be made to be oblivious of Mona Lisa. Even if some lawyer points out that the score for Mona Lisa is high, i can get similiarly high score on Starry Night, or a picture of capybara.
Well, you could say that surely some knowledge of Mona Lisa were transferred. But, you would miss that the room acts as isolation of knowledge, even with signalling allowed. In essence, this looks like zero-knowledge proof, where the only knowledge gained is a specific painting, without meaning attached.
The point is, there is no way to put the blame of copyright infringement to anyone. As if the copying were never happened in the first place.
Of course, it is just a thought experiment. I mean, it never is ethical to imprison someone for a painting, music, or any art. Hold your pants, because it is much more realistic than you thought. Or even, already been done.
There is no one stopping you from substituting the subject, researcher, or the door with a computer. In fact, this is exactly what AI is. A dumb computer, with no prior knowledge of the real world, dreaming of things that are real. The AI has no way of knowing what it just made, or capable of explaining how it's done (ignoring the "explainable AI" field).
I created this paradox long ago, ever since this appeared. A big inspiration comes from the Chinese Box Experiment. When Github Copilot announced, i thought, "Hey, isn't it just the experiment with computers?" And that's what compels me to refine the idea further. I gave it the name Mona Lisa Box because if you want to copy something, why not the most famous painting?
With DALL-E 2 and Stable Diffusion being very controversial among artists, maybe we should reconsider what being original means.
As someone who can't draw (that's why i program), i denounce any claim of originality with force. I has thunk really hard finding anything, anything remotely close to original. Something that i can say "Yes, i made this." But, years of searching found null result. Anytime i thought something new, turns out you can make it with really simple ingredients.
Well, this is nothing new. Programmers has been using others idea since beginning. We even make languages more modular, just to make it easier to integrate. Even "original" code very likely were made with estabilished/standard techniques, which anyone can retrace the path.
Since i'm not very literate with law, i can't give definitive solution to this paradox. I can, however, formulate a solution about the legality of AI generated work.
IMHO, AI work can be creative. AI cannot invent stuff, a human must put in the prompt. And very likely, the result is garbage. So the process of iteratively refining and tweaking is atleast, minimally creative. Similiar to how normal people draw, write, or code. Since AI destructively compiles all of it's training data, we can't say which part of what is copied, inspired, or simply randomly generated. I would err on the side of the ensemble (AI) becomes something new, beyond what the original input was. And AI engineers wants it, since such AI will generalize well.
I can't say where the law will go. But at least, this experiment will be an interest to law, philosophy of IP, and what is the limits to AI. Just like the Chinese Box experiment challenges validity of Turing Test.
© Dheatly23, 2022 (or not, i don't believe in it)
(Alternative heading title: WTF is US copyright office is doing‽)
Looking back at 1 year since, the US copyright office's stance on AI art is, simply nullify everything. Which i only expects from cowards who don't want to rule on potential infinite loophole.
Nullifying copyrightability on an entire field (which i shortened to field nullification) is something very rarely happens. It might have happened to song melody, since all the common combinations has already exhausted.
The problem of this stance is that the core problem is not in the use of AI. It doesn't matter if we used AI or Matrix-style machine enslaving humans for Mona Lisa Box. It's a fundamental misunderstanding of the problem. What the US copyright office is doing is equivalent to plastering holes in a wall made out of sponge.
So here's my prediction of what will happen do next. One, US congress passed law prohibiting using AI to smuggle copyrighted works, which is still not solve anything. Two, US congress annulls independent creation, grandfathered the rest, and make things much worse. Three, the problem is just kicked around further and further, maintaining the status quo. Four, some insane court decision allows Mona Lisa Box.
And yeah, that's my addition. See you next year, i guess?