The tech company OpenAI recently released the latest feature of its Generated Pre-trained Transformer 3 technology — the chat bot ChatGPT. The bot allows anyone online to have a dialogue with artificial intelligence, using human-like text.
Beyond the conversation, the bot can produce essays on command (I’ve read worse from graduate students). It can compose poetry and even create narrative plots at speeds rivaling the now memed pace of prolific author Brandon Sanderson.
The technology is, to put it lightly, phenomenal — but in its ability to produce text, can we really call it creative?
To test this question, I put the AI to work. I asked the bot (while thinking about my 4-year-old daughter) to write a prose paragraph, in the style of Herman Melville, about a girl eating an ice-cream cone. The results were quite remarkable:
The mimicry reads almost like Melville, but without “Moby-Dick,” the bot wouldn’t reference the “rail of the ship” nor the “rolling of the waves beneath her feet.”
The second prompt mirrored the first. I requested the same story, this time in the style of William Faulkner. Again, the results were impressive.
Like the Melville paragraph, this one borrowed words like “still” and “hot” from the work of Faulkner, adapting his cadence to describe the dense summer heat he experienced living in Oxford, Mississippi. But it was Faulkner who lived in that heat, not the bot.
Finally, I asked for the paragraph in the style of Cormac McCarthy. Once again, the bot produced a remarkable representation, but nothing generative or creative.
If I had asked for any literary style — or no style at all, or even “original” work from the bot — the responses would still be mimicry. As philosopher Sean Kelly argues beautifully: “(This) is what an artist does as an apprentice: copy and perfect the style of others instead of working in an authentic, original voice.”
As impressive as the computer prose imitations might be, they are no more than the outcome of algorithms processing large swaths of our language and data. The bot takes the sum total of Melville’s work (or Faulkner’s or McCarthy’s) and reproduces the average of that work (in connection with near infinite data via the algorithms) for its responses.
The nature of natural language processing is almost unfathomable, yes — and the technology will only increase in accuracy and efficacy. Eventually, the results could culminate into something appearing creative in expression or style. But the bot will never engage in genuine creation.
ChatGPT isn’t exactly writing, much less writing creatively. As Sean Kelly might say, it will never change our “way of thinking about what (literary writing) is that allows it to speak to what is needed now.” Likewise, none of the ChatGPT paragraphs would be possible if it weren’t for McCarthy, Faulkner or Melville. The paragraphs are simply computations. They are work without the art.
The philosopher Martin Heidegger wrote that any “work of art” consists of three distinct parts: the artist, the work and the art itself.
The first and the last are the elements generally familiar to all. We see a painting like the “Mona Lisa” (the art itself) and know that Leonardo da Vinci (the artist) painted it. What gets lost between the artist and the art is the work. The work element of a “work of art” is the sweat and living in the heat of Mississippi, the feeling and thriving Melville experienced at sea, or the suffering of the human condition worked into some artistic form in the plastic or literary arts.
The work is also the writing and composing, now automatable to some extent through AI. But that is not the totality of the work. There was some definite quantity of paint strokes (tens of thousands, if not hundreds of thousands) that ultimately worked into the strokes that would produce “Mona Lisa” — none visible but the last, yet the last depending on every stroke preceding. And as with da Vinci’s strokes, technologies like GPT-3 can, and likely will, do a lot of the heavy lifting going forward — but it will only ever be lifting.
The outsourcing of work in art is not new, of course. Renaissance and baroque artists like Rubens or Titian would build workshops and have their students do a lot of the work on a given painting. Once nearly complete, the master would come, finish off a few last things and stamp his name for completion. The work was not done by the artist, but rather finished at his hand, based on the work he previously established for himself. Similarly, skilled and experienced artists will likely follow suit into the future, outsourcing work to a computer while maintaining their own creative expression and dominion.
Yes, this technology is revolutionary, but it will not end artistic and intellectual life as we know it. Writing for The Atlantic, Stephen Marche is predicting the death of the college essay, similar to Francis Fukuyama and “The End of History.” Yet history continues its course. Many people pay upward of $50 a month to simulate hauling a bale of hay for daily exercise. I’m sure the universities will have no problem finding the humanist equivalent of the gym — we will need it.
The brilliant Kit Wilson recently wrote about ChatGPT, saying “(it’s) more or less inevitable, surely, that all of this will lead to the slow, self-perpetuating shrinking of the human imagination.” But most students today already have little to no imagination, let alone an ability to write. Perhaps, on the contrary, technologies like ChatGPT might spark their interest in developing both.
A hammer cannot create a man, but it can change him. I suppose there lies a future ahead where novels won’t be written without the aid of natural language processing, but I also imagine those novels could far exceed anything we can currently conceive of as great literature — because language processing will depend on the literary giants that came before and the work of those we will rightfully call artists.
David Perell, founder of Write of Passage recently wrote: “In a GPT world, where knowing a bunch of facts will become a commodity, developing an incredibly unique style is the best way to stand out as a writer.” The great artists and writers alike will do so. They will continue to work with all available tools and share their art for the benefit of humanity. Yes, mediocre television will likely multiply, along with the TikToks and low-level YouTube videos, but so too will run free the future Shakespeares and Cervantes: Prometheus unbound into the multiplying of our literary and technological flourishing.
Scott Raines is a writer and Ph.D. student at the University of Kansas.