“It’s AI Plagiarism!” …debunked

human art student makes a copy of a michaelangelo masterpiece

Understanding Human Reaction to Generative Creative Output

The initial astonishment of seeing AI art… so-called generative imagery. The fascination. The participation. (Lensa?) The virtue-signalling journalists. The complaining artists. The rallying cry: AI plagiarism!
…That’s the base accusation. I’ve heard it so many times now, from so many different fronts (artists, journalists, pundits, podcasters, “influencers”… even researchers!) that it’s beginning to hurt: on people’s first deepdive with AI creative outputs (today: pictures and essays), they proclaim:
“That’s not creativity.
. . . . . . .…that’s AI plagiarism!

Its not hurting me necessarily because it’s wrong, but rather because I can see it as a natural “knee-jerk” reaction to the awesome creative power of the new AIs, Class of 2022… a creative power that many of us find totally impossible to conceive of outside of a human spirit, much more so embodied in a… machine?

The thinking goes something like:

“Machines are mechanical, right? I mean, they may have a million miles of circuitry and a few billion transistors, but still… its logic! you give it A+B and it makes C… right? So how could that possibly be… creative?

(topic of a separate article: Neural Nets: connecting an unfathomably large number of logical components in parallel in order to transcend logic, embrace fractal chaos, and create genuine intuitive thought)

Most recently this sentiment — “Plagiarism! Copycat!” — appeared in a well-respected newsletter talking about how generative AI tech was “flooding the app store” and will soon be dominating the social media news feeds.

It concluded:

“ChatGPT would be docked for plagiarism
in any high school English class.”
I beg to differ. My fact check, from firsthand experience:
PATENTLY FALSE.
In fact, numerous professors from well-respected colleges and universities around the world have already declared that the rise of Text Generation (fka “ChatBot”) AIs spells, effectively, the death knell for the college essay challenge. And there are plenty of sophisticated web engines that test for plagiarism vs. previously published sources. And guess what? The AI-generated content passes those tests. With aplomb.
ChatGPT was trained on an unfathomably massive text dataset, in the same way that “AI.Artists” MidJourney, DALL-E, etc used a very large portion of the images (art, illustration, photographs) on the internet as a “training data set,” in order to understand how humans conceptualise visual art in terms of verbal descriptions (it takes the meta-info, captions, and context of nearby text as verbal descriptors of the image content, in addition to very large datasets of images with human-annotated & verified scene descriptions).
But let’s be clear: “training” is nothing like “plagiarism.”
I have had GPT write comedy skits, and some of the creativity of the lines it wrote had me falling out of my chair, laughing. Laughing so hard that I was *sure* that it must have stolen the key punchlines from something already published. So I did extensive Google searches for the key text phrases, even searching for a number of slight variations on the concept. Conclusion?
No Results Found… nothing online was even close.
Corollary: the AI output was both *original* (proven) and *creative* (opinion, value judgement) and therefore certainly not plagiarism, in fact, not even mildly *derivative* of existing tropes.
I’ve done similar experiments with the generative art engine outputs. Visual outputs that at first I deemed “too good / too creative to be done by a machine… it must have just found this image somewhere on Pinterest, Deviant Art, Etsy, Behance, etc.”
Put it up on Google Image search. Result: Again, nothing even close. The work was original. Every time. In other words, the AI’s creation had not existed moments before, in any form, until I put in the magic words and “summoned” it into existence. It was a direct product of what can only be properly called… the machine’s… imagination.

AI Plagiarism: True or False?

So why is this reaction — an immediate accusation of plagiarism — so common? Well, imho artists, writers and creators autopilot into the refrain of “plagiarism” for three main reasons:
1) they recognize that their original, copyrighted works are included in the core AI training data sets (true), and
2) the human artists, writers and creators have not been compensated in any meaningful or monetary way for their contribution to the AIs vast knowledge (true again), and
3) they cannot conceive of an AI that could ever possibly perform original creative thought (false).
Now as noted, artists are correct in their assertion that their original artworks, largely without their consent or permissionwere included in the training datasets.
So: let’s consider, for a moment, what “inclusion in a training data set” really means. It means that the artworks of any artist who has ever digitized their work, posted their work, shared their work, or had their work shared on the internet (read: 99.99% of all living artists) were included in the massive totality of what the AI analyzed during its intensive training. (which was intense, btw: it took all of about 8 human days, trained on more than 3 billion annotated digital images, analysing every pixel, and during that time consumed more total electricity than an average human city with a population of 100,000).

Plagiarism has Precedent

It’s important to note that human artists *also* train by studying the work of the masters, both historical and contemporary. Writers intensely study the works of both Shakespeare and Spielberg. First year art majors are required to painstakingly, by hand, copy the masterworks of Michaelangelo and Davinci, in order to learn composition and technique. So when they come into their own careers many years later, don’t you think that these human artists are influenced by the works of the masters that they studied?
Yet with human artists, we don’t call it “plagiarism”, we call it “inspiration” or “paying tribute to…” or “standing on the shoulders of…”
Picasso, when he was once criticized for being unoriginal and simply copying the style of other modern artists (the “cubist” school, specifically) went so far as to condone and applaud the Art of the Copy in his famous quip:
“Good artists copy.
. . Great artists steal.”
So what I’m getting at are three key points:
1) The AI trained (and trains) in a fashion very similar to classically-trained and -schooled human artists. It simply does it far faster, far more intensely, with a far larger data set (that being, the entire internet of words and pictures)
2) The AI’s outputs (both writing and artwork, and soon: music and film) are genuinely original creations, and even though at times, due to explicit human prompting, they may be constructed “in the style of… [human artist X]”, they are nonetheless truly original. In fact, many of the creations show a sophisticated sense of symbolism, metaphor, and even humor… previously thought of as exclusively human domains.
3) The AI’s outputs are therefor conclusively *not* plagiarism. Plagiarism is defined as “the practice of taking someone else’s work or ideas and passing them off as one’s own.” Synonyms include “copying, piracy, theft, stealing, poaching, appropriation… cribbing” This is not that. This is not copying. This is not piracy. This is creative inspiration.
In conclusion, accusations of plagiarism are a reactive and ill-conceived defense by humans who are feeling the threat of an AI that is better than most at something that, until this year, was thought of as the exclusive domain of humans: creativity. And while it is true that “the best” human artists are still “better” than the AI artists, the AI at this point (as with the domains of chess, poker, trivia, debate, computer programming, radiology, welding, and an ever growing number of fields) performs “better than the average human practitioner” along multiple axis: speed (faster), quality (better), and cost(cheaper).

“AI Plagiarism” Won’t Stop with “Art”

This should not concern only artists and creators. Art is only child’s play for the AI. Creative output is merely the first “shot across the bow” in AI’s relentless march to analyze, digest, and replicate every field of human endeavor. (Actually, AlphaZero was the clarion call way back in 2015, but since it was only Chess, not many people took notice. Chess is, theoretically (but not practically) a strictly “logical” game of 64 squares and 32 pieces that should be “solvable”… “creativity” is a far more abstract realm).
We all need to see the writing on the wall at this point. AI Plagiarism is not the real issue at all. The hard truth is: The AI is learning, it is growing, and it is both unstoppable and inevitable. It is time for we humans to reassess our roles, our values, what we think of as “work”, and most importantly, what we see as our individual & collective “purposes” in our short, shining lives.
I’m with you on this journey, fellow humans.
Wakey Wakey!
, , ,

Exit mobile version