Discover more from The Algorithmic Bridge
ChatGPT Is the Perfect Creative Machine—Except For One Big Little Thing
Have you heard of ShakespeareGPT?
I don’t know where William Shakespeare is now. Perhaps he’s in heaven. Perhaps his flesh and bones turned into dust and joined the stars on the other side of the universe. Perhaps his soul is still dancing around. Or perhaps he simply is no more.
I may not know where he is or where he was while alive. I don't know how he was as a person, as a son, or as a husband either, but I know him. You, too, know him. Well, everyone knows him because his words didn’t leave this world when he did; they stayed behind, with the rest of us, and traveled through time and space with the generations that followed until today.
He devoted his life to lending us a legacy that changed humanity forever. That legacy is as powerful today as it has ever been.
We’d know nothing about William Shakespeare, however, had he written Hamlet, Romeo and Juliet, or King Lear with ChatGPT.
Creativity as the mechanization of creation
“everything 'creative' is a remix of things that happened in the past, plus epsilon and times the quality of the feedback loop and the number of iterations.
people think they should maximize epsilon but the trick is to maximize the other two.”
Sam Altman wrote the above tweet on July 21st. That’s how he thinks creativity works.
Setting aside the pseudo-scientific formula, his first sentence is somewhat acceptable: Creativity combines “copying” others with an “epsilon”, which I’ll define as the small contribution that each of us puts into our creations (and I say small not to dismiss its importance but in sheer comparison to the giants’ shoulders on which we stand). Epsilon is that abstract, intangible part on a piece of art that would be non-existent had a different person made it. Feedback and iteration then allow this sum to evolve and develop and improve.
Altman’s second sentence, which he apparently considers the key to his insight, has sparked controversy. He says creatives should maximize copying and iterating, which is—I’ll assume, incidentally—akin to what ChatGPT does, delegating that mysterious “epsilon” to a secondary role; as I interpret his words not because it's a small part but because it borders on being irrelevant.
That’s not what creatives think, though, more like the exact opposite. We see why if we attend closely to Altman’s words; he never explicitly says what’s the goal of this formula is or what he’s maximizing for. If we applied his equation to the creative process and optimized as he suggests, what would we get?
Not creativity as I understand it.
If we dismiss that epsilon we all put into everything we create; those tiny idiosyncratic bits that critically separate inspiration from copying; those marks that imprint our presence in the world, what’s left is a cold, automatic, and unnaturally flawless algorithm to produce more.
We modernized the world by following that idea and paid the now evidently high cost of making it all look the exact same everywhere: we mechanized manufacturing, serialized every product, and traded off—to the absurd degree only possible under a socioeconomic system hungry of and obsessed with productivity—quality and meaning for quantity.
The key feature that ChatGPT lacks
Depending on who you ask, that’s not wrong. For people like Altman it’s actually the perfect process. For me, it’s a mechanism by which we’re slowly divesting the world from any leftovers of soul.
Whether you fall on one side or the other depends on what you understand by creativity and, most importantly, what you want out of it. We know the answer for Altman and OpenAI—because ChatGPT, if anything, lacks exactly that epsilon.
It has everything else: a bottomless memory filled with more text data than any human could consume in a lifetime, and the benefit of millions of users providing free feedback combined with how easily modern software, including neural networks, can be iteratively and cleverly improved on the deep pool of talent and money that OpenAI possesses.
But ChatGPT doesn’t have that epsilon; that spark that we can’t explain nor locate inside us. OpenAI researchers, of course, didn’t put it there because, as far as we know, it can't be encoded into a computer program. They expected it to emerge by itself—by feeding it with tons and tons of better or worse examples of human epsilons—but it didn’t.
No AI system can create like humans do.
Regardless of how important you think epsilon is—maybe everything has already been written, maybe we’re all just repeating, or maybe mixing things from the past is the only way to create—you can't deny there's something special there. Very special: It's something humans have that AIs don't.
A world without epsilons
Back somewhere in the late 16th century, our very human friend Shakespeare is writing some of the greatest pieces ever written. In between human experiences, feelings, and thoughts he finds inspiration.
Had he lived in today’s world, I think it’s safe to assume he wouldn’t have used ChatGPT. At least not as a substitute for his creative genius. If he did, if he had access to such a tool and unthinkably decided to replace his invaluable presence in his work, he would have erased his epsilon from the annals of history.
That epsilon might be tiny, even for Shakespeare, compared to the older influences that defined him as a person and as an author, but it was there, present in every word, sentence, and paragraph he ever wrote. What we imagine in our minds when we think of Shakespeare—not having known him personally but knowing his prose and poetry—is the compounded effect of his epsilon through the ages.
Had Shakespeare used an ancient ChatGPT, the ChatGPTs of today's world wouldn't be able to create a sonnet or play “in the style of Shakespeare” because there wouldn’t be such a thing as Shakespeare’s style. It would be ChatGPT’s style all the way down.
That's the paradox we face with ChatGPT; the paradox of a world without epsilons that Altman seems to prefer. ChatGPT, without an epsilon to itself, wouldn’t be a thing at all without the work of all the people that precede it. ChatGPT is a thing because of the Shakespeares of the past who collectively contributed to humanity’s archive.
Two important consequences stem from this observation: First, as we travel back to the beginning of human culture, it becomes rather clear that inspiration is nothing else but accumulated epsilons: the idiosyncratic perceptions, experiences, thoughts, and feelings that make up each and every person that has ever lived. The first cavern painters didn’t need inspiration from others, but a pure inclination to create from an inner world mixed with long-gone external beauty and alien-to-us lived experiences. The first hieroglyphics in the pyramids were faithful representations of the real things they represented. Whatever was written back then was in no way copying or mixed things from the past, or mere mindless iteration but the epsilon that we have by being human.
Second, what do we expect to find at the limit in the infinite if we keep creating through machines that increasingly erase our epsilons from a future yet to be written? If ChatGPT and all the other AI tools can't provide their epsilons and eventually start to feed on themselves, then the inevitable—and sad—outcome is that all epsilons will die out.
AI may not kill our bodies but it’s well on its way to kill our souls.